Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Accuracy Evaluation and Parameter Analysis of Land Surface Temperature Inversion Algorithm for Landsat-8 Data

Accuracy Evaluation and Parameter Analysis of Land Surface Temperature Inversion Algorithm for... Hindawi Advances in Meteorology Volume 2021, Article ID 9917145, 16 pages https://doi.org/10.1155/2021/9917145 Research Article Accuracy Evaluation and Parameter Analysis of Land Surface Temperature Inversion Algorithm for Landsat-8 Data 1 1,2 3 Jikang Wan , Min Zhu , and Wei Ding Hohai University, College of Computer and Information, Nanjing 211100, China Yellow River Institute of Hydraulic Research, Zhengzhou 450003, China University of Calgary, Department of Geomatics Engineering, Calgary AB T2N 1N4, Canada Correspondence should be addressed to Min Zhu; 190207050001@hhu.edu.cn Received 3 August 2021; Revised 29 August 2021; Accepted 8 September 2021; Published 24 September 2021 Academic Editor: Stefania Bonafoni Copyright © 2021 Jikang Wan et al. -is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Many researchers have developed a variety of land surface temperature (LST) inversion algorithms based on satellite data. -e main LST inversion algorithms include Radiative Transfer Equation (RTE), Single Channel (SC) algorithm, Mono Window (MW) algorithm, and Split Window (SW) algorithm. In this study, nine LST inversion algorithms were designed using Landsat-8 data and meteorological station data to test the inversion efficiency of different algorithms in different seasons and different locations. -e results show that the error of various LST inversion algorithms will increase with the rise of LST. R of the inversion results of ° ° each LST algorithm and the measured data are all greater than 0.73 C in winter and about 0.5 C in the other seasons. By analyzing the stability of various algorithms inside and outside the city, it is found that the stability of each LST inversion algorithm inside the city is better than that outside the city. For the same surface features, the inversion temperature inside the city is 3–5 C higher than that outside the city. In addition, the sensitivity of various inversion algorithms to parameters was also analyzed. -e influence of atmospheric transmittance on RTE, SC, and MW inversion algorithms is in logarithmic form. -e effect of emissivity on each algorithm is linear. -e influence of NDVI on the algorithms is mainly through the estimation of surface emissivity parameters to affect the inversion results. -e effect of ascending radiation on SC (LST4 and LST5) is linear and on RTE (LST1 and LST2) is logarithmic. -e effect of downslope radiation on SC and RTE is linear. -e influence of atmospheric water vapor content on SW (LST7) is nonlinear. method cannot represent the regional temperature. -e 1.Introduction timeliness of the method is insufficient. With the develop- Land surface temperature (LST) is an essential parameter in ment and research of space-based infrared remote sensing, it many research fields, such as ecology, climatology, urban is more and more popular to use satellite remote sensing to thermal environment, urban heat island, and hydrology retrieve regional and global surface temperatures. [1–3]. Especially in recent years, more attention has been In many satellite observation systems equipped with paid to carbon dioxide emissions [4, 5], polar climate thermal infrared sensors, Landsat satellites have become one of the most widely used satellites due to their observation warming and glacier melting [6, 7], urban heat island (UHI) [8–10], urban heatwave and human living health [11], forest history of more than 40 years. Since its launch on February fires [12], drought monitoring [13], soil evapotranspiration 11, 2013, the newest generation of Landsat-8 has been in [14], and precipitation and surface runoff [15, 16]. -e orbit for more than eight years. It has sent back a large traditional LST measurement method is based on the amount of Earth observation data at a rate of 550 images per temperature measurement data of meteorological stations day. Together with Landsat-7 ETM+, it forms a repeated for further spatial interpolation, which is a classical point- observation period with an interval of 8 days, which provides based measurement method [17]. Although this method has precious data for monitoring ecological environment high accuracy, the point-based temperature measurement changes on the Earth’s surface. Because of its advantages of 2 Advances in Meteorology high spatial resolution, large scanning field of view, and which seven multispectral bands have a spatial resolution of accessible data, researchers have applied it to further sci- 30 m. -rough resampling, the spatial resolution of the two entific research and ecological construction. thermal infrared bands is also 30 m. -e revisit period is 16 Since the launch of Landsat-8, various processing doc- days for global coverage. Since the surface reflectance is uments and parameters have been posted on the Landsat. affected by the atmospheric conditions at the time of image Many researchers have developed a variety of LST inversion acquisition and it is also a potential factor for studying the algorithms based on Landsat-8. -e main LST inversion surface thermal model, when screening the data, we select algorithms include the Radiative Transfer Equation (RTE) the imaging data under good weather condition, which was a [18, 19], Single Channel (SC) algorithm [20, 21], Mono clear day with less than 2% cloud cover, and use the thermal Window (MW) algorithm [22], and Split Window (SW) band data to retrieve the LST conduct research. In order to algorithm [21, 23–25]. In these surface temperature inver- investigate the accuracy of the Landsat-8 land surface sion algorithms, multiple basic parameters are needed for temperature inversion algorithm in season, the imaging input [26]. -e multiple parameters are estimated variables dates of January 31, 2017 (winter), May 7, 2017 (spring), July rather than standard ones. Accurate and reliable inversion 10, 2017 (summer), and October 30, 2017 (autumn), were algorithms should calculate the accuracy of the LST data selected. All images were named by the 8-digit number of used in the study to make the use of the product possible. If years, months, and day of the collection date for the con- the inversion algorithm has a significant systematic error or venience of recording. For example, the image on May 7, a significant instability error, the algorithm or product will 2017, as shown in Figure 1, was named 2017-05-07. not be popular among users [27]. -is study also collected and sorted the surface tem- In this study, LST inversion, spatial analysis, and mul- perature data of 20 meteorological stations in the study area. tiscale analysis methods were considered for Landsat data -e basic information of the meteorological stations is representing four seasons. -e commonly used LST inver- shown in Table 1. -e positions of these meteorological sion algorithms are evaluated. First of all, we designed nine stations in the study area are shown in the red dots in remote sensing inversion schemes for LST. We fitted the LST Figure 1. -e temperature data were collected from the China Meteorological Administration and measured si- calculated by the satellite with the measured temperature values at the ground stations at the same time. Based on the multaneously as remote sensing images. To study the ac- same scene remote sensing data, the accuracies of the nine curacy of the inversion algorithm of the LST inside and inversion algorithms were compared and analyzed. -e outside the city, points of interest (POIs) were established for accuracy of each LST inversion algorithm on the time scale four kinds of ground objects (water, vegetation, soil, and was analyzed separately on the images of different seasons. building), and 20 POIs were established for each ground -en, the inversion accuracies of all the LST inversion object. Pure pixels were selected for each point of interest methods on the regional scale (inside city and outside city) of from Landsat-8 multispectral data. High-resolution Google the same surface features are compared and analyzed. Fi- Earth images also examined the selection of all POIs. As nally, we test the dependence of each inversion algorithm on shown in Figure 1, those outside the city are represented by parameters by equal step size. -e aims of this study are as green dots, while yellow dots represent those inside the city. follows: (1) evaluate the errors of multiple inversion algo- rithms on the same scene data, (2) evaluate the error of each 3.Methods inversion algorithm in seasons (spring, summer, autumn, and winter), (3) evaluate the regional error of each inversion 3.1. Multispectral Image Processing. Using the Landsat-8 algorithm, and (4) perform dependence intensity and sen- website, the user can convert the Digital Numbers (DNs) of sitivity analysis of the evaluation algorithm on parameters. the image to the spectral radiance of the top of the atmo- sphere (TOA). By applying the equation 2.Study Areas and Data Compilation L � M Q + A , (1) λ L cal L 2.1. Study Areas. -e study area was selected as an area −2 −1 −1 where L (Watts·m ·srad µm ) refers to the TOA covered by Landsat-8 data (orbit number 123, 032). -e spectral radiance, M is the multiplicative rescaling factor of longitude and latitude of the data center are 40.3 N and the corresponding band, it can be obtained from the header 116.7 E. It mainly covers Beijing, the capital of China. -e file of the data with the field “Radiance_Mult_Band_x � ,” x climate type in the study area is semihumid and semiarid stands for band number. Q is the pixel values, and A is the cal L monsoon climate in the warm temperate zone, with four offset of the data. -e multispectral bands of Landsat-8 data distinct seasons, hot and rainy in summer and cold and dry can also be converted to the reflectivity of the top of the in winter. It provides a natural, convenient condition for atmosphere by the following equation: studying the difference of inversion algorithms on a seasonal ρ � M Q + A , (2) scale. λ ρ cal ρ where ρ is the reflectance of the top of the atmosphere at band λ without correction of the solar angle, M is the 2.2. Data Compilation. -is study contained four pieces of reflectivity adjustment factor of band λ, and it can be ob- Landsat-8 data based on USGS (United States Geological tained from the header file of the data with the field Survey, https://www.usgs.gov/). Landsat-8 has 11 bands, of Advances in Meteorology 3 115°30’0”E 116°0’0”E 116°30’0”E 117°0’0”E 117°30’0”E 118°0’0”E W E 010 20 40 60 80 Km 115°30’0”E 116°0’0”E 116°30’0”E 117°0’0”E 117°30’0”E 118°0’0”E Station Outer-ROI Inner-ROI Figure 1: Study areas. -e remote sensing data of “LC08_L1TP_123032_20170507_20170515_01_T1,” displayed in true color, is named 2017-05-07, with red dots representing the location of the meteorological station, green dots representing the points of interest outside the city, and yellow dots representing the points of interest inside the city. Table 1: Basic information of meteorological stations in the study area. Name Latitude Longitude Elevation (m) Land cover type ° ° Shunyi 40.13 N 116.61 E 28.6 Grassland ° ° Haidian 39.98 N 116.28 E 45.8 Grassland ° ° Yanqing 40.45 N 115.96 E 487.9 Natural vegetation mosaic ° ° Foyeding 40.60 N 116.13 E 1224.7 Natural vegetation mosaic ° ° Tanghekou 40.73 N 116.63 E 331.6 Natural vegetation mosaic ° ° Miyun 40.38 N 116.86 E 71.8 Grassland ° ° Huairou 40.37 N 116.63 E 75.7 Grassland ° ° Shangdianzi 40.65 N 117.11 E 293.3 Natural vegetation mosaic ° ° Pinggu 40.17 N 117.11 E 32.1 Grassland ° ° Tongzhou 39.85 N 116.75 E 19.8 Grassland ° ° Chaoyang 39.95 N 116.50 E 35.3 Grassland ° ° Changping 40.21 N 116.21 E 76.2 Grassland ° ° Zhaitang 39.96 N 115.68 E 440.3 Natural vegetation mosaic ° ° Mentougou 39.88 N 116.15 E 85.5 Grassland ° ° Beijing 39.80 N 116.46 E 31.3 Grassland ° ° Shijingshan 39.95 N 116.20 E 63 Grassland ° ° Fengtai 39.86 N 116.25 E 55.2 Grassland ° ° Daxing 39.71 N 116.35 E 37.5 Grassland ° ° Fangshan 39.76 N 116.20 E 48.9 Grassland ° ° Xiayunling 39.73 N 115.73 E 407.7 Grassland 39°30’0”N 40°0’0”N 40°30’0”N 41°0’0”N 39°30’0”N 40°0’0”N 40°30’0”N 41°0’0”N 4 Advances in Meteorology “Reflectance_Mult_Band_x � ,” where x stands for band where NDVI is the NDVI value of bare soil or areas soil number. A is the reflectivity adjustment parameter of band without vegetation coverage and NDVI is the NDVI value ρ veg λ, the header file of the data with the field of pixels completely covered by vegetation, namely, the “Reflectance_Add_Band_x � .” ρ can be further corrected NDVI value of pure vegetation pixels. Empirical values by equation (3) into atmospheric top reflectivity ρ : NDVI � 0.6 and NDVI � 0.05 were taken; that is, when λ veg soil the NDVI of a pixel is greater than 0.60, P value is 1.0, and ′ 􏼐M Q + A 􏼑 ρ cal ρ λ when NDVI is less than 0.05, P value is 0. (3) ρ � � , cos θ cos θ -e other method LSE2 is to firstly divide the surface z z into water, natural surface, and urban area according to the where θ is the solar zenith angle of the image center, and method proposed by Qin et al. [31] and then calculate the solar altitude angle θ can also be used, but sine function LSE for the three types of surface separately. -e calculation should be used. We used the solar altitude angle provided in is by the following equation: the image for our calculations. ⎧ ⎪ 0.995 water ⎫ ⎪ In this study, the COST model atmospheric correction ⎪ ⎪ ⎪ ⎪ ⎨ ⎬ algorithm of Pat and Chavez [28] is adopted to conduct 0.9625 + 0.0614P − 0.0461P natural surface ε � . v v ⎪ ⎪ ⎪ ⎪ Landsat-8 atmospheric correction. COST requires few pa- ⎪ ⎪ ⎩ 2 ⎭ 0.9589 + 0.086P − 0.0671P urban area rameters, and the correction is mainly carried out with the v v data of the image itself [29] by the following equation: (9) 􏽨M Q − Q 􏼁 + A 􏽩 ρ cal h ρ However, the concept of natural surfaces and urban (4) ρ � , λ cost areas is vague and cannot be quantitatively calculated. Based cos θ τ on remote sensing images of Beijing and its surrounding where Q is the modification value of atmospheric influence, h areas, we conducted local experiments and found that when which can be obtained by the darkest pixel method, NDVI is less than 0.05, it corresponds to water; when NDVI τ � cos[(90 − θ )π/180], and τ is the atmospheric trans- is greater than or equal to 0.05 or less than or equal to 0.6, it mittance estimated based on θ . After COST atmospheric corresponds to urban areas; and when NDVI is greater than correction, NDVI (normalized difference vegetation index) 0.6, it corresponds to the natural surface. -e calculation is is calculated by the following equation: in the following equation: ρ − ρ 􏼁 NIR COST red COST ⎧ ⎪ 0.995 NDVI < 0.05 ⎫ ⎪ NDVI � , (5) ⎪ ⎪ ⎪ ⎪ ⎨ ⎬ ρ + ρ 􏼁 NIR COST red COST 0.9625 + 0.0614P − 0.0461P 0.05 ≤ NDVI < 0.6 ε � . ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 2 ⎭ where ρ is the reflectance image of the near infrared nir COST 0.9589 + 0.086P − 0.0671P 0.6 ≤ NDVI v v band and ρ is the reflectance image of the red band. red COST (10) 3.2. 7ermal Infrared Image Processing. -e brightness temperature (BT) can be calculated using the following 3.4. Land Surface Temperature Inversion Algorithm. Based equation: on four commonly used LST inversion methods (RTE, SC, 2 MW, and SW), a total of nine schemes were designed based T � , (6) ln K /L 􏼁 + 1􏼁 on different parameters such as LSE and water vapor content 1 λ (w). -e input atmospheric parameters such as downwelling where T is the BT; K and K for the Landsat-8 band 10 are 1 2 radiance (L↓, W/m /sr/μm), upwelling radiance −2 −1 −1 774.89 Watts·m ·srad µm and 1321.08 K, respectively. 2 − 2 (L↑, W/m /sr/μm), water vapor content (g · cm ), and L is the thermal infrared band after radiation calibration atmospheric transmittance (τ) were calculated using the and COST atmospheric correction. Atmospheric Correction Parameter Calculator (ACPC, https://atmcorr.gsfc.nasa.gov) developed by National Aeronautics and Space Administration (NASA) of the US 3.3. Estimation of Land Surface Emissivity. In this study, we [32, 33]. applied two LSE (LSE1 and LSE2) estimation methods; both methods are based on NDVI. -e first method LSE1 refers to the NDVI threshold method proposed by Sobrino [30] 3.4.1. RTE Inversion Methods. -e RTE is a direct method calculated by the following equation: for LST retrieval using a single TIR band. -is can be given by the following equation: ε � 0.004P + 0.986, (7) L � εB T + (1 − ε)L↓ τ + L↑, (11) 􏼂 􏼁 􏼃 λ s where ε is the LSE and P is calculated according to NDVI using the following equation: − 2 − 1 − 1 where L (w · m · sr · μm ) is the brightness value of TIRS received by satellite sensor and λ is the tenth band of NDVI − NDVI 􏼁 soil P � , TIRS. At present, the accuracy of the eleventh band cali- (8) NDVI − NDVI 􏼐 􏼑 veg soil bration of TIRS is still not quantified, and USGS still does Advances in Meteorology 5 not encourage the use of eleventh band for relevant surface Landsat ETM + data and Landsat-8 data, which can be used temperature inversion, so we only consider the tenth band in in the following equation: the RTE and SC. ε refers to the LSE, B(T ) is the blackbody a · (1 − C − D) + [b · (1 − C − D) + C + D] · T − D · T 􏼈 􏼉 radiance energy, and B(T ) at a temperature of T is cal- T � , s s s culated by the inversion of the following equation: (19) B T 􏼁 � 􏼂L − L↑ − τ(1 − ε)L↓􏼃τε, (12) s λ where T is the effective mean atmospheric temperature and a and b values for the Landsat-8 band 10 are −67.355 and and, eventually, T can be obtained from the inversion of 0.458606, respectively. C � ε · τ and D � (1 − τ)[1 + Planck’s law as in the following equation: (1 − ε) · τ]. Table 2 provides empirical equations to estimate T T � . through air temperature (T ), since it is an essential parameter (13) s o ln K /B T􏼁 + 1􏼁 1 s of MW (Zhang et al. 2016). In this study, T was computed for the mid-latitude summer and winter region and T was ob- tained from the mean of all meteorological stations. 3.4.2. SC Inversion Methods. Jimenez-Muñoz ´ et al. [20] proposed a revised version of SC for LST retrieval using 3.4.4. SW Inversion Methods. Although the USGS does not Landsat TIR data. Concerning the SC, T is obtained from encourage users to use band 11 calculations, some re- the following equations: searchers have developed a Split Window algorithm for − 1 � c􏽨ε ψ L + ψ 􏼁 + ψ 􏽩 + δ, (14) Landsat-8 that references other satellite two-channel algo- s 1 2 3 rithms. Jimenez-Muñoz ´ et al. [21] proposed a revised version 2 2 of SW for LST retrieval using Landsat TIRS data. -e al- T T c � , δ ≈ T − , (15) gorithm is simple to calculate. After the brightness tem- 􏼐b L􏼑 perature of band 10 and band 11 of TIR is obtained, it can be calculated by the following equation: where b is equal to 1320 K for Landsat-8 band 10 and L and T are obtained by equation (1) and equation (6), respectively. T � T + c T − T 􏼁 + c T − T 􏼁 + c s 10 1 10 11 2 10 11 0 (20) ψ , ψ , ψ is a function of water vapor content (w) expressed 1 2 3 + c + c w􏼁 1 − ε 􏼁 + c + c w􏼁 Δε, 3 4 m 5 6 in the following equation: where T and T are the brightness temperatures of band ψ � p w + p w + p , 10 11 1 11 12 13 10 and band 11 and ε is the average LSE of band 10 and (16) ψ � p w + p w + p , 2 21 22 23 band 11. Δε is the LSE interpolation of band 10 and band 11. In this study, the LSE of main ground objects was obtained ψ � p w + p w + p , 3 31 32 33 through ASTER spectral library, and the estimated value of three NDVI segments in equation (10) was matched through where p (i � 1, 2, 3, j � 1, 2, 3) is the parameter related to w. ij the calculation of the average value of LSE of relevant ground Perhaps because the calibration of TIRS band 11 is still objects. Table 3 shows the estimated LSE. w is the water unstable, Jimenez-Muñoz ´ et al. only gave specific parameters vapor content (w), and the values of parameters c ∼ c are for TIRS band 10: 0 6 −0.268, 1.378, 0.183, 54.3, −2.238, −129.2, and 16.4, 0.04019 0.02916 1.01523 ⎪ ⎪ ⎧ ⎫ ⎪ ⎪ respectively. ⎨ ⎬ p � −0.38333 −1.50294 0.20324 . (17) Rozenstein et al. [25] proposed an SW algorithm for ⎪ ⎪ ⎪ ⎪ ⎩ ⎭ inversion of land surface temperature using Landsat-8 TIR 0.00918 1.36072 −0.27514 data based on the MW algorithm of QIN [23]. -e algorithm − 2 is expressed by the following equations: When w ≥ 3g · cm , the accuracy of LST will be sig- nificantly affected, so Jimenez-Muñoz ´ suggests using T � A + A T − A T , (21) s 0 1 10 2 11 equation (18) to calculate ψ , ψ , ψ : 1 2 3 1 10 A � , (22) ψ � , E � D C − D C , (23) −L↓ − L↑ 0 11 10 10 11 (18) ψ � , D 1 − C − D 11 10 10 E � , (24) ψ � L↓. D 1 − C − D 10 11 11 E � , (25) 3.4.3. MW Inversion Methods. Mono Window (MW) al- C � ε τ (θ), (26) i i i gorithm was originally developed for Landsat TM data. D � 􏼂1 − τ (θ)􏼃􏼂1 + 1 − ε􏼁 τ (θ)􏼃, (27) Later, some scholars evaluated the ability of MW to process i i i i 6 Advances in Meteorology Table 2: -e linear equations for the calculation of the effective Table 4: a and b over different temperature ranges. i i mean atmospheric temperature (T ) from the near-surface air Temperature ranges a b a b 10 10 11 11 temperature (T ). 0∼30 C −59.1391 0.4213 −63.3921 0.4565 Region Linear equations 0∼40 C −60.9196 0.4276 −65.2240 0.4629 USA 1976 region T � 25.940 + 0.8805 × T ° 10∼40 C −62.8065 0.4338 −67.1728 0.4694 a o Tropical region T � 17.977 + 0.9172 × T ° a o 10∼50 C −64.6081 0.4399 −69.0251 0.4756 Mid-latitude summer region T � 16.011 + 0.9262 × T a o Mid-latitude winter region T � 19.270 + 0.9112 × T a o Table 5: Relationship between atmospheric transmittance and atmospheric water vapor content. Table 3: Estimation of the associated LSE. Profile Estimation equation R SEE Band NDVI < 0.05 0.05 ≤ NDVI < 0.6 0.6 ≤ NDVI τ � −0.1146w + 1.0286 0.9882 0.0094 1976 U.S. Standard TIRS 10 0.9908 0.9451 0.9824 τ � −0.1568w + 1.0083 0.9947 0.0086 TIRS 11 0.9901 0.9503 0.9819 τ � −0.1134w + 1.0355 0.986 0.1010 Mid-latitude summer τ � −0.1546w + 1.0078 0.996 0.0073 where A � E a − E a , A � 1 + A − E b , A � 0 1 10 2 11 1 1 10 2 A + E b , A , A , A are the parameters determined by 2 11 0 1 2 Table 6: Different LST inversion methods. atmospheric transmittance and LSE, and a , b are, respec- i i Model Model + parameter Model ID tively, the regression coefficients of band 10 and band 11 RTE (LSE1, τ, L↑, L↓ ) LST1 determined according to different temperature ranges; this is RTE RTE (LSE2, τ, L↑, L↓) LST2 shown in Table 4. In this study, parameters with temperature ° ° SC (w) LST3 ranging from 0 C to 40 C were selected. ε is the LSE of band SC SC (LSE1, τ, L↑, L↓) LST4 10 and band 11. τ (θ) is the atmospheric transmittance SC (LSE2, τ, L↑, L↓) LST5 corresponding to the solar zenith angle. Rozenstein et al. SW (by Jimenez-Muñoz ´ et al.) (LSE2, w) LST6 provided the relationship between atmospheric transmit- SW SW (by Rozenstein et al.) (LSE2, w) LST7 tance and atmospheric water vapor content, as shown in MW (LSE1, τ, L↑, L↓) LST8 Table 5; namely, atmospheric transmittance was estimated MW MW (LSE2, τ, L↑, L↓) LST9 by atmospheric water vapor content, which was also the calculation method in this paper. In this study, the pa- rameters determined by mid-latitude summer are selected. Inputs of parameters such as LSE, up radiance, down radiance, atmospheric transmittance, atmospheric water and gas content, and atmospheric average temperature are 3.5. Validation Method and Performance Metrics. In this needed in the inversion algorithm. Each of the above pa- study, based on four commonly used LST inversion methods rameters is not directly measured by the instrument but has and different parameters, a total of nine schemes are been estimated and processed. -e errors of these inputs of designed, which are shown in Table 6. parameters will affect the accuracy of LST inversion. Sen- -e T-based technique, examined in this research, is a sitivity analysis is an application of how the errors (nu- direct way of comparing the satellite-derived LST with merical, statistical, or otherwise) of the model output are measured temperatures at meteorological stations [34–38]. divided and allocated to different sources of uncertainty in -e capability of the T-based technique depends mostly on the model input [41]. -e following equation is utilized: the accuracy of the measured temperatures at meteorological stations and pixel scale effect. -e major benefit of the S � T (x) − T (x + Δx), (30) e s s T-based method is that the accuracy of LST algorithm can be where x represents one of the input parameters and Δx is the accurately evaluated [39]. In this study, satellite-derived LST step size of the parameter setting. In the study species, at- and station-based LST were analyzed considering the per- formance metrics such as Root Mean Square Error (RMSE) mospheric transmittance (τ), emissivity, and NDVI were set as 0.01, upwelling radiance and downwelling radiance were and average Bias [40]. -e formulas of these metrics are as set as 0.1 unit, water vapor content (w) was set as 0.1 unit, follows: mean surface temperature was set as 1 unit, and S was the 􏽳�������������� � LST difference calculated for each increase in step size; T − T 􏽐􏼂 􏼃 L8 station (28) RMSE � , T (x + Δx) and T (x) refer to the LST calculated for s s “x + Δx” and “x,” respectively. 􏽐 􏼂T − T 􏼃 L8 Station (29) BIAS � , 4.Results and Analysis where T and T are the Landsat-8-derived LST and In order to accurately evaluate the accuracy of the LST L8 Station station-based LST, respectively, and n refers to the number inversion algorithm, measured data from 20 meteorological of pieces of data; in this study, n is equal to 20. stations in the study area were used to evaluate the stability Advances in Meteorology 7 of the inversion results of different LST inversion algorithms temperature, the error of the inversion results of the algo- in different seasons. In addition, considering the difference rithm was smaller than that of the measured data. With the of inversion results of the same reflectivity ground objects by increase of temperature, the inversion results of all the in- different LST inversion algorithms, we designed two groups version algorithms were higher than those of the measured of experimental areas inside the city and outside the city. data. LST1, LST2, LST4, and LST5 have a relatively small Each experimental area has four types of ground objects, error, followed by LST3 LST6, LST7; the inversion errors of respectively, and each type of ground object has 20 POIs. LST8 and LST9 were relatively large. Finally, we analyzed the dependence of different LST in- We performed linear fitting between nine LST inversion version algorithms on parameters by equal step size. results and measured temperatures in four seasons. -e fitting results are shown in Figure 3. According to the statistical fitting results, it can also be seen that, in winter, all 4.1. Comparison of the Results of Different LST Inversion the relevant results were relatively ideal, with R greater than Algorithms. Nine LST inversion algorithms were used to 0.73. Except for the two SW algorithms LST6 and LST7, the calculate Landsat-8 data in four seasons in the research area. fitting slopes of the other algorithms were all less than 1. -e -e results were compared with the measured temperatures fitting results in autumn show that R was around 0.5, and all of 20 meteorological stations. -e results are shown in the fitting slopes were less than 1. However, the relevant Figure 2. By comparing the inversion results of each algo- intercept data are relatively large. In spring and summer, the rithm with the measured data, the statistical Root Mean proper slopes were more significant than 1. -e absolute Square Error (RMSE) and average Bias are shown in Table 7 value of the fitting intercept was also relatively large. It is of Appendix. It can be obtained from Figure 2 and important to note that the fitting results of the LST7 al- Appendix’s Table 7 that the inversion results of the LST gorithm in spring, summer, and autumn show no significant algorithm are entirely consistent with the graph trend of the phenomenon; that is to say, the error between the inversion measured data of meteorological stations. -e high elevation results of the LST7 algorithm and the measured results is meteorological sites (e.g., Foyeding, Yanqing, Tanghekou, unacceptable. and Zhaitang) were low. In the graphic display, the line with Overall, the inversion results of the RTE and SC algo- the measured data of adjacent stations shows a concave rithm calculated by LSE1 and LSE2 have an excellent fitting shape. All LST inversion algorithms also have the same effect with the measured data; the inversion results of the SC characteristics. In terms of seasonal scale, the inversion algorithm are calculated by w. -e inversion result of MW results of the nine algorithms in winter have the highest was higher compared to the measured temperature. -e matching degree with the measured data of meteorological inversion effect of SW was not particularly ideal, which may stations. -e values of RMSE and Bias were relatively small. be caused by the instability of the radiation calibration of the Followed by the data in autumn, the matching degrees in 11th band of TIRS. spring and summer were inferior. -e values of RMSE and Bias were relatively large. In winter, the inversion results of LST1, LST2, LST3, LST4, and LST5 were the closest to the 4.2. Results on Different POIs. It is well known that intensive measured data of meteorological stations, with difference of human activities affect the surrounding environment. In- less than 2 C, followed by the inversion results of LST6, dustrial production and emissions affect atmospheric which were 3 C higher than the measured data, and the transmittance, water vapor content, and upward and downward radiation. All LST inversion algorithms need difference between the inversion results of LST7, LST8, and LST9 and the measured data was about 7 C. In spring and these parameters to support the calculation. -erefore, it is summer, almost all the algorithm inversion results were necessary to explore whether the inversion results of similar higher than the measured temperature. At this time, the surface features with the same reflectivity in densely pop- closest to the measured temperature were LST1, LST2, LST4, ulated cities and outside cities close to the natural surface and LST5, with difference of about 5 C, followed by LST3 will be affected by various inversion algorithms and to what inversion results, about 10 C. -e inversion results of LST8 extent. -e four selected ground objects (20 POIs for each and LST9 were 20 C higher than the measured results, and ground object) were analyzed inside and outside the city. the reliability of the results will be seriously questioned. In Considering the statistical error, 10%–90% of the points were selected as the upper and lower whiskers of the box autumn, the inversion results of LST1, LST2, LST4, and LST5 were slightly lower than the measured data by 2 C, the map, and the upper and lower boundaries of the box rep- resented 25%–75% of the values, respectively. -e result is inversion results of LST3 were 5 C higher than the measured data, and the inversion results of LST6, LST7, LST8, and shown in Figure 4. LST9 were 8 C higher than the measured data. As shown in Figure 4, in the same season, the same LST From the perspective of the algorithm inversion results, inversion algorithm results show that the height of the box the inversion results of LST1, LST2, LST4, and LST5 were the map inside the city was less than the height of the box map closest to the measured data of meteorological stations, with outside the city. -e standard deviation of inversion results the inversion error within ±5 C, followed by the inversion of all LST inversion algorithms on POI inside the city was results of LST3 at about 10 C. -e worst are the inversion less than the standard deviation of inversion results outside results of LST6, LST7, LST8, and LST9, with the inversion the city. On the one hand, the main reason may be the urban heat island effect and on the other hand it may be the error around 20 C. In general, in the season with low 8 Advances in Meteorology 5 50 –5 –10 –15 MEASURED LST4 LST7 MEASURED LST4 LST7 LST1 LST5 LST8 LST1 LST5 LST8 LST2 LST6 LST9 LST2 LST6 LST9 LST3 LST3 (a) (b) MEASURED LST4 LST7 MEASURED LST4 LST7 LST1 LST5 LST8 LST1 LST5 LST8 LST2 LST6 LST9 LST2 LST6 LST9 LST3 LST3 (c) (d) Figure 2: Inversion results of 9 LST inversion algorithms at 20 meteorological stations in different seasons. (a) 2017-01-31 (winter), (b) 2017- 05-17 (spring), (c) 2017-7-10 (summer), and (d) 2017-10-30 (autumn). influence of intensive human activities inside the city on the inversion results will be different in different outer areas. As parameters of the inversion algorithm. In addition, in a a result, the inversion results of land surface temperature are relatively narrow space, using the same estimated parame- relatively scattered. ters for calculation, the inversion results are relatively Both inside and outside the city, the height of the water concentrated. However, the outer space of the city is rela- box chart is minimal. -e POI inversion temperature dif- tively broad, which belongs to the natural surface. -ere are ference between inside the city and outside the city is about differences in estimated atmospheric parameters, and the 1 C in autumn. -e average inversion temperature of POI Temperature (°C) Temperature (°C) Shunyi Shunyi Haidian Haidian Yanqing Yanqing Foyeding Foyeding Tanghekou Tanghekou Miyun Miyun Huairou Huairou Shangdianzi Shangdianzi Pinggu Pinggu Tongzhou Tongzhou Chaoyang Chaoyang Changping Changping Zhaitang Zhaitang Mentougou Mentougou Beijing Beijing Shijingshan Shijingshan Fengtai Fengtai Daxing Daxing Fangshan Fangshan Xiayunling Xiayunling Temperature (°C) Temperature (°C) Shunyi Shunyi Haidian Haidian Yanqing Yanqing Foyeding Foyeding Tanghekou Tanghekou Miyun Miyun Huairou Huairou Shangdianzi Shangdianzi Pinggu Pinggu Tongzhou Tongzhou Chaoyang Chaoyang Changping Changping Zhaitang Zhaitang Mentougou Mentougou Beijing Beijing Shijingshan Shijingshan Fengtai Fengtai Daxing Daxing Fangshan Fangshan Xiayunling Xiayunling Advances in Meteorology 9 Table 7: Root Mean Square Error (RMSE), average bias, slope, intercept, and R of LST inversion and field measured temperature. Season Model ID RMSE Bias Slope Intercept R LST1 0.67 −0.15 0.94 0.04 0.775 LST2 3.24 −0.73 0.94 −0.54 0.772 LST3 18.89 4.22 0.91 4.52 0.775 LST4 0.27 −0.06 0.98 0.01 0.794 Winter LST5 3.23 −0.72 0.94 −0.53 0.772 LST6 13.54 3.03 1.16 2.51 0.738 LST7 11.80 2.64 1.89 −0.24 0.771 LST8 22.42 5.01 0.93 5.24 0.776 LST9 19.20 4.29 0.93 4.52 0.772 LST1 10.18 2.28 1.42 −12.30 0.535 LST2 9.54 2.13 1.38 −11.03 0.531 LST3 39.09 8.74 1.33 −2.57 0.531 LST4 12.33 2.76 1.49 −14.05 0.472 Spring LST5 9.79 2.19 1.38 −11.02 0.530 LST6 82.87 18.53 2.11 −19.69 0.469 LST7 63.14 14.12 1.04 12.86 0.055 LST8 91.98 20.57 1.44 5.31 0.532 LST9 90.20 20.17 1.52 2.28 0.561 LST1 11.12 2.49 1.32 −8.99 0.459 LST2 11.49 2.57 1.28 −7.39 0.498 LST3 65.02 14.54 1.16 8.88 0.450 LST4 11.47 2.57 1.33 −9.16 0.458 Summer LST5 11.86 2.65 1.28 −7.52 0.499 LST6 84.98 19.00 1.49 1.65 0.411 LST7 72.96 16.32 1.36 3.32 0.121 LST8 53.96 12.07 1.35 0.32 0.465 LST9 55.89 12.50 1.25 0.28 0.500 LST1 1.68 −0.38 0.69 4.35 0.592 LST2 1.63 −0.37 0.72 3.89 0.522 LST3 12.74 2.85 0.66 7.94 0.591 LST4 0.45 −0.10 0.70 4.36 0.607 Autumn LST5 1.53 −0.34 0.72 3.90 0.519 LST6 27.62 6.17 0.93 7.30 0.459 LST7 21.57 4.82 0.52 12.06 0.055 LST8 32.28 7.22 0.70 11.81 0.599 LST9 31.58 7.06 0.74 10.97 0.573 inside the city is about 3 C higher than that outside the city vegetation pixels, their reflectivity is inconsistent, which may in other seasons. It shows that the standard deviation of all lead to specific errors. LST inversion algorithms on POI of the water body is -e box graph of soil POI temperature retrieved by the minimal, which also conforms to the characteristics of the SW (LST6 and LST7) algorithm is relatively large; in par- water body being relatively stable, and LSE is close to the ticular, the LST7 algorithm is very obvious. -e height of the black body. box graph of soil POI temperature retrieved by other al- Surprisingly, the height of the box chart of vegetation in gorithms is relatively tiny, with a difference of about 4 C all categories is considerable. Among all the POI inversion outside the city and 3 C inside the city. It shows that the temperatures of vegetation outside the city in spring, the POI standard deviation of all LST inversion algorithms on soil inversion temperature difference of vegetation outside the POI is minimal. It is worth noting that, in the selection of soil city is about 5 POI in this paper, the soil is dry and bright areas were C (this value does not count LST7 because it is precarious). -e POI inversion temperature difference of selected instead of the dark soil with high moisture content other vegetations is 8 C. It indicates that the standard de- to avoid the error caused by the significant difference in viation of all LST inversion algorithms on vegetation POI is reflectivity. substantial, indicating that inversion algorithms are more Except for LST6 and LST7, the height of the box graph of sensitive to vegetation inversion, which will be explained in building POI inversion by other LST inversion algorithms is ° ° detail in the next section. In addition, it is also possible that small, with a difference of about 5 C outside the city and 3 C when selecting pure pixels of vegetation, the vegetation types inside the city, indicating that the standard deviation of of pixels are different. Some pixels are grassland, some are inversion results of all LST inversion algorithms on building woodland, and some are shrubs. Although these are all POI is also small. 10 Advances in Meteorology 12 12 12 12 y=0.98x+0.01 y=0.94x–0.54 10 y=0.94x+0.04 10 10 10 2 R =0.794 R =0.772 8 8 8 8 R =0.775 6 6 6 6 y=0.91x+4.52 2017-01-31 4 4 4 4 2 2 2 2 R =0.775 0 0 0 0 –2 –2 –2 –2 –4 –4 –4 –4 –4 –2 0 2 4 6 8 10 12 –4 –2 0246 8 10 12 –4 –2 0246 810 12 –4 –2 0 2 4 6 810 12 MEASURED MEASURED MEASURED MEASURED 12 12 12 12 12 y=0.94x–0.53 10 10 10 10 10 R =0.772 8 8 8 8 8 6 6 6 6 6 4 4 4 y=1.89x–0.24 4 y=0.93x+5.24 4 y=0.93x+4.52 y=1.169x+2.51 2 2 2 2 2 2 2 2 R =0.771 R =0.776 R =0.772 R =0.738 0 0 0 0 0 –2 –2 –2 –2 –2 –4 –4 –4 –4 –4 –4 –2 0 2 4 6 8 10 12 –4 –2 0246 8 10 12 –4 –2 0246 8 10 12 –4 –2 0246 810 12 –4 –2 0 2 4 6 810 12 MEASURED MEASURED MEASURED MEASURED MEASURED 60 60 60 60 y=1.42x–12.30 y=1.33x–2.57 y=1.49x–14.05 55 55 y=1.38x–11.03 55 55 2 2 2 2 50 R =0.535 50 R =0.531 50 R =0.531 50 R =0.472 45 45 45 45 2017-05-07 40 40 40 40 35 35 35 35 30 30 30 30 25 25 25 25 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 MEASURED MEASURED MEASURED MEASURED 60 60 60 60 60 y=2.11x–19.69 y=1.38x–11.02 55 55 55 y=1.04x+12.86 55 y=1.44x+5.31 55 y=1.52x+2.28 2 2 2 R =0.469 R =0.530 50 50 50 R =0.055 50 R =0.532 50 R =0.561 45 45 45 45 45 40 40 40 40 40 35 35 35 35 35 30 30 30 30 30 25 25 25 25 25 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 MEASURED MEASURED MEASURED MEASURED MEASURED 66 66 y=1.28x–7.39 66 66 y=1.16x+8.88 y=1.32x–8.99 2 y=1.33x–9.16 60 2 60 R =0.498 60 60 R =0.450 2 R =0.459 R =0.458 54 54 54 54 2017-07-10 48 48 48 48 42 42 42 42 36 36 36 36 30 30 30 30 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 MEASURED MEASURED MEASURED MEASURED 66 66 66 66 66 y=1.28x–7.52 y=1.49x+1.65 y=1.36x+3.32 y=1.35x+0.32 y=1.25x+0.28 60 60 60 2 60 60 R =0.499 2 R =0.411 2 R =0.121 R =0.500 R =0.465 54 54 54 54 54 48 48 48 48 48 42 42 42 42 42 36 36 36 36 36 30 30 30 30 30 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 MEASURED MEASURED MEASURED MEASURED MEASURED 28 28 28 28 y=0.69x+4.35 y=0.72x+3.89 y=0.66x+7.94 y=0.70x+4.36 2 2 2 2 24 R =0.592 24 R =0.522 24 R =0.591 24 R =0.607 20 20 20 20 2017-10-30 16 16 16 16 12 12 12 12 8 8 8 8 812 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 MEASURED MEASURED MEASURED MEASURED 28 28 28 28 28 y=0.72x+3.90 24 R =0.519 24 24 24 24 20 20 20 20 20 y=0.93x+7.30 y=0.52x+12.56 y=0.70x–11.81 y=0.74x+10.97 2 2 16 16 R =0.459 16 16 16 R =0.599 R =0.055 R =0.57 12 12 12 12 12 8 8 8 8 8 812 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 MEASURED MEASURED MEASURED MEASURED MEASURED Figure 3: LST inversion results are fitted with the measured temperature. LST5 LST5 LST5 LST5 LST6 LST1 LST6 LST1 LST6 LST1 LST6 LST1 LST7 LST2 LST7 LST2 LST7 LST2 LST7 LST2 LST8 LST3 LST8 LST3 LST8 LST3 LST8 LST3 LST9 LST4 LST9 LST4 LST9 LST4 LST9 LST4 Advances in Meteorology 11 2017-01-31 2017-05-07 2017-07-10 2017-10-30 –5 40 –10 –15 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 10 60 70 30 60 25 4 50 20 2 30 40 15 20 30 10 –2 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 50 10 35 0 20 –5 –10 20 5 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 70 80 35 70 30 10 60 60 25 50 20 40 15 30 10 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 60 40 60 35 50 40 6 15 2 20 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 25 70 30 15 60 10 20 5 40 –5 30 10 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 70 50 15 40 10 30 5 20 0 10 –5 30 30 0 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 70 80 40 50 25 30 30 10 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 Figure 4: Inversion results of POI inside and outside the city by various algorithms. Inside-build Outside-build Inside-soil Outside-soil Inside-veg Outside-veg Inside-water Outside-water Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) 12 Advances in Meteorology upline radiation on the RTE (LST1 and LST2) algorithm is 4.3. Sensitivity Analysis of Model Parameters. -is paper adopts the control variable method to conduct sensitivity logarithmic. However, the curvature of the curve is min- imal. With the increase of upline radiation, the surface analysis for different parameters to explore the sensitivity of various LST inversion algorithms to parameters. First, temperature retrieved by the RTE algorithm also decreases. assume all the DN values on the image are the same (in As shown in Figure 5(e), the influence of downward ra- this study, take the average of the band after atmospheric diation on the SC and RTE algorithm is linear. As correction; band 10 is set to 10.4 and band 11 is set to 9.4). downward radiation increases by 1 unit, the inversion land -en select a group of parameters as variables, and the surface temperature decreases by 0.2 units. As shown in rest as invariants to control the variation of variable Figure 5(f), the influence of atmospheric water vapor content on LST3 and LST6 is linear. As the atmospheric parameters for sensitivity analysis. Finally, the sensitivity of the parameters was analyzed. As the data with good water vapor content increases by 0.1 units, the surface temperature retrieved by LST3 and LST6 increases by about weather conditions were selected in the remote sensing image screening previously, the interval of atmospheric 0.3 units. -e influence of atmospheric water vapor content on LST7 is nonlinear. With the increase of atmospheric transmittance is set as 0.5∼1.0 with a step size of 0.01 unit, and the interval of surface reflectance is set as 0.9∼1.0 water vapor, the LST retrieved by LST7 also increases, but with a step size of 0.01 unit. -e interval of DNVI is set as the saturated phenomenon appears after 2.5. -is result is −0.2∼1.0 with a step size of 0.01 units. -e interval of consistent with the result of Jimenez et al. (2010). In upward and downward radiation is set as 0∼5, and the Figure 5(g), the average land surface temperature param- step size is set as 0.1 units; the interval of atmospheric eters required in the MW algorithm are presented. It can be water vapor content is set as 0–2.5, and the step size is set seen that, with the increase of the average land surface temperature by 1 unit, the inversion of the land surface as 0.1 units; the interval of average surface temperature is set as 0–100, and the step size is set as 1 unit. -e influence temperature by MW decreases by 0.01 units. of each parameter on the results in different algorithms was obtained by controlling variables. For the compa- 5.Discussion rability of the results, LST was normalized, as shown in Figure 5. NASA’s ACPC obtains the atmospheric parameters required It can be seen from Figure 5(a) that the influence of for the LST inversion method based on the MODTRAN atmospheric transmittance on the inversion algorithms of radiation transfer code. It is not possible to find atmospheric RTE, SC, and MW is a logarithmic function. With the in- profiles in situ (radiosonde data, etc.) at any time and any crease of atmospheric transmittance, the inversion of land place. -erefore, even if this use (simulating atmospheric surface temperature gradually decreases. -e atmospheric profiling information with ACPC) affects the accuracy of the transmittance has a minor effect on the curvature of RTE method, it is clear from our results and the literature (LST1 and LST2) and a more significant effect on the cur- [8, 10, 14, 18, 20, 22–24, 29–33] that NASA’s ACPC provides vature of SC (LST4 and LST5) and MW (LST8 and LST9). an excellent and effective simulation. -e development of Figure 5(b) shows that the influence of specific emissivity on the LST inversion method has its source of error, as it in- all algorithms is linear. As the specific emissivity increases, volves parameterization steps for inverting the coefficients the land surface temperature decreases linearly. When the and estimating some initial parameters. -erefore, it is emissivity increases by 0.1 units, the inversion surface necessary to compare the errors of various algorithms in the temperature decreases by 1 unit. Figure 5(c) shows that the system. In general, the inversion results of the RTE algo- influence of NDVI on LST1, LST3, LST4, and LST8 is rithm and SC algorithm calculated by LSE1 and LSE2 have piecewise linear. Segmentation is because, during parameter an excellent fitting effect with the measured data. -e in- setting, specific emissivity is segmenting according to NDVI version results of MW are higher compared to the measured mandatory threshold. Specific emissivity is set as a fixed temperature. -e inversion effect of SW is not particularly value for parts with NDVI less than 0.05 and DNVI greater ideal, which may be due to the instability of radiation cal- than 0.6. -e inversion temperature is also a fixed value. For ibration at band 11 of TIRS. Hanqiu [42] used SC (LST3 and the part with DNVI greater than 0.05 and less than 0.6, the LST4) and SW (LST8 and LST9) algorithms to invert the LST specific emissivity changed with the change of NDVI. -e in Fuzhou on August 4, 2013, and concluded that the SC inversion temperature decreased linearly with the increase of result was better than the SW result. -e inversion error of NDVI. -e influence of NDVI on LST2, LST5, and LST9 is LST4 was −2.77 C. However, since the atmospheric water also segmenting. However, when NDVI is more than 0.05, vapor content parameter was more than 4 on that day, the the influence of NDVI on LST presents a quadratic function; error of LST3 is −17.44 C, and the inversion errors of LST8 ° ° when NDVI is equal to 0.4, the function has a minimum and LST9 are −13.61 C and −5.75 C, respectively. Wang [41] value; when NDVI is more than 0.05 and less than 0.4, the compared the sensitivity analysis of the three algorithms, SC, inversion LST decreases the NDVI. When the NDVI is more MW, and SW, and found that SW had the lowest sensitivity than 0.4, the inversion LST increases the NDVI. to the input parameter error. Asia Siddiqui et al. [19] selected In Figure 5(d), the influence of uplink radiation on the the data of two landscapes in January and April to study the SC (LST4 and LST5) algorithm is linear. With the increase heat island in Bangalore, India, based on the RTE method of uplink radiation, the land surface temperature inversion and found that the relationship between LST and NDVI was by the SC algorithm decreases linearly. -e influence of firstly positively correlated and then negatively correlated. Advances in Meteorology 13 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0.5 0.6 0.7 0.8 0.9 1.0 0.90 0.92 0.94 0.96 0.98 1.00 –0.2 0.0 0.2 0.4 0.6 0.8 1.0 Transmissivity Emissivity NDV1 LST1 LST5 LST1 LST4 LST8 LST1 LST5 LST2 LST8 LST2 LST5 LST9 LST2 LST8 LST4 LST9 LST3 LST7 LST3 LST9 LST4 (a) (b) (c) 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0123 45 0123 45 0.0 0.5 1.0 1.5 2.0 2.5 3.0 up_radiance down_radiance water_vapour LST1 LST4 LST1 LST4 LST3 LST2 LST5 LST2 LST5 LST6 LST7 (d) (e) (f) 1.0 0.8 0.6 0.4 0.2 0.0 020 40 60 80 100 Ta LST8 LST9 (g) Figure 5: Sensitivity analysis of parameters of various LST inversion algorithms. -e vertical axis represents the normalized LST, and the horizontal axis represents the parameters of each variable. -e units of parameters are given in Section 3. -e various LST inversion algorithms are most stable in temperature from rural areas and found that RTE and SC winter, and the stability becomes worse in other seasons. presented similar results regardless of the season, while MW Aliihsan Sekertekin and Stefania Bonafoni [40] used differed from RTE and SC for all seasons, especially in Landsat-5, Landsat-7, and Landsat-8 data to retrieve surface summer. -e parameter error had little influence on the Normalized temperature Normalized temperature Normalized temperature Normalized temperature Normalized temperature Normalized temperature Normalized temperature 14 Advances in Meteorology parameters calculation, calculation results by ACPC dif- results in the humid environment. MW and SC methods are more sensitive to the error of input parameters, especially ference are not very big; however, when we use the estimated atmospheric profile parameters for further calculation, there under hot and humid conditions. -is result is consistent with our inversion results in urban areas. On the one hand, will be error propagation. Sensitivity analysis of the LST with the increase of temperature, the properties of ground inversion algorithm can help researchers to understand the objects (specific heat capacity and specific emissivity) have source of error and reduce the error as much as possible. distinct differences, leading to a significant difference in reflectivity and then resulting in a significant difference in 6.Conclusions the inversion of surface temperature. On the other hand, considering the errors of surface temperature inversion of In all seasons, the inversion results of the nine LST algo- similar surface features POI from inside and outside the city, rithms are in complete agreement with the graph trends of the increase of temperature will have a particular influence the measured data of all meteorological stations. In winter, on the parameters of the LST inversion algorithm, which the inversion results of the nine algorithms have the highest may lead to errors in the inversion results. matching degree with the measured data of meteorological -e stability of various inversion algorithms for tem- stations, and the RMSE and Bias values are relatively small. perature inversion inside and outside the city is tested. -e -e second is autumn’s LST1, LST2, LST4, and LST5 in- results show that the standard deviations of inversion results version results of the nearest meteorological station, with for POIs of LST inversion algorithms in urban areas are inversion error of ±5 C; the following inversion result of smaller than those of inversion results outside the city. On LST3 is about 10 C; the worst are the inversion results of the one hand, the main reason may be that the intensive LST6, LST7, LST8 and LST9, with inversion error of about human activities in cities have a particular influence on land 20 C. In the low-temperature season, the error of the al- surface temperature inversion parameters. On the other gorithm inversion results is less than that of the measured hand, the inversion results of the same reflectivity charac- data. With the increase of temperature, the inversion results teristics will be relatively concentrated when the same es- of each inversion algorithm are all higher compared to the timated parameters are used for calculation in a very narrow measured data. -e inversion errors of LST1, LST2, LST4, space inside the city. However, the space outside the city is and LST5 are small, followed by LST3, LST6, LST7, LST8, relatively broad and belongs to the natural surface, so the and LST9. estimated atmospheric parameters in different regions will According to the relevant results of the nine LST in- be different, which may lead to the relatively scattered in- version results and the measured temperature, the fitting version results of land surface temperature. results in winter are all relatively ideal, with R greater than It is essential to analyze the sensitivity/uncertainty of the 0.73. In other seasons, the fitting effect is general, and R is LST inversion algorithm by considering all the input pa- about 0.5. In general, the inversion results of the RTE al- rameters. Sekertekin and Bonafoni’s [43] simultaneous gorithm and SC algorithm calculated by LSE1 and LSE2 have ground-based LST measurements were collected from At- an excellent fitting effect with the measured data, followed by mospheric Radiation Measurement (ARM) and Surface the inversion results of the SC algorithm calculated by w. By Radiation Budget Observing Network (SURFRAD) stations, analyzing the stability of various inversion algorithms inside located at different rural environments of the United States and outside the city, it can be seen that the standard de- to examine the efficiency of different LST algorithms, using viation of inversion results of each LST inversion algorithm both daytime and nighttime Landsat-8 data and in situ for POI inside the city is less than that outside the city. measurements. Concerning the in situ sensitivity results, the -e influence of atmospheric transmittance on RTE, SC, effects on LST of the uncertainty of the downwelling and and MW inversion algorithms is a logarithmic function. upwelling radiance were almost identical in daytime and With the increase of atmospheric transmittance, the in- nighttime. -e accuracy of the LST retrieval methods for version surface temperature decreases gradually. -e in- daytime Landsat-8 data varied between 2.17 K Root Mean fluence of emissivity on all algorithms is linear. -e Square Error (RMSE) and 5.47 K RMSE considering all LST inversion of LST decreases by 1 for every 0.1 increase of methods and LSE models. MW with two different LSE emissivity. -e influence of NDVI on LST1, LST3, LST4, and models presented the best results for the daytime. Wang LST8 is piecewise linear, and the influence on LST2, LST5, et al. [44] presented an improved Mono Window (IMW) and LST9 is in the form of a quadratic function. When NDVI algorithm for LSTretrieval from the Landsat-8 TIRS band 10 is equal to 0.4, the function has a minimum value. When data, and sensitivity analysis conducted for the IMW al- NDVI is more than 0.05 and less than 0.4, the inversion land gorithm revealed that the possible error in estimating the surface temperature decreases. When NDVI is more than required atmospheric water vapor content has the most 0.4, the inversion LST increases; this is the opposite of the significant impact on the probable LSTestimation error. -is previous results. -e effect of uplink radiation on the SC result is similar to our inversion results of LST3, LST6, and (LST4 and LST5) algorithm is linear. With the increase of LST7. Although from our results and literature we can see uplink radiation, the LST retrieved by the SC algorithm clearly that NASA ACPC provides a satisfying and effective decreases linearly. -e influence of upline radiation on the simulation, in our precise and fuzzy provided coordinates, RTE (LST1 and LST2) algorithm is logarithmic. With the respectively, phenology, area type (urban and rural), and increase of upline radiation, the land surface temperature satellite altitude parameters such as atmospheric profile retrieved by the RTE algorithm decreases. -e influence of Advances in Meteorology 15 [8] D. Zhou, J. Xiao, and S. Bonafoni, “Satellite remote sensing of downslope radiation on SC and RTE algorithms is linear. surface urban heat islands: progress, challenges, and per- With the increase of downslope radiation by 1 unit, the spectives,” Remote Sensing, vol. 11, no. 1, 2018. inversion LST decreases by 0.2 units. -e surface temper- [9] C. Keeratikasikorn and S. Bonafoni, “Urban heat island ature retrieved by LST3 and LST6 increases by about 0.3 analysis over the land use zoning plan of bangkok by means of units for every 0.1 unit increase of atmospheric water vapor landsat 8 imagery,” Remote Sensing, vol. 10, no. 3, 2018. content. -e influence of atmospheric water vapor content [10] J. A. Sobrino, R. Oltra-Carrio, ´ and G. Soria, ` “Evaluation of the on LST7 is nonlinear. With the increase of atmospheric surface urban heat island effect in the city of Madrid by water vapor, the surface temperature retrieved by LST7 also thermal remote sensing,” International Journal of Remote increases, but, after 2.5 units, it shows saturation Sensing, vol. 34, no. 9-10, pp. 3177–3192, 2012. phenomenon. [11] J. Tan, Y. Zheng, X. Tang et al., “-e urban heat island and its impact on heat waves and human health in Shanghai,” In- ternational Journal of Biometeorology, vol. 54, no. 1, pp. 75–84, Data Availability [12] C. Maffei, S. Alfieri, and M. Menenti, “Relating spatiotem- Some or all data, models, or codes that support the findings poral patterns of forest fires burned area and duration to of this study are available from the corresponding author diurnal land surface temperature anomalies,” Remote Sensing, upon reasonable request. vol. 10, no. 11, 2018. [13] Z. Wan, P. Wang, and X. Li, “Using MODIS land surface temperature and normalized difference vegetation index Conflicts of Interest products for monitoring drought in the southern great plains, -e authors declare that there are no conflicts of interest USA,” International Journal of Remote Sensing, vol. 25, no. 1, pp. 61–72, 2010. regarding the publication of this paper. [14] J. Sun, G. D. Salvucci, and D. Entekhabi, “Estimates of evapotranspiration from MODIS and AMSR-E land surface Acknowledgments temperature and moisture over the Southern Great Plains,” Remote Sensing of Environment, vol. 127, pp. 44–59, 2012. -e National Key Research and Development Project of [15] J. Qin, S. Liang, R. Liu, H. Zhang, and B. Hu, “A weak- China (no. 2017YFA0605004) and Basic R&D Special Fund constraint-based data assimilation scheme for estimating of Central Government for Non-Profit Research Institutes surface turbulent fluxes,” IEEE Geoscience and Remote Sensing (Grant/Award HKY-JBYW-2021-02) are acknowledged. Letters, vol. 4, no. 4, pp. 649–653, 2007. [16] B. Yong, L. Ren, Y. Hong et al., “First evaluation of the cli- matological calibration algorithm in the real-time TMPA References precipitation estimates over two basins at high and low lat- itudes,” Water Resources Research, vol. 49, no. 5, [1] M. Anderson, J. Norman, W. Kustas, R. Houborg, P. Starks, pp. 2461–2472, 2013. and N. Agam, “A thermal-based remote sensing technique for [17] O. Hall, G. Falorni, and R. L. Bras, “Characterization and routine mapping of land-surface carbon, water and energy quantification of data voids in the shuttle radar topography fluxes from field to regional scales,” Remote Sensing of En- mission data,” IEEE Geoscience and Remote Sensing Letters, vironment, vol. 112, no. 12, pp. 4227–4241, 2008. vol. 2, no. 2, pp. 177–181, 2005. [2] P. Dash, F.-M. Gottsche, ¨ F.-S. Olesen, and H. Fischer, “Land [18] J. C. Price, “Estimating surface temperatures from satellite surface temperature and emissivity estimation from passive thermal infrared data-A simple formulation for the atmo- sensor data: theory and practice-current trends,” Interna- spheric effect,” Remote Sensing of Environment, vol. 13, no. 4, tional Journal of Remote Sensing, vol. 23, no. 13, pp. 2563– pp. 353–361, 1983. 2594, 2010. [19] A. Siddiqui, G. Kushwaha, A. Raoof, P. A. Verma, and [3] R. E. Dickinson, “Land surface processes and climate-surface Y. Kant, “Bangalore: urban heating or urban cooling?” 7e albedos and energy balance,” Advances in Geophysics, in Egyptian Journal of Remote Sensing and Space Science, vol. 24, Proceedings of a Symposium Commemorating the Two-Hun- no. 2, pp. 265–272, 2021. dredth Anniversary of the Academy of Sciences of Lisbon, [20] J. C. Jimenez-Muñoz and J. A. Sobrino, “A single-channel pp. 305–353, Lisbon, Portugal, October 1983. algorithm for land-surface temperature retrieval from ASTER [4] D. Tong, Q. Zhang, Y. Zheng et al., “Committed emissions data,” IEEE Geoscience and Remote Sensing Letters, vol. 7, from existing energy infrastructure jeopardize 1.5 C climate no. 1, pp. 176–179, 2010. target,” Nature, vol. 572, no. 7769, pp. 373–377, 2019. [21] J. C. Jimenez-Muñoz, ´ J. Cristobal, J. A. Sobrino et al., “Re- [5] G. P. Peters, R. M. Andrew, J. G. Canadell et al., “Carbon vision of the single-channel algorithm for land surface dioxide emissions continue to grow amidst slowly emerging temperature retrieval from landsat thermal-infrared data,” climate policies,” Nature Climate Change, vol. 10, no. 1, pp. 3–6, 2019. IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 1, pp. 339–349, 2009. [6] B. Talukder, R. Matthew, G. W. Vanloon, M. J. Bunch, K. W. Hipel, and J. Orbinski, “Melting of Himalayan glaciers [22] Z. Qin, A. Karnieli, and P. Berliner, “A mono-window al- gorithm for retrieving land surface temperature from Landsat and planetary health,” Current Opinion in Environmental Sustainability, vol. 50, pp. 98–108, 2021. TM data and its application to the Israel-Egypt border re- gion,” International Journal of Remote Sensing, vol. 22, no. 18, [7] J. Xin, X. Sun, and L. Liu, “Quantifying the contribution of climate and underlying surface changes to alpine runoff al- pp. 3719–3746, 2010. [23] K. Mao, Z. Qin, J. Shi, and P. Gong, “A practical split-window terations associated with glacier melting,” Hydrological Pro- cesses, vol. 35, no. 3, 2021. algorithm for retrieving land-surface temperature from 16 Advances in Meteorology MODIS data,” International Journal of Remote Sensing, [40] A. Sekertekin and S. Bonafoni, “Land surface temperature retrieval from landsat 5, 7, and 8 over rural areas: assessment vol. 26, no. 15, pp. 3181–3204, 2007. [24] Z. M. W. a. J. Dozier, “A generalized split- window algorithm of different retrieval algorithms and emissivity models and toolbox implementation,” Remote Sensing, vol. 12, no. 2, 2020. for retrieving land-surface temperature from space,” IEEE Transactions on Geoscience and Remote Sensing, vol. 34, no. 4, [41] L. Wang, Y. Lu, and Y. Yao, “Comparison of three algorithms for the retrieval of land surface temperature from landsat 8 pp. 892–905, 1996. [25] O. Rozenstein, Z. Qin, Y. Derimian, and A. Karnieli, “Der- images,” Sensors, vol. 19, no. 22, 2019. [42] X. Hanqiu, “Retrieval of reflectivity and surface temperature ivation of land surface temperature for Landsat-8 TIRS using a from the newly launched Landsat 8 satellite,” Chinese journal split window algorithm,” Sensors, vol. 14, no. 4, of geophysice, vol. 58, no. 3, pp. 741–747, 2015. pp. 5768–5780, 2014. [43] A. Sekertekin and S. Bonafoni, “Sensitivity analysis and [26] F.-M. G. Prasanjit Dash, F.-S. Olesen, and H. Fischer, “Re- validation of daytime and nighttime land surface temperature trieval of land surface temperature and emissivity from sat- retrievals from Landsat 8 using different algorithms and ellite data: physics, theoretical limitations and current emissivity models,” Remote Sensing, vol. 17, no. 12, p. 2776, methods,” Journal of the Indian Society of Remote Sensing, vol. 29, no. 1, pp. 24–30, 2001. [44] F. Wang, Z. Qin, C. Song, L. Tu, A. Karnieli, and S. Zhao, “An [27] Y. Yu, D. Tarpley, J. L. Privette et al., “Validation of GOES-R improved mono-window algorithm for land surface tem- satellite land surface temperature algorithm using SURFRAD perature retrieval from Landsat 8 thermal infrared sensor ground measurements and statistical estimates of error data,” Remote Sensing, vol. 4, no. 7, pp. 4268–4289, 2015. properties,” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 3, pp. 704–713, 2012. [28] S. Pat and J. Chavez, “Image-based atmospheric corrections - revisited and improved,” Photogrammetric Engineering & Remote Sensing, vol. 62, no. 9, pp. 1025–1036, 1996. [29] X. Hanqiu, “Evaluation of two absolute radiometric nor- malization algorithms for pre-processing of landsat imagery,” Journal of China University of Geosciences, vol. 17, no. 2, pp. 146–157, 2006. [30] J. A. Sobrino, J. C. Jimenez-Munoz, G. Soria et al., “Land surface emissivity retrieval from different VNIR and TIR sensors,” IEEE Transactions on Geoscience and Remote Sensing, vol. 46, no. 2, pp. 316–327, 2008. [31] Z. Qin, W. Li, B. Xu, Z. Chen, and J. Liu, “-e estimation of land surface emissivity for Landsat TM6,” Remote sensing for land & resources, vol. 3, no. 3, pp. 28–36, 2004. [32] A. Julia, J. L. B. Barsi, and J. R. Schott, “An atmospheric correction parameter calculator for a single thermal band earth-sensing instrument,” in IEEE International Geoscience & Remote Sensing Symposium, Yokohama, Japan, August 2003. [33] J. A. Barsi, J. J. Butler, and J. R. Schott, “Validation of a web- based atmospheric correction tool for single thermal band instruments,” Earth Observing Systems X, vol. 10, 2005. [34] N. K. Malakar, G. C. Hulley, S. J. Hook, K. Laraby, M. Cook, and J. R. Schott, “An operational land surface temperature product for landsat thermal data: methodology and valida- tion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 10, pp. 5717–5735, 2018. [35] Z. Zhang, G. He, M. Wang, T. Long, G. Wang, and X. Zhang, “Validation of the generalized single-channel algorithm using Landsat 8 imagery and SURFRAD ground measurements,” Remote Sensing Letters, vol. 7, no. 8, pp. 810–816, 2016. [36] C. Coll, V. Caselles, J. Galve et al., “Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS data,” Remote Sensing of Environment, vol. 97, no. 3, pp. 288–300, 2005. [37] X. Meng, J. Cheng, and S. Zhao, “Estimating land surface temperature from landsat-8 data using the NOAA JPSS en- terprise algorithm,” Remote Sensing, vol. 11, no. 2, 2019. [38] M. Wang, Z. Zhang, and T. Hu, “A practical single-channel algorithm for land surface temperature retrieval: application to landsat series data,” Journal of Geophysical Research: At- mosphere, vol. 33, 2018. [39] Z.-L. Li, B.-H. Tang, H. Wu et al., “Satellite-derived land surface temperature: current status and perspectives,” Remote Sensing of Environment, vol. 131, pp. 14–37, 2013. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Advances in Meteorology Hindawi Publishing Corporation

Accuracy Evaluation and Parameter Analysis of Land Surface Temperature Inversion Algorithm for Landsat-8 Data

Advances in Meteorology , Volume 2021 – Sep 24, 2021

Loading next page...
 
/lp/hindawi-publishing-corporation/accuracy-evaluation-and-parameter-analysis-of-land-surface-temperature-bXWML9xLQv
Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2021 Jikang Wan et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
1687-9309
eISSN
1687-9317
DOI
10.1155/2021/9917145
Publisher site
See Article on Publisher Site

Abstract

Hindawi Advances in Meteorology Volume 2021, Article ID 9917145, 16 pages https://doi.org/10.1155/2021/9917145 Research Article Accuracy Evaluation and Parameter Analysis of Land Surface Temperature Inversion Algorithm for Landsat-8 Data 1 1,2 3 Jikang Wan , Min Zhu , and Wei Ding Hohai University, College of Computer and Information, Nanjing 211100, China Yellow River Institute of Hydraulic Research, Zhengzhou 450003, China University of Calgary, Department of Geomatics Engineering, Calgary AB T2N 1N4, Canada Correspondence should be addressed to Min Zhu; 190207050001@hhu.edu.cn Received 3 August 2021; Revised 29 August 2021; Accepted 8 September 2021; Published 24 September 2021 Academic Editor: Stefania Bonafoni Copyright © 2021 Jikang Wan et al. -is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Many researchers have developed a variety of land surface temperature (LST) inversion algorithms based on satellite data. -e main LST inversion algorithms include Radiative Transfer Equation (RTE), Single Channel (SC) algorithm, Mono Window (MW) algorithm, and Split Window (SW) algorithm. In this study, nine LST inversion algorithms were designed using Landsat-8 data and meteorological station data to test the inversion efficiency of different algorithms in different seasons and different locations. -e results show that the error of various LST inversion algorithms will increase with the rise of LST. R of the inversion results of ° ° each LST algorithm and the measured data are all greater than 0.73 C in winter and about 0.5 C in the other seasons. By analyzing the stability of various algorithms inside and outside the city, it is found that the stability of each LST inversion algorithm inside the city is better than that outside the city. For the same surface features, the inversion temperature inside the city is 3–5 C higher than that outside the city. In addition, the sensitivity of various inversion algorithms to parameters was also analyzed. -e influence of atmospheric transmittance on RTE, SC, and MW inversion algorithms is in logarithmic form. -e effect of emissivity on each algorithm is linear. -e influence of NDVI on the algorithms is mainly through the estimation of surface emissivity parameters to affect the inversion results. -e effect of ascending radiation on SC (LST4 and LST5) is linear and on RTE (LST1 and LST2) is logarithmic. -e effect of downslope radiation on SC and RTE is linear. -e influence of atmospheric water vapor content on SW (LST7) is nonlinear. method cannot represent the regional temperature. -e 1.Introduction timeliness of the method is insufficient. With the develop- Land surface temperature (LST) is an essential parameter in ment and research of space-based infrared remote sensing, it many research fields, such as ecology, climatology, urban is more and more popular to use satellite remote sensing to thermal environment, urban heat island, and hydrology retrieve regional and global surface temperatures. [1–3]. Especially in recent years, more attention has been In many satellite observation systems equipped with paid to carbon dioxide emissions [4, 5], polar climate thermal infrared sensors, Landsat satellites have become one of the most widely used satellites due to their observation warming and glacier melting [6, 7], urban heat island (UHI) [8–10], urban heatwave and human living health [11], forest history of more than 40 years. Since its launch on February fires [12], drought monitoring [13], soil evapotranspiration 11, 2013, the newest generation of Landsat-8 has been in [14], and precipitation and surface runoff [15, 16]. -e orbit for more than eight years. It has sent back a large traditional LST measurement method is based on the amount of Earth observation data at a rate of 550 images per temperature measurement data of meteorological stations day. Together with Landsat-7 ETM+, it forms a repeated for further spatial interpolation, which is a classical point- observation period with an interval of 8 days, which provides based measurement method [17]. Although this method has precious data for monitoring ecological environment high accuracy, the point-based temperature measurement changes on the Earth’s surface. Because of its advantages of 2 Advances in Meteorology high spatial resolution, large scanning field of view, and which seven multispectral bands have a spatial resolution of accessible data, researchers have applied it to further sci- 30 m. -rough resampling, the spatial resolution of the two entific research and ecological construction. thermal infrared bands is also 30 m. -e revisit period is 16 Since the launch of Landsat-8, various processing doc- days for global coverage. Since the surface reflectance is uments and parameters have been posted on the Landsat. affected by the atmospheric conditions at the time of image Many researchers have developed a variety of LST inversion acquisition and it is also a potential factor for studying the algorithms based on Landsat-8. -e main LST inversion surface thermal model, when screening the data, we select algorithms include the Radiative Transfer Equation (RTE) the imaging data under good weather condition, which was a [18, 19], Single Channel (SC) algorithm [20, 21], Mono clear day with less than 2% cloud cover, and use the thermal Window (MW) algorithm [22], and Split Window (SW) band data to retrieve the LST conduct research. In order to algorithm [21, 23–25]. In these surface temperature inver- investigate the accuracy of the Landsat-8 land surface sion algorithms, multiple basic parameters are needed for temperature inversion algorithm in season, the imaging input [26]. -e multiple parameters are estimated variables dates of January 31, 2017 (winter), May 7, 2017 (spring), July rather than standard ones. Accurate and reliable inversion 10, 2017 (summer), and October 30, 2017 (autumn), were algorithms should calculate the accuracy of the LST data selected. All images were named by the 8-digit number of used in the study to make the use of the product possible. If years, months, and day of the collection date for the con- the inversion algorithm has a significant systematic error or venience of recording. For example, the image on May 7, a significant instability error, the algorithm or product will 2017, as shown in Figure 1, was named 2017-05-07. not be popular among users [27]. -is study also collected and sorted the surface tem- In this study, LST inversion, spatial analysis, and mul- perature data of 20 meteorological stations in the study area. tiscale analysis methods were considered for Landsat data -e basic information of the meteorological stations is representing four seasons. -e commonly used LST inver- shown in Table 1. -e positions of these meteorological sion algorithms are evaluated. First of all, we designed nine stations in the study area are shown in the red dots in remote sensing inversion schemes for LST. We fitted the LST Figure 1. -e temperature data were collected from the China Meteorological Administration and measured si- calculated by the satellite with the measured temperature values at the ground stations at the same time. Based on the multaneously as remote sensing images. To study the ac- same scene remote sensing data, the accuracies of the nine curacy of the inversion algorithm of the LST inside and inversion algorithms were compared and analyzed. -e outside the city, points of interest (POIs) were established for accuracy of each LST inversion algorithm on the time scale four kinds of ground objects (water, vegetation, soil, and was analyzed separately on the images of different seasons. building), and 20 POIs were established for each ground -en, the inversion accuracies of all the LST inversion object. Pure pixels were selected for each point of interest methods on the regional scale (inside city and outside city) of from Landsat-8 multispectral data. High-resolution Google the same surface features are compared and analyzed. Fi- Earth images also examined the selection of all POIs. As nally, we test the dependence of each inversion algorithm on shown in Figure 1, those outside the city are represented by parameters by equal step size. -e aims of this study are as green dots, while yellow dots represent those inside the city. follows: (1) evaluate the errors of multiple inversion algo- rithms on the same scene data, (2) evaluate the error of each 3.Methods inversion algorithm in seasons (spring, summer, autumn, and winter), (3) evaluate the regional error of each inversion 3.1. Multispectral Image Processing. Using the Landsat-8 algorithm, and (4) perform dependence intensity and sen- website, the user can convert the Digital Numbers (DNs) of sitivity analysis of the evaluation algorithm on parameters. the image to the spectral radiance of the top of the atmo- sphere (TOA). By applying the equation 2.Study Areas and Data Compilation L � M Q + A , (1) λ L cal L 2.1. Study Areas. -e study area was selected as an area −2 −1 −1 where L (Watts·m ·srad µm ) refers to the TOA covered by Landsat-8 data (orbit number 123, 032). -e spectral radiance, M is the multiplicative rescaling factor of longitude and latitude of the data center are 40.3 N and the corresponding band, it can be obtained from the header 116.7 E. It mainly covers Beijing, the capital of China. -e file of the data with the field “Radiance_Mult_Band_x � ,” x climate type in the study area is semihumid and semiarid stands for band number. Q is the pixel values, and A is the cal L monsoon climate in the warm temperate zone, with four offset of the data. -e multispectral bands of Landsat-8 data distinct seasons, hot and rainy in summer and cold and dry can also be converted to the reflectivity of the top of the in winter. It provides a natural, convenient condition for atmosphere by the following equation: studying the difference of inversion algorithms on a seasonal ρ � M Q + A , (2) scale. λ ρ cal ρ where ρ is the reflectance of the top of the atmosphere at band λ without correction of the solar angle, M is the 2.2. Data Compilation. -is study contained four pieces of reflectivity adjustment factor of band λ, and it can be ob- Landsat-8 data based on USGS (United States Geological tained from the header file of the data with the field Survey, https://www.usgs.gov/). Landsat-8 has 11 bands, of Advances in Meteorology 3 115°30’0”E 116°0’0”E 116°30’0”E 117°0’0”E 117°30’0”E 118°0’0”E W E 010 20 40 60 80 Km 115°30’0”E 116°0’0”E 116°30’0”E 117°0’0”E 117°30’0”E 118°0’0”E Station Outer-ROI Inner-ROI Figure 1: Study areas. -e remote sensing data of “LC08_L1TP_123032_20170507_20170515_01_T1,” displayed in true color, is named 2017-05-07, with red dots representing the location of the meteorological station, green dots representing the points of interest outside the city, and yellow dots representing the points of interest inside the city. Table 1: Basic information of meteorological stations in the study area. Name Latitude Longitude Elevation (m) Land cover type ° ° Shunyi 40.13 N 116.61 E 28.6 Grassland ° ° Haidian 39.98 N 116.28 E 45.8 Grassland ° ° Yanqing 40.45 N 115.96 E 487.9 Natural vegetation mosaic ° ° Foyeding 40.60 N 116.13 E 1224.7 Natural vegetation mosaic ° ° Tanghekou 40.73 N 116.63 E 331.6 Natural vegetation mosaic ° ° Miyun 40.38 N 116.86 E 71.8 Grassland ° ° Huairou 40.37 N 116.63 E 75.7 Grassland ° ° Shangdianzi 40.65 N 117.11 E 293.3 Natural vegetation mosaic ° ° Pinggu 40.17 N 117.11 E 32.1 Grassland ° ° Tongzhou 39.85 N 116.75 E 19.8 Grassland ° ° Chaoyang 39.95 N 116.50 E 35.3 Grassland ° ° Changping 40.21 N 116.21 E 76.2 Grassland ° ° Zhaitang 39.96 N 115.68 E 440.3 Natural vegetation mosaic ° ° Mentougou 39.88 N 116.15 E 85.5 Grassland ° ° Beijing 39.80 N 116.46 E 31.3 Grassland ° ° Shijingshan 39.95 N 116.20 E 63 Grassland ° ° Fengtai 39.86 N 116.25 E 55.2 Grassland ° ° Daxing 39.71 N 116.35 E 37.5 Grassland ° ° Fangshan 39.76 N 116.20 E 48.9 Grassland ° ° Xiayunling 39.73 N 115.73 E 407.7 Grassland 39°30’0”N 40°0’0”N 40°30’0”N 41°0’0”N 39°30’0”N 40°0’0”N 40°30’0”N 41°0’0”N 4 Advances in Meteorology “Reflectance_Mult_Band_x � ,” where x stands for band where NDVI is the NDVI value of bare soil or areas soil number. A is the reflectivity adjustment parameter of band without vegetation coverage and NDVI is the NDVI value ρ veg λ, the header file of the data with the field of pixels completely covered by vegetation, namely, the “Reflectance_Add_Band_x � .” ρ can be further corrected NDVI value of pure vegetation pixels. Empirical values by equation (3) into atmospheric top reflectivity ρ : NDVI � 0.6 and NDVI � 0.05 were taken; that is, when λ veg soil the NDVI of a pixel is greater than 0.60, P value is 1.0, and ′ 􏼐M Q + A 􏼑 ρ cal ρ λ when NDVI is less than 0.05, P value is 0. (3) ρ � � , cos θ cos θ -e other method LSE2 is to firstly divide the surface z z into water, natural surface, and urban area according to the where θ is the solar zenith angle of the image center, and method proposed by Qin et al. [31] and then calculate the solar altitude angle θ can also be used, but sine function LSE for the three types of surface separately. -e calculation should be used. We used the solar altitude angle provided in is by the following equation: the image for our calculations. ⎧ ⎪ 0.995 water ⎫ ⎪ In this study, the COST model atmospheric correction ⎪ ⎪ ⎪ ⎪ ⎨ ⎬ algorithm of Pat and Chavez [28] is adopted to conduct 0.9625 + 0.0614P − 0.0461P natural surface ε � . v v ⎪ ⎪ ⎪ ⎪ Landsat-8 atmospheric correction. COST requires few pa- ⎪ ⎪ ⎩ 2 ⎭ 0.9589 + 0.086P − 0.0671P urban area rameters, and the correction is mainly carried out with the v v data of the image itself [29] by the following equation: (9) 􏽨M Q − Q 􏼁 + A 􏽩 ρ cal h ρ However, the concept of natural surfaces and urban (4) ρ � , λ cost areas is vague and cannot be quantitatively calculated. Based cos θ τ on remote sensing images of Beijing and its surrounding where Q is the modification value of atmospheric influence, h areas, we conducted local experiments and found that when which can be obtained by the darkest pixel method, NDVI is less than 0.05, it corresponds to water; when NDVI τ � cos[(90 − θ )π/180], and τ is the atmospheric trans- is greater than or equal to 0.05 or less than or equal to 0.6, it mittance estimated based on θ . After COST atmospheric corresponds to urban areas; and when NDVI is greater than correction, NDVI (normalized difference vegetation index) 0.6, it corresponds to the natural surface. -e calculation is is calculated by the following equation: in the following equation: ρ − ρ 􏼁 NIR COST red COST ⎧ ⎪ 0.995 NDVI < 0.05 ⎫ ⎪ NDVI � , (5) ⎪ ⎪ ⎪ ⎪ ⎨ ⎬ ρ + ρ 􏼁 NIR COST red COST 0.9625 + 0.0614P − 0.0461P 0.05 ≤ NDVI < 0.6 ε � . ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 2 ⎭ where ρ is the reflectance image of the near infrared nir COST 0.9589 + 0.086P − 0.0671P 0.6 ≤ NDVI v v band and ρ is the reflectance image of the red band. red COST (10) 3.2. 7ermal Infrared Image Processing. -e brightness temperature (BT) can be calculated using the following 3.4. Land Surface Temperature Inversion Algorithm. Based equation: on four commonly used LST inversion methods (RTE, SC, 2 MW, and SW), a total of nine schemes were designed based T � , (6) ln K /L 􏼁 + 1􏼁 on different parameters such as LSE and water vapor content 1 λ (w). -e input atmospheric parameters such as downwelling where T is the BT; K and K for the Landsat-8 band 10 are 1 2 radiance (L↓, W/m /sr/μm), upwelling radiance −2 −1 −1 774.89 Watts·m ·srad µm and 1321.08 K, respectively. 2 − 2 (L↑, W/m /sr/μm), water vapor content (g · cm ), and L is the thermal infrared band after radiation calibration atmospheric transmittance (τ) were calculated using the and COST atmospheric correction. Atmospheric Correction Parameter Calculator (ACPC, https://atmcorr.gsfc.nasa.gov) developed by National Aeronautics and Space Administration (NASA) of the US 3.3. Estimation of Land Surface Emissivity. In this study, we [32, 33]. applied two LSE (LSE1 and LSE2) estimation methods; both methods are based on NDVI. -e first method LSE1 refers to the NDVI threshold method proposed by Sobrino [30] 3.4.1. RTE Inversion Methods. -e RTE is a direct method calculated by the following equation: for LST retrieval using a single TIR band. -is can be given by the following equation: ε � 0.004P + 0.986, (7) L � εB T + (1 − ε)L↓ τ + L↑, (11) 􏼂 􏼁 􏼃 λ s where ε is the LSE and P is calculated according to NDVI using the following equation: − 2 − 1 − 1 where L (w · m · sr · μm ) is the brightness value of TIRS received by satellite sensor and λ is the tenth band of NDVI − NDVI 􏼁 soil P � , TIRS. At present, the accuracy of the eleventh band cali- (8) NDVI − NDVI 􏼐 􏼑 veg soil bration of TIRS is still not quantified, and USGS still does Advances in Meteorology 5 not encourage the use of eleventh band for relevant surface Landsat ETM + data and Landsat-8 data, which can be used temperature inversion, so we only consider the tenth band in in the following equation: the RTE and SC. ε refers to the LSE, B(T ) is the blackbody a · (1 − C − D) + [b · (1 − C − D) + C + D] · T − D · T 􏼈 􏼉 radiance energy, and B(T ) at a temperature of T is cal- T � , s s s culated by the inversion of the following equation: (19) B T 􏼁 � 􏼂L − L↑ − τ(1 − ε)L↓􏼃τε, (12) s λ where T is the effective mean atmospheric temperature and a and b values for the Landsat-8 band 10 are −67.355 and and, eventually, T can be obtained from the inversion of 0.458606, respectively. C � ε · τ and D � (1 − τ)[1 + Planck’s law as in the following equation: (1 − ε) · τ]. Table 2 provides empirical equations to estimate T T � . through air temperature (T ), since it is an essential parameter (13) s o ln K /B T􏼁 + 1􏼁 1 s of MW (Zhang et al. 2016). In this study, T was computed for the mid-latitude summer and winter region and T was ob- tained from the mean of all meteorological stations. 3.4.2. SC Inversion Methods. Jimenez-Muñoz ´ et al. [20] proposed a revised version of SC for LST retrieval using 3.4.4. SW Inversion Methods. Although the USGS does not Landsat TIR data. Concerning the SC, T is obtained from encourage users to use band 11 calculations, some re- the following equations: searchers have developed a Split Window algorithm for − 1 � c􏽨ε ψ L + ψ 􏼁 + ψ 􏽩 + δ, (14) Landsat-8 that references other satellite two-channel algo- s 1 2 3 rithms. Jimenez-Muñoz ´ et al. [21] proposed a revised version 2 2 of SW for LST retrieval using Landsat TIRS data. -e al- T T c � , δ ≈ T − , (15) gorithm is simple to calculate. After the brightness tem- 􏼐b L􏼑 perature of band 10 and band 11 of TIR is obtained, it can be calculated by the following equation: where b is equal to 1320 K for Landsat-8 band 10 and L and T are obtained by equation (1) and equation (6), respectively. T � T + c T − T 􏼁 + c T − T 􏼁 + c s 10 1 10 11 2 10 11 0 (20) ψ , ψ , ψ is a function of water vapor content (w) expressed 1 2 3 + c + c w􏼁 1 − ε 􏼁 + c + c w􏼁 Δε, 3 4 m 5 6 in the following equation: where T and T are the brightness temperatures of band ψ � p w + p w + p , 10 11 1 11 12 13 10 and band 11 and ε is the average LSE of band 10 and (16) ψ � p w + p w + p , 2 21 22 23 band 11. Δε is the LSE interpolation of band 10 and band 11. In this study, the LSE of main ground objects was obtained ψ � p w + p w + p , 3 31 32 33 through ASTER spectral library, and the estimated value of three NDVI segments in equation (10) was matched through where p (i � 1, 2, 3, j � 1, 2, 3) is the parameter related to w. ij the calculation of the average value of LSE of relevant ground Perhaps because the calibration of TIRS band 11 is still objects. Table 3 shows the estimated LSE. w is the water unstable, Jimenez-Muñoz ´ et al. only gave specific parameters vapor content (w), and the values of parameters c ∼ c are for TIRS band 10: 0 6 −0.268, 1.378, 0.183, 54.3, −2.238, −129.2, and 16.4, 0.04019 0.02916 1.01523 ⎪ ⎪ ⎧ ⎫ ⎪ ⎪ respectively. ⎨ ⎬ p � −0.38333 −1.50294 0.20324 . (17) Rozenstein et al. [25] proposed an SW algorithm for ⎪ ⎪ ⎪ ⎪ ⎩ ⎭ inversion of land surface temperature using Landsat-8 TIR 0.00918 1.36072 −0.27514 data based on the MW algorithm of QIN [23]. -e algorithm − 2 is expressed by the following equations: When w ≥ 3g · cm , the accuracy of LST will be sig- nificantly affected, so Jimenez-Muñoz ´ suggests using T � A + A T − A T , (21) s 0 1 10 2 11 equation (18) to calculate ψ , ψ , ψ : 1 2 3 1 10 A � , (22) ψ � , E � D C − D C , (23) −L↓ − L↑ 0 11 10 10 11 (18) ψ � , D 1 − C − D 11 10 10 E � , (24) ψ � L↓. D 1 − C − D 10 11 11 E � , (25) 3.4.3. MW Inversion Methods. Mono Window (MW) al- C � ε τ (θ), (26) i i i gorithm was originally developed for Landsat TM data. D � 􏼂1 − τ (θ)􏼃􏼂1 + 1 − ε􏼁 τ (θ)􏼃, (27) Later, some scholars evaluated the ability of MW to process i i i i 6 Advances in Meteorology Table 2: -e linear equations for the calculation of the effective Table 4: a and b over different temperature ranges. i i mean atmospheric temperature (T ) from the near-surface air Temperature ranges a b a b 10 10 11 11 temperature (T ). 0∼30 C −59.1391 0.4213 −63.3921 0.4565 Region Linear equations 0∼40 C −60.9196 0.4276 −65.2240 0.4629 USA 1976 region T � 25.940 + 0.8805 × T ° 10∼40 C −62.8065 0.4338 −67.1728 0.4694 a o Tropical region T � 17.977 + 0.9172 × T ° a o 10∼50 C −64.6081 0.4399 −69.0251 0.4756 Mid-latitude summer region T � 16.011 + 0.9262 × T a o Mid-latitude winter region T � 19.270 + 0.9112 × T a o Table 5: Relationship between atmospheric transmittance and atmospheric water vapor content. Table 3: Estimation of the associated LSE. Profile Estimation equation R SEE Band NDVI < 0.05 0.05 ≤ NDVI < 0.6 0.6 ≤ NDVI τ � −0.1146w + 1.0286 0.9882 0.0094 1976 U.S. Standard TIRS 10 0.9908 0.9451 0.9824 τ � −0.1568w + 1.0083 0.9947 0.0086 TIRS 11 0.9901 0.9503 0.9819 τ � −0.1134w + 1.0355 0.986 0.1010 Mid-latitude summer τ � −0.1546w + 1.0078 0.996 0.0073 where A � E a − E a , A � 1 + A − E b , A � 0 1 10 2 11 1 1 10 2 A + E b , A , A , A are the parameters determined by 2 11 0 1 2 Table 6: Different LST inversion methods. atmospheric transmittance and LSE, and a , b are, respec- i i Model Model + parameter Model ID tively, the regression coefficients of band 10 and band 11 RTE (LSE1, τ, L↑, L↓ ) LST1 determined according to different temperature ranges; this is RTE RTE (LSE2, τ, L↑, L↓) LST2 shown in Table 4. In this study, parameters with temperature ° ° SC (w) LST3 ranging from 0 C to 40 C were selected. ε is the LSE of band SC SC (LSE1, τ, L↑, L↓) LST4 10 and band 11. τ (θ) is the atmospheric transmittance SC (LSE2, τ, L↑, L↓) LST5 corresponding to the solar zenith angle. Rozenstein et al. SW (by Jimenez-Muñoz ´ et al.) (LSE2, w) LST6 provided the relationship between atmospheric transmit- SW SW (by Rozenstein et al.) (LSE2, w) LST7 tance and atmospheric water vapor content, as shown in MW (LSE1, τ, L↑, L↓) LST8 Table 5; namely, atmospheric transmittance was estimated MW MW (LSE2, τ, L↑, L↓) LST9 by atmospheric water vapor content, which was also the calculation method in this paper. In this study, the pa- rameters determined by mid-latitude summer are selected. Inputs of parameters such as LSE, up radiance, down radiance, atmospheric transmittance, atmospheric water and gas content, and atmospheric average temperature are 3.5. Validation Method and Performance Metrics. In this needed in the inversion algorithm. Each of the above pa- study, based on four commonly used LST inversion methods rameters is not directly measured by the instrument but has and different parameters, a total of nine schemes are been estimated and processed. -e errors of these inputs of designed, which are shown in Table 6. parameters will affect the accuracy of LST inversion. Sen- -e T-based technique, examined in this research, is a sitivity analysis is an application of how the errors (nu- direct way of comparing the satellite-derived LST with merical, statistical, or otherwise) of the model output are measured temperatures at meteorological stations [34–38]. divided and allocated to different sources of uncertainty in -e capability of the T-based technique depends mostly on the model input [41]. -e following equation is utilized: the accuracy of the measured temperatures at meteorological stations and pixel scale effect. -e major benefit of the S � T (x) − T (x + Δx), (30) e s s T-based method is that the accuracy of LST algorithm can be where x represents one of the input parameters and Δx is the accurately evaluated [39]. In this study, satellite-derived LST step size of the parameter setting. In the study species, at- and station-based LST were analyzed considering the per- formance metrics such as Root Mean Square Error (RMSE) mospheric transmittance (τ), emissivity, and NDVI were set as 0.01, upwelling radiance and downwelling radiance were and average Bias [40]. -e formulas of these metrics are as set as 0.1 unit, water vapor content (w) was set as 0.1 unit, follows: mean surface temperature was set as 1 unit, and S was the 􏽳�������������� � LST difference calculated for each increase in step size; T − T 􏽐􏼂 􏼃 L8 station (28) RMSE � , T (x + Δx) and T (x) refer to the LST calculated for s s “x + Δx” and “x,” respectively. 􏽐 􏼂T − T 􏼃 L8 Station (29) BIAS � , 4.Results and Analysis where T and T are the Landsat-8-derived LST and In order to accurately evaluate the accuracy of the LST L8 Station station-based LST, respectively, and n refers to the number inversion algorithm, measured data from 20 meteorological of pieces of data; in this study, n is equal to 20. stations in the study area were used to evaluate the stability Advances in Meteorology 7 of the inversion results of different LST inversion algorithms temperature, the error of the inversion results of the algo- in different seasons. In addition, considering the difference rithm was smaller than that of the measured data. With the of inversion results of the same reflectivity ground objects by increase of temperature, the inversion results of all the in- different LST inversion algorithms, we designed two groups version algorithms were higher than those of the measured of experimental areas inside the city and outside the city. data. LST1, LST2, LST4, and LST5 have a relatively small Each experimental area has four types of ground objects, error, followed by LST3 LST6, LST7; the inversion errors of respectively, and each type of ground object has 20 POIs. LST8 and LST9 were relatively large. Finally, we analyzed the dependence of different LST in- We performed linear fitting between nine LST inversion version algorithms on parameters by equal step size. results and measured temperatures in four seasons. -e fitting results are shown in Figure 3. According to the statistical fitting results, it can also be seen that, in winter, all 4.1. Comparison of the Results of Different LST Inversion the relevant results were relatively ideal, with R greater than Algorithms. Nine LST inversion algorithms were used to 0.73. Except for the two SW algorithms LST6 and LST7, the calculate Landsat-8 data in four seasons in the research area. fitting slopes of the other algorithms were all less than 1. -e -e results were compared with the measured temperatures fitting results in autumn show that R was around 0.5, and all of 20 meteorological stations. -e results are shown in the fitting slopes were less than 1. However, the relevant Figure 2. By comparing the inversion results of each algo- intercept data are relatively large. In spring and summer, the rithm with the measured data, the statistical Root Mean proper slopes were more significant than 1. -e absolute Square Error (RMSE) and average Bias are shown in Table 7 value of the fitting intercept was also relatively large. It is of Appendix. It can be obtained from Figure 2 and important to note that the fitting results of the LST7 al- Appendix’s Table 7 that the inversion results of the LST gorithm in spring, summer, and autumn show no significant algorithm are entirely consistent with the graph trend of the phenomenon; that is to say, the error between the inversion measured data of meteorological stations. -e high elevation results of the LST7 algorithm and the measured results is meteorological sites (e.g., Foyeding, Yanqing, Tanghekou, unacceptable. and Zhaitang) were low. In the graphic display, the line with Overall, the inversion results of the RTE and SC algo- the measured data of adjacent stations shows a concave rithm calculated by LSE1 and LSE2 have an excellent fitting shape. All LST inversion algorithms also have the same effect with the measured data; the inversion results of the SC characteristics. In terms of seasonal scale, the inversion algorithm are calculated by w. -e inversion result of MW results of the nine algorithms in winter have the highest was higher compared to the measured temperature. -e matching degree with the measured data of meteorological inversion effect of SW was not particularly ideal, which may stations. -e values of RMSE and Bias were relatively small. be caused by the instability of the radiation calibration of the Followed by the data in autumn, the matching degrees in 11th band of TIRS. spring and summer were inferior. -e values of RMSE and Bias were relatively large. In winter, the inversion results of LST1, LST2, LST3, LST4, and LST5 were the closest to the 4.2. Results on Different POIs. It is well known that intensive measured data of meteorological stations, with difference of human activities affect the surrounding environment. In- less than 2 C, followed by the inversion results of LST6, dustrial production and emissions affect atmospheric which were 3 C higher than the measured data, and the transmittance, water vapor content, and upward and downward radiation. All LST inversion algorithms need difference between the inversion results of LST7, LST8, and LST9 and the measured data was about 7 C. In spring and these parameters to support the calculation. -erefore, it is summer, almost all the algorithm inversion results were necessary to explore whether the inversion results of similar higher than the measured temperature. At this time, the surface features with the same reflectivity in densely pop- closest to the measured temperature were LST1, LST2, LST4, ulated cities and outside cities close to the natural surface and LST5, with difference of about 5 C, followed by LST3 will be affected by various inversion algorithms and to what inversion results, about 10 C. -e inversion results of LST8 extent. -e four selected ground objects (20 POIs for each and LST9 were 20 C higher than the measured results, and ground object) were analyzed inside and outside the city. the reliability of the results will be seriously questioned. In Considering the statistical error, 10%–90% of the points were selected as the upper and lower whiskers of the box autumn, the inversion results of LST1, LST2, LST4, and LST5 were slightly lower than the measured data by 2 C, the map, and the upper and lower boundaries of the box rep- resented 25%–75% of the values, respectively. -e result is inversion results of LST3 were 5 C higher than the measured data, and the inversion results of LST6, LST7, LST8, and shown in Figure 4. LST9 were 8 C higher than the measured data. As shown in Figure 4, in the same season, the same LST From the perspective of the algorithm inversion results, inversion algorithm results show that the height of the box the inversion results of LST1, LST2, LST4, and LST5 were the map inside the city was less than the height of the box map closest to the measured data of meteorological stations, with outside the city. -e standard deviation of inversion results the inversion error within ±5 C, followed by the inversion of all LST inversion algorithms on POI inside the city was results of LST3 at about 10 C. -e worst are the inversion less than the standard deviation of inversion results outside results of LST6, LST7, LST8, and LST9, with the inversion the city. On the one hand, the main reason may be the urban heat island effect and on the other hand it may be the error around 20 C. In general, in the season with low 8 Advances in Meteorology 5 50 –5 –10 –15 MEASURED LST4 LST7 MEASURED LST4 LST7 LST1 LST5 LST8 LST1 LST5 LST8 LST2 LST6 LST9 LST2 LST6 LST9 LST3 LST3 (a) (b) MEASURED LST4 LST7 MEASURED LST4 LST7 LST1 LST5 LST8 LST1 LST5 LST8 LST2 LST6 LST9 LST2 LST6 LST9 LST3 LST3 (c) (d) Figure 2: Inversion results of 9 LST inversion algorithms at 20 meteorological stations in different seasons. (a) 2017-01-31 (winter), (b) 2017- 05-17 (spring), (c) 2017-7-10 (summer), and (d) 2017-10-30 (autumn). influence of intensive human activities inside the city on the inversion results will be different in different outer areas. As parameters of the inversion algorithm. In addition, in a a result, the inversion results of land surface temperature are relatively narrow space, using the same estimated parame- relatively scattered. ters for calculation, the inversion results are relatively Both inside and outside the city, the height of the water concentrated. However, the outer space of the city is rela- box chart is minimal. -e POI inversion temperature dif- tively broad, which belongs to the natural surface. -ere are ference between inside the city and outside the city is about differences in estimated atmospheric parameters, and the 1 C in autumn. -e average inversion temperature of POI Temperature (°C) Temperature (°C) Shunyi Shunyi Haidian Haidian Yanqing Yanqing Foyeding Foyeding Tanghekou Tanghekou Miyun Miyun Huairou Huairou Shangdianzi Shangdianzi Pinggu Pinggu Tongzhou Tongzhou Chaoyang Chaoyang Changping Changping Zhaitang Zhaitang Mentougou Mentougou Beijing Beijing Shijingshan Shijingshan Fengtai Fengtai Daxing Daxing Fangshan Fangshan Xiayunling Xiayunling Temperature (°C) Temperature (°C) Shunyi Shunyi Haidian Haidian Yanqing Yanqing Foyeding Foyeding Tanghekou Tanghekou Miyun Miyun Huairou Huairou Shangdianzi Shangdianzi Pinggu Pinggu Tongzhou Tongzhou Chaoyang Chaoyang Changping Changping Zhaitang Zhaitang Mentougou Mentougou Beijing Beijing Shijingshan Shijingshan Fengtai Fengtai Daxing Daxing Fangshan Fangshan Xiayunling Xiayunling Advances in Meteorology 9 Table 7: Root Mean Square Error (RMSE), average bias, slope, intercept, and R of LST inversion and field measured temperature. Season Model ID RMSE Bias Slope Intercept R LST1 0.67 −0.15 0.94 0.04 0.775 LST2 3.24 −0.73 0.94 −0.54 0.772 LST3 18.89 4.22 0.91 4.52 0.775 LST4 0.27 −0.06 0.98 0.01 0.794 Winter LST5 3.23 −0.72 0.94 −0.53 0.772 LST6 13.54 3.03 1.16 2.51 0.738 LST7 11.80 2.64 1.89 −0.24 0.771 LST8 22.42 5.01 0.93 5.24 0.776 LST9 19.20 4.29 0.93 4.52 0.772 LST1 10.18 2.28 1.42 −12.30 0.535 LST2 9.54 2.13 1.38 −11.03 0.531 LST3 39.09 8.74 1.33 −2.57 0.531 LST4 12.33 2.76 1.49 −14.05 0.472 Spring LST5 9.79 2.19 1.38 −11.02 0.530 LST6 82.87 18.53 2.11 −19.69 0.469 LST7 63.14 14.12 1.04 12.86 0.055 LST8 91.98 20.57 1.44 5.31 0.532 LST9 90.20 20.17 1.52 2.28 0.561 LST1 11.12 2.49 1.32 −8.99 0.459 LST2 11.49 2.57 1.28 −7.39 0.498 LST3 65.02 14.54 1.16 8.88 0.450 LST4 11.47 2.57 1.33 −9.16 0.458 Summer LST5 11.86 2.65 1.28 −7.52 0.499 LST6 84.98 19.00 1.49 1.65 0.411 LST7 72.96 16.32 1.36 3.32 0.121 LST8 53.96 12.07 1.35 0.32 0.465 LST9 55.89 12.50 1.25 0.28 0.500 LST1 1.68 −0.38 0.69 4.35 0.592 LST2 1.63 −0.37 0.72 3.89 0.522 LST3 12.74 2.85 0.66 7.94 0.591 LST4 0.45 −0.10 0.70 4.36 0.607 Autumn LST5 1.53 −0.34 0.72 3.90 0.519 LST6 27.62 6.17 0.93 7.30 0.459 LST7 21.57 4.82 0.52 12.06 0.055 LST8 32.28 7.22 0.70 11.81 0.599 LST9 31.58 7.06 0.74 10.97 0.573 inside the city is about 3 C higher than that outside the city vegetation pixels, their reflectivity is inconsistent, which may in other seasons. It shows that the standard deviation of all lead to specific errors. LST inversion algorithms on POI of the water body is -e box graph of soil POI temperature retrieved by the minimal, which also conforms to the characteristics of the SW (LST6 and LST7) algorithm is relatively large; in par- water body being relatively stable, and LSE is close to the ticular, the LST7 algorithm is very obvious. -e height of the black body. box graph of soil POI temperature retrieved by other al- Surprisingly, the height of the box chart of vegetation in gorithms is relatively tiny, with a difference of about 4 C all categories is considerable. Among all the POI inversion outside the city and 3 C inside the city. It shows that the temperatures of vegetation outside the city in spring, the POI standard deviation of all LST inversion algorithms on soil inversion temperature difference of vegetation outside the POI is minimal. It is worth noting that, in the selection of soil city is about 5 POI in this paper, the soil is dry and bright areas were C (this value does not count LST7 because it is precarious). -e POI inversion temperature difference of selected instead of the dark soil with high moisture content other vegetations is 8 C. It indicates that the standard de- to avoid the error caused by the significant difference in viation of all LST inversion algorithms on vegetation POI is reflectivity. substantial, indicating that inversion algorithms are more Except for LST6 and LST7, the height of the box graph of sensitive to vegetation inversion, which will be explained in building POI inversion by other LST inversion algorithms is ° ° detail in the next section. In addition, it is also possible that small, with a difference of about 5 C outside the city and 3 C when selecting pure pixels of vegetation, the vegetation types inside the city, indicating that the standard deviation of of pixels are different. Some pixels are grassland, some are inversion results of all LST inversion algorithms on building woodland, and some are shrubs. Although these are all POI is also small. 10 Advances in Meteorology 12 12 12 12 y=0.98x+0.01 y=0.94x–0.54 10 y=0.94x+0.04 10 10 10 2 R =0.794 R =0.772 8 8 8 8 R =0.775 6 6 6 6 y=0.91x+4.52 2017-01-31 4 4 4 4 2 2 2 2 R =0.775 0 0 0 0 –2 –2 –2 –2 –4 –4 –4 –4 –4 –2 0 2 4 6 8 10 12 –4 –2 0246 8 10 12 –4 –2 0246 810 12 –4 –2 0 2 4 6 810 12 MEASURED MEASURED MEASURED MEASURED 12 12 12 12 12 y=0.94x–0.53 10 10 10 10 10 R =0.772 8 8 8 8 8 6 6 6 6 6 4 4 4 y=1.89x–0.24 4 y=0.93x+5.24 4 y=0.93x+4.52 y=1.169x+2.51 2 2 2 2 2 2 2 2 R =0.771 R =0.776 R =0.772 R =0.738 0 0 0 0 0 –2 –2 –2 –2 –2 –4 –4 –4 –4 –4 –4 –2 0 2 4 6 8 10 12 –4 –2 0246 8 10 12 –4 –2 0246 8 10 12 –4 –2 0246 810 12 –4 –2 0 2 4 6 810 12 MEASURED MEASURED MEASURED MEASURED MEASURED 60 60 60 60 y=1.42x–12.30 y=1.33x–2.57 y=1.49x–14.05 55 55 y=1.38x–11.03 55 55 2 2 2 2 50 R =0.535 50 R =0.531 50 R =0.531 50 R =0.472 45 45 45 45 2017-05-07 40 40 40 40 35 35 35 35 30 30 30 30 25 25 25 25 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 MEASURED MEASURED MEASURED MEASURED 60 60 60 60 60 y=2.11x–19.69 y=1.38x–11.02 55 55 55 y=1.04x+12.86 55 y=1.44x+5.31 55 y=1.52x+2.28 2 2 2 R =0.469 R =0.530 50 50 50 R =0.055 50 R =0.532 50 R =0.561 45 45 45 45 45 40 40 40 40 40 35 35 35 35 35 30 30 30 30 30 25 25 25 25 25 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 25 30 35 40 45 50 55 60 MEASURED MEASURED MEASURED MEASURED MEASURED 66 66 y=1.28x–7.39 66 66 y=1.16x+8.88 y=1.32x–8.99 2 y=1.33x–9.16 60 2 60 R =0.498 60 60 R =0.450 2 R =0.459 R =0.458 54 54 54 54 2017-07-10 48 48 48 48 42 42 42 42 36 36 36 36 30 30 30 30 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 MEASURED MEASURED MEASURED MEASURED 66 66 66 66 66 y=1.28x–7.52 y=1.49x+1.65 y=1.36x+3.32 y=1.35x+0.32 y=1.25x+0.28 60 60 60 2 60 60 R =0.499 2 R =0.411 2 R =0.121 R =0.500 R =0.465 54 54 54 54 54 48 48 48 48 48 42 42 42 42 42 36 36 36 36 36 30 30 30 30 30 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 30 36 42 48 54 60 66 MEASURED MEASURED MEASURED MEASURED MEASURED 28 28 28 28 y=0.69x+4.35 y=0.72x+3.89 y=0.66x+7.94 y=0.70x+4.36 2 2 2 2 24 R =0.592 24 R =0.522 24 R =0.591 24 R =0.607 20 20 20 20 2017-10-30 16 16 16 16 12 12 12 12 8 8 8 8 812 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 MEASURED MEASURED MEASURED MEASURED 28 28 28 28 28 y=0.72x+3.90 24 R =0.519 24 24 24 24 20 20 20 20 20 y=0.93x+7.30 y=0.52x+12.56 y=0.70x–11.81 y=0.74x+10.97 2 2 16 16 R =0.459 16 16 16 R =0.599 R =0.055 R =0.57 12 12 12 12 12 8 8 8 8 8 812 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 8 12 16 20 24 28 MEASURED MEASURED MEASURED MEASURED MEASURED Figure 3: LST inversion results are fitted with the measured temperature. LST5 LST5 LST5 LST5 LST6 LST1 LST6 LST1 LST6 LST1 LST6 LST1 LST7 LST2 LST7 LST2 LST7 LST2 LST7 LST2 LST8 LST3 LST8 LST3 LST8 LST3 LST8 LST3 LST9 LST4 LST9 LST4 LST9 LST4 LST9 LST4 Advances in Meteorology 11 2017-01-31 2017-05-07 2017-07-10 2017-10-30 –5 40 –10 –15 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 10 60 70 30 60 25 4 50 20 2 30 40 15 20 30 10 –2 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 50 10 35 0 20 –5 –10 20 5 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 70 80 35 70 30 10 60 60 25 50 20 40 15 30 10 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 60 40 60 35 50 40 6 15 2 20 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 25 70 30 15 60 10 20 5 40 –5 30 10 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 70 50 15 40 10 30 5 20 0 10 –5 30 30 0 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 20 70 80 40 50 25 30 30 10 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 LST1 LST2 LST3 LST4 LST5 LST6 LST7 LST8 LST9 Figure 4: Inversion results of POI inside and outside the city by various algorithms. Inside-build Outside-build Inside-soil Outside-soil Inside-veg Outside-veg Inside-water Outside-water Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) Temperature (°C) 12 Advances in Meteorology upline radiation on the RTE (LST1 and LST2) algorithm is 4.3. Sensitivity Analysis of Model Parameters. -is paper adopts the control variable method to conduct sensitivity logarithmic. However, the curvature of the curve is min- imal. With the increase of upline radiation, the surface analysis for different parameters to explore the sensitivity of various LST inversion algorithms to parameters. First, temperature retrieved by the RTE algorithm also decreases. assume all the DN values on the image are the same (in As shown in Figure 5(e), the influence of downward ra- this study, take the average of the band after atmospheric diation on the SC and RTE algorithm is linear. As correction; band 10 is set to 10.4 and band 11 is set to 9.4). downward radiation increases by 1 unit, the inversion land -en select a group of parameters as variables, and the surface temperature decreases by 0.2 units. As shown in rest as invariants to control the variation of variable Figure 5(f), the influence of atmospheric water vapor content on LST3 and LST6 is linear. As the atmospheric parameters for sensitivity analysis. Finally, the sensitivity of the parameters was analyzed. As the data with good water vapor content increases by 0.1 units, the surface temperature retrieved by LST3 and LST6 increases by about weather conditions were selected in the remote sensing image screening previously, the interval of atmospheric 0.3 units. -e influence of atmospheric water vapor content on LST7 is nonlinear. With the increase of atmospheric transmittance is set as 0.5∼1.0 with a step size of 0.01 unit, and the interval of surface reflectance is set as 0.9∼1.0 water vapor, the LST retrieved by LST7 also increases, but with a step size of 0.01 unit. -e interval of DNVI is set as the saturated phenomenon appears after 2.5. -is result is −0.2∼1.0 with a step size of 0.01 units. -e interval of consistent with the result of Jimenez et al. (2010). In upward and downward radiation is set as 0∼5, and the Figure 5(g), the average land surface temperature param- step size is set as 0.1 units; the interval of atmospheric eters required in the MW algorithm are presented. It can be water vapor content is set as 0–2.5, and the step size is set seen that, with the increase of the average land surface temperature by 1 unit, the inversion of the land surface as 0.1 units; the interval of average surface temperature is set as 0–100, and the step size is set as 1 unit. -e influence temperature by MW decreases by 0.01 units. of each parameter on the results in different algorithms was obtained by controlling variables. For the compa- 5.Discussion rability of the results, LST was normalized, as shown in Figure 5. NASA’s ACPC obtains the atmospheric parameters required It can be seen from Figure 5(a) that the influence of for the LST inversion method based on the MODTRAN atmospheric transmittance on the inversion algorithms of radiation transfer code. It is not possible to find atmospheric RTE, SC, and MW is a logarithmic function. With the in- profiles in situ (radiosonde data, etc.) at any time and any crease of atmospheric transmittance, the inversion of land place. -erefore, even if this use (simulating atmospheric surface temperature gradually decreases. -e atmospheric profiling information with ACPC) affects the accuracy of the transmittance has a minor effect on the curvature of RTE method, it is clear from our results and the literature (LST1 and LST2) and a more significant effect on the cur- [8, 10, 14, 18, 20, 22–24, 29–33] that NASA’s ACPC provides vature of SC (LST4 and LST5) and MW (LST8 and LST9). an excellent and effective simulation. -e development of Figure 5(b) shows that the influence of specific emissivity on the LST inversion method has its source of error, as it in- all algorithms is linear. As the specific emissivity increases, volves parameterization steps for inverting the coefficients the land surface temperature decreases linearly. When the and estimating some initial parameters. -erefore, it is emissivity increases by 0.1 units, the inversion surface necessary to compare the errors of various algorithms in the temperature decreases by 1 unit. Figure 5(c) shows that the system. In general, the inversion results of the RTE algo- influence of NDVI on LST1, LST3, LST4, and LST8 is rithm and SC algorithm calculated by LSE1 and LSE2 have piecewise linear. Segmentation is because, during parameter an excellent fitting effect with the measured data. -e in- setting, specific emissivity is segmenting according to NDVI version results of MW are higher compared to the measured mandatory threshold. Specific emissivity is set as a fixed temperature. -e inversion effect of SW is not particularly value for parts with NDVI less than 0.05 and DNVI greater ideal, which may be due to the instability of radiation cal- than 0.6. -e inversion temperature is also a fixed value. For ibration at band 11 of TIRS. Hanqiu [42] used SC (LST3 and the part with DNVI greater than 0.05 and less than 0.6, the LST4) and SW (LST8 and LST9) algorithms to invert the LST specific emissivity changed with the change of NDVI. -e in Fuzhou on August 4, 2013, and concluded that the SC inversion temperature decreased linearly with the increase of result was better than the SW result. -e inversion error of NDVI. -e influence of NDVI on LST2, LST5, and LST9 is LST4 was −2.77 C. However, since the atmospheric water also segmenting. However, when NDVI is more than 0.05, vapor content parameter was more than 4 on that day, the the influence of NDVI on LST presents a quadratic function; error of LST3 is −17.44 C, and the inversion errors of LST8 ° ° when NDVI is equal to 0.4, the function has a minimum and LST9 are −13.61 C and −5.75 C, respectively. Wang [41] value; when NDVI is more than 0.05 and less than 0.4, the compared the sensitivity analysis of the three algorithms, SC, inversion LST decreases the NDVI. When the NDVI is more MW, and SW, and found that SW had the lowest sensitivity than 0.4, the inversion LST increases the NDVI. to the input parameter error. Asia Siddiqui et al. [19] selected In Figure 5(d), the influence of uplink radiation on the the data of two landscapes in January and April to study the SC (LST4 and LST5) algorithm is linear. With the increase heat island in Bangalore, India, based on the RTE method of uplink radiation, the land surface temperature inversion and found that the relationship between LST and NDVI was by the SC algorithm decreases linearly. -e influence of firstly positively correlated and then negatively correlated. Advances in Meteorology 13 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0.5 0.6 0.7 0.8 0.9 1.0 0.90 0.92 0.94 0.96 0.98 1.00 –0.2 0.0 0.2 0.4 0.6 0.8 1.0 Transmissivity Emissivity NDV1 LST1 LST5 LST1 LST4 LST8 LST1 LST5 LST2 LST8 LST2 LST5 LST9 LST2 LST8 LST4 LST9 LST3 LST7 LST3 LST9 LST4 (a) (b) (c) 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0123 45 0123 45 0.0 0.5 1.0 1.5 2.0 2.5 3.0 up_radiance down_radiance water_vapour LST1 LST4 LST1 LST4 LST3 LST2 LST5 LST2 LST5 LST6 LST7 (d) (e) (f) 1.0 0.8 0.6 0.4 0.2 0.0 020 40 60 80 100 Ta LST8 LST9 (g) Figure 5: Sensitivity analysis of parameters of various LST inversion algorithms. -e vertical axis represents the normalized LST, and the horizontal axis represents the parameters of each variable. -e units of parameters are given in Section 3. -e various LST inversion algorithms are most stable in temperature from rural areas and found that RTE and SC winter, and the stability becomes worse in other seasons. presented similar results regardless of the season, while MW Aliihsan Sekertekin and Stefania Bonafoni [40] used differed from RTE and SC for all seasons, especially in Landsat-5, Landsat-7, and Landsat-8 data to retrieve surface summer. -e parameter error had little influence on the Normalized temperature Normalized temperature Normalized temperature Normalized temperature Normalized temperature Normalized temperature Normalized temperature 14 Advances in Meteorology parameters calculation, calculation results by ACPC dif- results in the humid environment. MW and SC methods are more sensitive to the error of input parameters, especially ference are not very big; however, when we use the estimated atmospheric profile parameters for further calculation, there under hot and humid conditions. -is result is consistent with our inversion results in urban areas. On the one hand, will be error propagation. Sensitivity analysis of the LST with the increase of temperature, the properties of ground inversion algorithm can help researchers to understand the objects (specific heat capacity and specific emissivity) have source of error and reduce the error as much as possible. distinct differences, leading to a significant difference in reflectivity and then resulting in a significant difference in 6.Conclusions the inversion of surface temperature. On the other hand, considering the errors of surface temperature inversion of In all seasons, the inversion results of the nine LST algo- similar surface features POI from inside and outside the city, rithms are in complete agreement with the graph trends of the increase of temperature will have a particular influence the measured data of all meteorological stations. In winter, on the parameters of the LST inversion algorithm, which the inversion results of the nine algorithms have the highest may lead to errors in the inversion results. matching degree with the measured data of meteorological -e stability of various inversion algorithms for tem- stations, and the RMSE and Bias values are relatively small. perature inversion inside and outside the city is tested. -e -e second is autumn’s LST1, LST2, LST4, and LST5 in- results show that the standard deviations of inversion results version results of the nearest meteorological station, with for POIs of LST inversion algorithms in urban areas are inversion error of ±5 C; the following inversion result of smaller than those of inversion results outside the city. On LST3 is about 10 C; the worst are the inversion results of the one hand, the main reason may be that the intensive LST6, LST7, LST8 and LST9, with inversion error of about human activities in cities have a particular influence on land 20 C. In the low-temperature season, the error of the al- surface temperature inversion parameters. On the other gorithm inversion results is less than that of the measured hand, the inversion results of the same reflectivity charac- data. With the increase of temperature, the inversion results teristics will be relatively concentrated when the same es- of each inversion algorithm are all higher compared to the timated parameters are used for calculation in a very narrow measured data. -e inversion errors of LST1, LST2, LST4, space inside the city. However, the space outside the city is and LST5 are small, followed by LST3, LST6, LST7, LST8, relatively broad and belongs to the natural surface, so the and LST9. estimated atmospheric parameters in different regions will According to the relevant results of the nine LST in- be different, which may lead to the relatively scattered in- version results and the measured temperature, the fitting version results of land surface temperature. results in winter are all relatively ideal, with R greater than It is essential to analyze the sensitivity/uncertainty of the 0.73. In other seasons, the fitting effect is general, and R is LST inversion algorithm by considering all the input pa- about 0.5. In general, the inversion results of the RTE al- rameters. Sekertekin and Bonafoni’s [43] simultaneous gorithm and SC algorithm calculated by LSE1 and LSE2 have ground-based LST measurements were collected from At- an excellent fitting effect with the measured data, followed by mospheric Radiation Measurement (ARM) and Surface the inversion results of the SC algorithm calculated by w. By Radiation Budget Observing Network (SURFRAD) stations, analyzing the stability of various inversion algorithms inside located at different rural environments of the United States and outside the city, it can be seen that the standard de- to examine the efficiency of different LST algorithms, using viation of inversion results of each LST inversion algorithm both daytime and nighttime Landsat-8 data and in situ for POI inside the city is less than that outside the city. measurements. Concerning the in situ sensitivity results, the -e influence of atmospheric transmittance on RTE, SC, effects on LST of the uncertainty of the downwelling and and MW inversion algorithms is a logarithmic function. upwelling radiance were almost identical in daytime and With the increase of atmospheric transmittance, the in- nighttime. -e accuracy of the LST retrieval methods for version surface temperature decreases gradually. -e in- daytime Landsat-8 data varied between 2.17 K Root Mean fluence of emissivity on all algorithms is linear. -e Square Error (RMSE) and 5.47 K RMSE considering all LST inversion of LST decreases by 1 for every 0.1 increase of methods and LSE models. MW with two different LSE emissivity. -e influence of NDVI on LST1, LST3, LST4, and models presented the best results for the daytime. Wang LST8 is piecewise linear, and the influence on LST2, LST5, et al. [44] presented an improved Mono Window (IMW) and LST9 is in the form of a quadratic function. When NDVI algorithm for LSTretrieval from the Landsat-8 TIRS band 10 is equal to 0.4, the function has a minimum value. When data, and sensitivity analysis conducted for the IMW al- NDVI is more than 0.05 and less than 0.4, the inversion land gorithm revealed that the possible error in estimating the surface temperature decreases. When NDVI is more than required atmospheric water vapor content has the most 0.4, the inversion LST increases; this is the opposite of the significant impact on the probable LSTestimation error. -is previous results. -e effect of uplink radiation on the SC result is similar to our inversion results of LST3, LST6, and (LST4 and LST5) algorithm is linear. With the increase of LST7. Although from our results and literature we can see uplink radiation, the LST retrieved by the SC algorithm clearly that NASA ACPC provides a satisfying and effective decreases linearly. -e influence of upline radiation on the simulation, in our precise and fuzzy provided coordinates, RTE (LST1 and LST2) algorithm is logarithmic. With the respectively, phenology, area type (urban and rural), and increase of upline radiation, the land surface temperature satellite altitude parameters such as atmospheric profile retrieved by the RTE algorithm decreases. -e influence of Advances in Meteorology 15 [8] D. Zhou, J. Xiao, and S. Bonafoni, “Satellite remote sensing of downslope radiation on SC and RTE algorithms is linear. surface urban heat islands: progress, challenges, and per- With the increase of downslope radiation by 1 unit, the spectives,” Remote Sensing, vol. 11, no. 1, 2018. inversion LST decreases by 0.2 units. -e surface temper- [9] C. Keeratikasikorn and S. Bonafoni, “Urban heat island ature retrieved by LST3 and LST6 increases by about 0.3 analysis over the land use zoning plan of bangkok by means of units for every 0.1 unit increase of atmospheric water vapor landsat 8 imagery,” Remote Sensing, vol. 10, no. 3, 2018. content. -e influence of atmospheric water vapor content [10] J. A. Sobrino, R. Oltra-Carrio, ´ and G. Soria, ` “Evaluation of the on LST7 is nonlinear. With the increase of atmospheric surface urban heat island effect in the city of Madrid by water vapor, the surface temperature retrieved by LST7 also thermal remote sensing,” International Journal of Remote increases, but, after 2.5 units, it shows saturation Sensing, vol. 34, no. 9-10, pp. 3177–3192, 2012. phenomenon. [11] J. Tan, Y. Zheng, X. Tang et al., “-e urban heat island and its impact on heat waves and human health in Shanghai,” In- ternational Journal of Biometeorology, vol. 54, no. 1, pp. 75–84, Data Availability [12] C. Maffei, S. Alfieri, and M. Menenti, “Relating spatiotem- Some or all data, models, or codes that support the findings poral patterns of forest fires burned area and duration to of this study are available from the corresponding author diurnal land surface temperature anomalies,” Remote Sensing, upon reasonable request. vol. 10, no. 11, 2018. [13] Z. Wan, P. Wang, and X. Li, “Using MODIS land surface temperature and normalized difference vegetation index Conflicts of Interest products for monitoring drought in the southern great plains, -e authors declare that there are no conflicts of interest USA,” International Journal of Remote Sensing, vol. 25, no. 1, pp. 61–72, 2010. regarding the publication of this paper. [14] J. Sun, G. D. Salvucci, and D. Entekhabi, “Estimates of evapotranspiration from MODIS and AMSR-E land surface Acknowledgments temperature and moisture over the Southern Great Plains,” Remote Sensing of Environment, vol. 127, pp. 44–59, 2012. -e National Key Research and Development Project of [15] J. Qin, S. Liang, R. Liu, H. Zhang, and B. Hu, “A weak- China (no. 2017YFA0605004) and Basic R&D Special Fund constraint-based data assimilation scheme for estimating of Central Government for Non-Profit Research Institutes surface turbulent fluxes,” IEEE Geoscience and Remote Sensing (Grant/Award HKY-JBYW-2021-02) are acknowledged. Letters, vol. 4, no. 4, pp. 649–653, 2007. [16] B. Yong, L. Ren, Y. Hong et al., “First evaluation of the cli- matological calibration algorithm in the real-time TMPA References precipitation estimates over two basins at high and low lat- itudes,” Water Resources Research, vol. 49, no. 5, [1] M. Anderson, J. Norman, W. Kustas, R. Houborg, P. Starks, pp. 2461–2472, 2013. and N. Agam, “A thermal-based remote sensing technique for [17] O. Hall, G. Falorni, and R. L. Bras, “Characterization and routine mapping of land-surface carbon, water and energy quantification of data voids in the shuttle radar topography fluxes from field to regional scales,” Remote Sensing of En- mission data,” IEEE Geoscience and Remote Sensing Letters, vironment, vol. 112, no. 12, pp. 4227–4241, 2008. vol. 2, no. 2, pp. 177–181, 2005. [2] P. Dash, F.-M. Gottsche, ¨ F.-S. Olesen, and H. Fischer, “Land [18] J. C. Price, “Estimating surface temperatures from satellite surface temperature and emissivity estimation from passive thermal infrared data-A simple formulation for the atmo- sensor data: theory and practice-current trends,” Interna- spheric effect,” Remote Sensing of Environment, vol. 13, no. 4, tional Journal of Remote Sensing, vol. 23, no. 13, pp. 2563– pp. 353–361, 1983. 2594, 2010. [19] A. Siddiqui, G. Kushwaha, A. Raoof, P. A. Verma, and [3] R. E. Dickinson, “Land surface processes and climate-surface Y. Kant, “Bangalore: urban heating or urban cooling?” 7e albedos and energy balance,” Advances in Geophysics, in Egyptian Journal of Remote Sensing and Space Science, vol. 24, Proceedings of a Symposium Commemorating the Two-Hun- no. 2, pp. 265–272, 2021. dredth Anniversary of the Academy of Sciences of Lisbon, [20] J. C. Jimenez-Muñoz and J. A. Sobrino, “A single-channel pp. 305–353, Lisbon, Portugal, October 1983. algorithm for land-surface temperature retrieval from ASTER [4] D. Tong, Q. Zhang, Y. Zheng et al., “Committed emissions data,” IEEE Geoscience and Remote Sensing Letters, vol. 7, from existing energy infrastructure jeopardize 1.5 C climate no. 1, pp. 176–179, 2010. target,” Nature, vol. 572, no. 7769, pp. 373–377, 2019. [21] J. C. Jimenez-Muñoz, ´ J. Cristobal, J. A. Sobrino et al., “Re- [5] G. P. Peters, R. M. Andrew, J. G. Canadell et al., “Carbon vision of the single-channel algorithm for land surface dioxide emissions continue to grow amidst slowly emerging temperature retrieval from landsat thermal-infrared data,” climate policies,” Nature Climate Change, vol. 10, no. 1, pp. 3–6, 2019. IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 1, pp. 339–349, 2009. [6] B. Talukder, R. Matthew, G. W. Vanloon, M. J. Bunch, K. W. Hipel, and J. Orbinski, “Melting of Himalayan glaciers [22] Z. Qin, A. Karnieli, and P. Berliner, “A mono-window al- gorithm for retrieving land surface temperature from Landsat and planetary health,” Current Opinion in Environmental Sustainability, vol. 50, pp. 98–108, 2021. TM data and its application to the Israel-Egypt border re- gion,” International Journal of Remote Sensing, vol. 22, no. 18, [7] J. Xin, X. Sun, and L. Liu, “Quantifying the contribution of climate and underlying surface changes to alpine runoff al- pp. 3719–3746, 2010. [23] K. Mao, Z. Qin, J. Shi, and P. Gong, “A practical split-window terations associated with glacier melting,” Hydrological Pro- cesses, vol. 35, no. 3, 2021. algorithm for retrieving land-surface temperature from 16 Advances in Meteorology MODIS data,” International Journal of Remote Sensing, [40] A. Sekertekin and S. Bonafoni, “Land surface temperature retrieval from landsat 5, 7, and 8 over rural areas: assessment vol. 26, no. 15, pp. 3181–3204, 2007. [24] Z. M. W. a. J. Dozier, “A generalized split- window algorithm of different retrieval algorithms and emissivity models and toolbox implementation,” Remote Sensing, vol. 12, no. 2, 2020. for retrieving land-surface temperature from space,” IEEE Transactions on Geoscience and Remote Sensing, vol. 34, no. 4, [41] L. Wang, Y. Lu, and Y. Yao, “Comparison of three algorithms for the retrieval of land surface temperature from landsat 8 pp. 892–905, 1996. [25] O. Rozenstein, Z. Qin, Y. Derimian, and A. Karnieli, “Der- images,” Sensors, vol. 19, no. 22, 2019. [42] X. Hanqiu, “Retrieval of reflectivity and surface temperature ivation of land surface temperature for Landsat-8 TIRS using a from the newly launched Landsat 8 satellite,” Chinese journal split window algorithm,” Sensors, vol. 14, no. 4, of geophysice, vol. 58, no. 3, pp. 741–747, 2015. pp. 5768–5780, 2014. [43] A. Sekertekin and S. Bonafoni, “Sensitivity analysis and [26] F.-M. G. Prasanjit Dash, F.-S. Olesen, and H. Fischer, “Re- validation of daytime and nighttime land surface temperature trieval of land surface temperature and emissivity from sat- retrievals from Landsat 8 using different algorithms and ellite data: physics, theoretical limitations and current emissivity models,” Remote Sensing, vol. 17, no. 12, p. 2776, methods,” Journal of the Indian Society of Remote Sensing, vol. 29, no. 1, pp. 24–30, 2001. [44] F. Wang, Z. Qin, C. Song, L. Tu, A. Karnieli, and S. Zhao, “An [27] Y. Yu, D. Tarpley, J. L. Privette et al., “Validation of GOES-R improved mono-window algorithm for land surface tem- satellite land surface temperature algorithm using SURFRAD perature retrieval from Landsat 8 thermal infrared sensor ground measurements and statistical estimates of error data,” Remote Sensing, vol. 4, no. 7, pp. 4268–4289, 2015. properties,” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 3, pp. 704–713, 2012. [28] S. Pat and J. Chavez, “Image-based atmospheric corrections - revisited and improved,” Photogrammetric Engineering & Remote Sensing, vol. 62, no. 9, pp. 1025–1036, 1996. [29] X. Hanqiu, “Evaluation of two absolute radiometric nor- malization algorithms for pre-processing of landsat imagery,” Journal of China University of Geosciences, vol. 17, no. 2, pp. 146–157, 2006. [30] J. A. Sobrino, J. C. Jimenez-Munoz, G. Soria et al., “Land surface emissivity retrieval from different VNIR and TIR sensors,” IEEE Transactions on Geoscience and Remote Sensing, vol. 46, no. 2, pp. 316–327, 2008. [31] Z. Qin, W. Li, B. Xu, Z. Chen, and J. Liu, “-e estimation of land surface emissivity for Landsat TM6,” Remote sensing for land & resources, vol. 3, no. 3, pp. 28–36, 2004. [32] A. Julia, J. L. B. Barsi, and J. R. Schott, “An atmospheric correction parameter calculator for a single thermal band earth-sensing instrument,” in IEEE International Geoscience & Remote Sensing Symposium, Yokohama, Japan, August 2003. [33] J. A. Barsi, J. J. Butler, and J. R. Schott, “Validation of a web- based atmospheric correction tool for single thermal band instruments,” Earth Observing Systems X, vol. 10, 2005. [34] N. K. Malakar, G. C. Hulley, S. J. Hook, K. Laraby, M. Cook, and J. R. Schott, “An operational land surface temperature product for landsat thermal data: methodology and valida- tion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 10, pp. 5717–5735, 2018. [35] Z. Zhang, G. He, M. Wang, T. Long, G. Wang, and X. Zhang, “Validation of the generalized single-channel algorithm using Landsat 8 imagery and SURFRAD ground measurements,” Remote Sensing Letters, vol. 7, no. 8, pp. 810–816, 2016. [36] C. Coll, V. Caselles, J. Galve et al., “Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS data,” Remote Sensing of Environment, vol. 97, no. 3, pp. 288–300, 2005. [37] X. Meng, J. Cheng, and S. Zhao, “Estimating land surface temperature from landsat-8 data using the NOAA JPSS en- terprise algorithm,” Remote Sensing, vol. 11, no. 2, 2019. [38] M. Wang, Z. Zhang, and T. Hu, “A practical single-channel algorithm for land surface temperature retrieval: application to landsat series data,” Journal of Geophysical Research: At- mosphere, vol. 33, 2018. [39] Z.-L. Li, B.-H. Tang, H. Wu et al., “Satellite-derived land surface temperature: current status and perspectives,” Remote Sensing of Environment, vol. 131, pp. 14–37, 2013.

Journal

Advances in MeteorologyHindawi Publishing Corporation

Published: Sep 24, 2021

References