Open Access
Add to BibSonomy
Abstract
The land surface temperature (LST) is a key parameter for energy balance, evapotranspiration and climate change. In this study, two new methods of LST retrieval from passive microwave data are developed: one is deriving LST only using single-channel dual-polarized data based on the relationship between the emissivity and microwave polarization difference index (MPDI) (denoted as Method 1); the other one is deriving LST using the traditional multi-channel method with prior knowledge of the normalized difference vegetation index (NDVI) (denoted as Method 3). Taking Moderate Resolution Imaging Spectroradiometer (MODIS) LST products as the actual LSTs, the coefficients for these algorithms are determined. From the results for the year 2008, it is demonstrated that the root mean square errors (RMSEs) for the LST retrieval using Method 3 are the smallest and range from 2.92 K to 3.44 K, the RMSEs for the LST retrieval using traditional multi-channel method (denoted as Method 2) range from 3.07 K to 4.05 K, and the worst results come from Method 1, whose RMSEs range from 3.11 K to 4.13 K at a frequency of 89 GHz. This could be caused by the fact that the NDVI provides substantial emissivity knowledge in Method 3, and much richer vegetation could result in a more accurate emissivity estimation.
© 2017 Optical Society of America
1. Introduction
Land surface temperature (LST) is one of the most important variables in the physical process of surface energy and water balance at local and global scales [1–4] and has been widely used in research areas such as vegetation assessment [5], hydrological cycle [6], heat flux estimation [7], evapotranspiration and climate change studies [8, 9]. Additionally, the LST is recognized as a high-priority parameter of the International Geosphere and Biosphere Program (IGBP) [10]. Due to the impossibility of acquiring ground measurements at a large scale, it is necessary to find an appropriate way to obtain LST data. With the development of remote sensing from space, satellite data, such as thermal infrared (TIR) data and passive microwave (PMW) data, offer the possibility for measuring the LST over the entire globe [11–14].
To date, most LST retrieval methods are based on TIR remotely sensed data, including multiple-channel algorithms [15–17], temperature and emissivity separation algorithms [18–20] and single-channel algorithms [21, 22]. These algorithms have made great progress, and their accuracies can reach up to 1.0 K. However, the main drawback of TIR satellite data is that it will be greatly influenced by weather and atmospheric conditions. When the land surface is covered by clouds, optical remote sensing data such as TIR data will not work [17].
Compared with TIR remotely sensed data, PMW data can overcome atmospheric influences. PMW emissions can penetrate clouds, allowing for the retrieval of the LST under nearly all weather conditions [23]. Several algorithms to retrieve the LST from PMW data have been proposed, including empirical statistical methods, neural networks and physical models. Among them, the physical models are derived from microwave radiation transfer theory with clear physical meanings. However, due to problems with their underdetermined equations (if the radiance is measured in N channels, there will always be N + 1 unknowns, corresponding to N emissivities in each channel and an unknown LST for N equations), physical methods are usually established under certain assumptions or with prior information. Such as, Weng et al. retrieved LST by assuming that the emissivities were the same at the lowest frequencies (19.35 and 22.235 GHz), so that the effects of unknown emissivity can be eliminated by combining the measurements [24]. Fily et al. used an observed linear relationship between the horizontal and vertical emissivities at 19 and 37 GHz to retrieve derive the LST and the fraction of water surface over snow-free areas in sub-arctic and mid-latitude areas [25]. Royer et al. derived LST by establishing the empirical relationship between the vertical polarization and horizontal polarization channel emissivity, and assuming the atmospheric upwelling radiation, downwelling radiation and transmittance are constant [26]. Gao et al. used the physical basis for relating polarization ratio (PR) to estimate emissivity, and then retrieved LST [27]. However, in practice, prior knowledge (such as the emissivity) is difficult to obtain, making the model difficult to create. Meanwhile, empirical statistical methods and neural networks could provide an effective way to retrieve LST from PMW data, and they exhibit a good performance over certain study areas. McFarland et al. found that there is a simple linear relationship between the LST and brightness temperature without the atmospheric influence [28].Owe et al. used the high-frequency (37GHz) vertically polarized observations from satellite radiometers such as Nimbus-7 SMMR or the Special Sensor Microwave/Imager (SSM/I) to estimate LST [29]. Holmes et al. also used the relationship between the Ts and the measured 37 GHz to derived LST at global scales [30]. Chen et al. established a regression model between AMSR-E brightness and observation temperatures (Ts) from meteorological observation stations over Guangdong Province [31]. Salama et al. used the single frequency (37GHz) approach over the Tibetan Plateau for the period 1987 to 2008 [32]. Zhou et al. proposed a temporally land cover-based look-up table (TL-LUT) method to estimate LSTs over Chinese landmass [33]. Zurk et al. used the multilayered perceptron (MLP) neural network to retrieve LST. The MLP neural network was trained with a subset of the generated temperatures, and the remaining temperatures were inverted using a back propagation method [34]. Aires et al. used neural network approach to simultaneously retrieve surface temperature, atmospheric water vapor, cloud liquid water path and emissivites [35].
Due to the fact that the Moderate Resolution Imaging Spectroradiometer (MODIS) and Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E) are onboard the same Aqua satellite platform with the same data acquisition times, an excellent opportunity is provided to retrieve the LST by establishing a statistical relationship between AMSR-E brightness temperature data and MODIS LST products. Gao et al. [36] attempted to retrieve the LST only with the combination of AMSR-E brightness temperatures; due to their lack of consideration of the impacts of emissivity, the accuracy of their LST retrieval was approximately 4.7 K. Liu et al. [37] used the multi-channel method to retrieve LST, the accuracy of which was approximately 2 K. However, this method is only used for desert areas.
In this study, the normalized difference vegetation index (NDVI), which corresponds to knowledge of the emissivity, is added into the traditional multi-channel method (denoted as Method 2) to improve the LST retrieval accuracy (denoted as Method 3). At the same time, a LST retrieval algorithm based on the microwave polarization difference index (MPDI) is used, which may provide a new guide for the retrieval of surface temperatures from microwave data (denoted as Method 1). Furthermore, the three methods are inter-compared using AMSR-E data and taking MODIS LST products as ground measurements.
In this paper, section 2 describes the study area, data sets and data processing scheme. In section 3, descriptions of these three LST retrieval algorithms and the results are presented. The conclusions are drawn in section 4.
2. Study area, data sets and data processing
2.1 Study area
In consideration of various land cover types, the region encompassing the latitudes from 30° N to 50° N and the longitudes from 10° W to 10° E is selected as the study area, which incorporates the Iberian Peninsula, part of the Maghreb region and part of southwestern Europe. The Iberian Peninsula, also known as Iberia, is located in the southwest corner of Europe. The peninsula is principally divided by Portugal and Spain and is mostly composed of their territory. With an area of approximately 582,000 km2, it is the second largest European peninsula [38]. The Maghreb region is usually defined as much or most of the region of western North Africa or Northwest Africa. The whole study area covers 17 typical surface types (Fig. 1).
Fig. 1 Location of the study area. The land cover is derived from the MODIS land cover products (MCD12Q1) in 2008 with a 0.01° spatial resolution.
2.2 Data set descriptions
• AMSR-E data
The AMSR-E sensor is situated onboard the Aqua satellite, whose local overpass times are approximately 13:30 in ascending mode and 01:30 in descending mode simultaneously with the MODIS system. It is a twelve-channel, six-frequency passive-microwave radiometer system that measures horizontally (H) and vertically (V) polarized brightness temperatures at 6.9 GHz, 10.7 GHz, 18.7 GHz, 23.8 GHz, 36.5 GHz and 89.0 GHz. The viewing angle of AMSR-E data is 53°, therefore, the viewing angle is not taken into consideration in this study. The spatial resolutions of the individual measurements vary from 5.4 km at 89 GHz to 56 km at 6.9 GHz [39]. The AMSR-E daily brightness temperature products with a 0.25° spatial resolution during the period from 2008 to 2009 are downloaded from National Snow and Ice Data Center (NSIDC) and are used in this study.
• MODIS surface products
The MODIS/Aqua land surface temperature and emissivity (LST/E) product (short name: MYD11B1) and monthly vegetation index product (short name: MYD13A3) are selected during the period from the year 2006 to 2010. The MYD11B1 product, which is used as the actual LST, is tile-based and gridded in a sinusoidal projection and is produced daily at a resolution of 5.6 km [40]. The global MYD13A3 data provides monthly at a 1-km spatial resolution as a gridded level-3 product in a sinusoidal projection [41]. The study area is represented by the tiles h17v04, h17v05, h18v04 and h18v05 in the sinusoidal grid, and every tile in either product is transformed onto a geographic projection through the MODIS reprojection tool (MRT). For the MYD11B1 product, the scientific data sets of the daytime LSTs and quality control (QC) are selected; for the MYD13A3 product, the NDVI data sets are derived.
2.3 Data processing
Because of the different spatial resolutions, a pixel aggregation algorithm [see Eq. (1)] is used to acquire the MODIS LST at the same spatial resolution of AMSR-E data (0.25°).
in which R is the aggregated value of the target pixel, N is the total pixel count in the target pixel, ωj is the weight of pixel j, Sj,p is the partial area of pixel j overlapping with the target pixel, Sj is the total area of pixel j, and Rj is the radiance value of the pixel j.The QC information in the MODIS LST product provides the quality level of each pixel; it provides information for whether the algorithm results are nominal or abnormal or if other defined conditions are encountered within a pixel [40]. In this study, QC information is used to determine the usefulness of the LST data when aggregating the MODIS LST and AMSR-E products in a spatial resolution of 0.25°. An aggregated pixel is recognized as useful when the number (N) of the qualified MODIS pixels (QC = 0 or QC = 8) is larger than 20 (80% of 25). In addition, the pixels that are classified as water, snow or ice are excluded from the LST retrieval in this study
3. LST retrieval methods and results
3.1 Passive microwave radiation transfer theory
The passive microwave radiation transfer equation, which can be written as Eq. (2), describes the total radiation intensity observed by a satellite radiometer [42]. It can be found that the radiation received by a sensor is sourced not only from surface radiation but also from atmospheric upward and downward radiations, which occupy a tiny fraction.
where Bf is the Planck function, Tf is the brightness temperature at a frequency f, is the atmospheric transmittance at a frequency f in the viewing direction (zenith angle from nadir), is the LST, is the ground emissivity, and and are the upwelling and downwelling path radiances, respectively.The Planck function describes the relationship between the spectral radiance emitted by a black body and its temperature, and is shown in Eq. (3).
in which is the blackbody radiance in W·m−2·sr−1·Hz−1, T is the blackbody temperature in K, h is the Planck constant (6.63 × 10−34 J·s), f is the frequency in Hz, k is the Boltzmann constant (1.38 × 10−23 J·K−1), and c is the speed of light (2.992458 × 108 m·s−1). According to the Raleigh-Jeans approximation for the Planck function, Eq. (2) can be simplified as3.2 LST retrieval with the prior knowledge of the MPDI
Since the wavelengths of microwaves are much larger than TIR wavelengths, their emission is influenced slightly by the atmosphere [43]. Equation (4) can be simplified further as Eq. (5) when ignoring the effects of the atmosphere, especially in the case of low frequencies.
As demonstrated in Eq. (5), the surface emissivity is a key parameter for the retrieval of the LST. The MPDI is defined as the ratio of the difference between the brightness temperatures in the V and H polarizations to the sum of the two terms, which is shown in Eq. (6). It is used to describe the characteristics of the land surface cover density, which is used to retrieve soil moisture.
in which MPDIf means the MPDI at a frequency f, and TbV and TbH are the brightness temperatures in the V and H polarizations, respectively.In view of this, an attempt for estimating the emissivity from the MPDI is made. With the prior knowledge of the AMSR-E brightness temperatures, the corresponding MPDI can be derived using Eq. (6). Subsequently, the emissivity is calculated from the AMSR-E brightness temperatures and MODIS LSTs using Eq. (5). The relationship between the emissivity and MPDI for the data sets in 2008 is shown in Fig. 2. It can be found that there’re significant differences between horizontally and vertically polarized microwave emissions. And the changes in polarized effects with frequency are partially dependent on intervening scatters and absorbers/emitters such as clouds and precipitation. Moreover, it is worth noting that the relationship in the H polarization is better than that in the V polarization; the coefficient of determination (R2) is larger than 0.8 in the H polarization and is smaller than 0.25 in the V polarization, while the emissivity in the V polarization changes much less than that in the H polarization. This illustrates that the same noise would result in fewer errors in the V polarization than in the H polarization.
With the aid of the data set shown in Fig. 2, a quadratic polynomial is used to describe the relationship between the emissivity and MPDI [see Eq. (7)]. The unknown coefficients and the results for the emissivities obtained from Eq. (7) are shown in Table 1. From Table 1, it can be seen that the root mean square errors (RMSEs) of the emissivities retrieved using Eq. (7) are approximately 0.01 in both the V and H polarizations.
where and MPDIf are the emissivity and MPDI at a frequency f, respectively, , and are unknown coefficients, and p is the V or H polarization.
Table 1. The coefficients aM1,f,p, bM1,f,p and cM1,f,p, and the results for the emissivities derived from Eq. (7).
Combining Eq. (5) and Eq. (7), Eq. (8) can be obtained
First, the coefficients aM1,f,p, bM1 f,p and cM1 f,p can be obtained from Eq. (7) with the prior knowledge of the MPDI and emissivity throughout all of 2008; then, the LST is retrieved using Eq. (8) for different months. Figure 3 shows the LST retrieval results for different polarizations using Eq. (8), which is denoted as Method 1. It can be found that the RMSEs, which are the results of comparison between the AMSR-E derived LSTs and the MODIS LSTs retrieved from thermal infrared data, for the two polarizations are almost the same, and the difference between them is within 0.06 K. Moreover, the results get better when the frequency increases, and the best results appear at the frequency of 89 GHz, for which the RMSEs range from 3.1 K to 4.1 K, while the worst RMSEs range from 4.3 K to 5.7 K at the frequency of 6.9 GHz. This could be caused by the fact that, in the absence of pronounced atmospheric scattering and re-emission, higher frequency channels provide increased accuracies in the determination of LSTs due to a shallower emitting depth than for lower frequencies [44].
Fig. 3 LST retrieval using Method 1 for different polarizations. (a) H polarization; (b) V polarization
3.3 Multi-channel algorithm
Different microwave channels contain different atmospheric and surface information. Theoretically, the 6.9 GHz channel has a longer wavelength and contains more surface information; meanwhile, the 23.8 GHz channel is affected by the atmospheric water vapor content, and thus, the measurements contain much more atmospheric information. The measurements obtained at 37 GHz are valuable for the retrieval of LSTs because this channel balances a reduced sensitivity to soil surface characteristics with a relatively high atmospheric transmittance [36]. The brightness temperatures at 89.0 GHz show a strong correlation with the MODIS LSTs due to a smaller penetration depth and a higher emissivity at 89.0 GHz [45]. From Eq. (4), a linear relationship between the brightness temperature and LST can be clearly observed. Thus, a linear regression model can be established to retrieve the LST from the microwave brightness temperature among various AMSR-E channels [see Eq. (9)] by taking MODIS LSTs as ground LST measurements.
in which Ts is the LST, TbV and TbH are the brightness temperatures in the V polarization and H polarization, respectively, fi is the frequency (6.9 GHz, 10.7 GHz, 18.7 GHz, 23.8 GHz, 36.5 GHz and 89.0 GHz corresponding to i from 1 to 6), and a_tM2, b_ti,M2 and c_ti, M2 are the fitting coefficients in Method 2.To choose the optimal channels for the LST retrieval, a stepwise linear regression method, which is a method of regressing multiple variables while simultaneously removing those that are unimportant, is used to select the channels. This method essentially performs a multiple regression a number of times, and the weakest correlated variable is removed during each iteration. At the end, the remaining variables can explain the distribution the most effectively. The results of regression analyses for the data set in 2008 are displayed in Table 2. It can be found that the RMSE for the retrieved LST using a combination of 9 channels (TbVf6, TbVf1, TbVf5, TbVf4, TbHf1, TbHf5, TbHf4, TbHf3, TbHf6) is the smallest, and the RMSE will not decrease when more channels are added into the retrieval algorithm. Therefore, these 9 channels (TbVf1, TbVf4, TbVf5, TbVf6, TbHf1, TbHf3, TbHf4, TbHf5, TbHf6) are selected to derive the LST using Eq. (10), and the fitting coefficients are determined using the whole data set in 2008 (Table 3).
Table 2. RMSEs for the LST retrieval from different channel combinations.
Table 3. The coefficients in Eq. (10)
Figure 4(a) shows the RMSE of the linear fitting. It is worth noting that most of the points are distributed near the 1:1 line (black dotted line). The R2 and RMSE are 0.92 and 3.45 K, respectively.
Fig. 4 Scatterplot between the AMSR-E-derived LST and the MODIS LST in the year 2008. The colors in the scatterplot correspond to the density of points. (a) for Method 2; (b) for Method 3
Subsequently, the coefficients in Table 3 are applied to the twelve different months in 2008, and the corresponding RMSEs are shown in Fig. 5. To reduce the uncertainties in the quantity of samples on the algorithm performance evaluation, the monthly averaged RMSE is used. It can be found that the RMSEs of Method 2 range from 3.1 K to 4.1 K and that the RMSEs are larger than 3.7 K in April, May and October, while the RMSEs are all lower than 3.5 K in the other months.
Fig. 5 RMSEs of LST retrieval using Eq. (10) and Eq. (11) during the year 2008 over the twelve months
3.4 Multi-channel algorithm with the prior knowledge of the NDVI
The NDVI is one of the parameters that can characterize the emissivity, and therefore, the NDVI value is added into Eq. (10) to reduce the uncertainty resulting from the lack of emissivity in the LST retrieval (see Eq. (11), denoted as Method 3). The coefficients are shown in Table 4.
in which dM3 is also a fitting coefficient.Table 4. The coefficients in Eq. (11)
With the aid of the whole data set in 2008, the 11 coefficients in Eq. (11) are fitted (Table 4), and the corresponding RMSE is 3.13 K. Figure 4(b) shows the scatterplot between the AMSR-E-derived LST using Method 3 and the MODIS LST in the year 2008. It is worth noting that most of the points are distributed near the 1:1 line (black dotted line), and the R2 and RMSE are 0.94 and 3.13 K, respectively.
Subsequently, the coefficients in Table 3 are applied to the twelve different months in 2008, and the corresponding RMSEs are shown in Fig. 5. It can be found that the RMSEs of Method 3 range from 2.7 K to 3.7 K and that the RMSEs are larger than 3.6 K in April, May and October, while the RMSEs are all lower than 3.2 K in the other months.
Moreover, to validate the generalization of the three methods (two newly developed algorithms: Method 1 and Method 3, and one published algorithm: Method 2), the aforementioned three methods and their corresponding coefficients are applied to monthly AMSR-E data for the year 2006, 2007, 2009, and 2010, and the results are shown in Fig. 6. It is worth noting that the results are relatively better for Method 2 and Method 3 than Method 1, as the RMSEs for Method 2 and Method 3 are both lower than 3.8 K for the different months of the four years, while Method 1 (89 GHz in the V polarization) performs more poorly with RMSEs ranging from 2.8 K to 4.6 K. This illustrates that Method 2 and 3 are more generalized than Method 1. Moreover, it can be found that in summer (months 5, 6, 7, and 8), Method 2 and Method 3 perform much better than Method 1. The possible reason maybe that the atmospheric effect is lager in summer than in other seasons. Method 2 and Method 3 can reduce the atmospheric effect by taking advantage of the linear combination of brightness temperature. Therefore, the results of Method 2 and Method 3 in summer are much better than that of Method 1.
Fig. 6 LST retrieval using the coefficients of 2008 for the data sets of 2006, 2007, 2009, and 2010.
4. Conclusions
In this study, two new methods of LST retrieval from passive microwave data are developed, namely, Method 1 and Method 3. Method 1 retrieves LST using single-channel dual-polarized data without any other prior knowledge such as emissivity, and this method is just suitable for one channel data; Method 3 derives LST using the multi-channel method with the information of NDVI. On the basis of the AMSE-R data and MODIS LST, the two new methods are compared with the traditional multi-channel method (Method 2) over the study area encompassing the latitudes from 30° N to 50° N and the longitudes from 10° W to 10° E.
The results demonstrated that RMSEs for Method 3 range from 2.92 K to 3.44 K during 2008, whereas the corresponding RMSEs for Method 2 range from 3.07 K to 4.05 K; the worst results come from Method 1, whose RMSEs range from 3.35 K to 4.15 K at a frequency of 89 GHz. It can be clearly found that the LST retrievals using Method 2 and Method 3 performed better than that using Method 1. This could be caused by the uncertainty in the emissivity estimation due to the simplification of the radiative transfer equation and the empirical formula; thus, the error in the LST retrieval using a single channel is larger than that using multiple channels. Moreover, compared with Method 2, the accuracies of the LSTs retrieved from Method 3 can be improved by approximately 0.3 K on average. This could be caused by the fact that the NDVI provides much more knowledge regarding the emissivity in Method 3. Consequently, Method 3 is the most suitable approach for the retrieval of LSTs from passive microwave data.
In addition, a validation of the generalization of these three methods is analyzed. The LST is retrieved from the AMSR-E data during the year 2006, 2007, 2009 and 2010 using the coefficients from 2008. The results show that Method 2 and Method 3 exhibit a better generalization approach than Method 1 (89 GHz in the V polarization). The RMSEs of Methods 2 and 3 are both under 3.8 K, and a higher accuracy can be derived using Method 3.
Moreover, in the case of cloud cover, the MODIS LST over study area is not available, except for ground station LST measurements, which are representative for the corresponding values at small areas (cm2 ~m2), rather than those at large pixel-size scale (102 km2). Therefore, the LST retrieval under cloudy conditions is not carried out in this study. Actually, clouds and precipitation affect microwave radiation strongly in frequency and polarization. The penetration depth of microwave signal depends on the ratio between the hydrometeor size and the wavelength, and on the phase of the hydrometeors, ice, or liquid. And the changes in polarized effects with frequency are partially dependent on intervening scatters and absorbers/emitters such as clouds and precipitation. This could be used to enhance microwave LST retrieval in cloudy conditions. Therefore, the performances of these three methods under cloudy conditions will be carried out in the future work.
Funding
This work has been supported by National Nature Science Foundation of China (41231170, 41571352 and 41671370).
Acknowledgments
The thanks are given to the anonymous reviewers for their valuable suggestions and comments.
References and links
1. J. Hansen, R. Ruedy, M. Sato, and K. Lo, “Global surface temperature change,” Rev. Geophys. 48(4), RG4004 (2010). [CrossRef]
2. A. Basist, N. C. Grody, T. C. Peterson, and C. N. Williams, “Using the Special Sensor Microwave/Imager to monitor land surface temperatures, wetness, and snow cover,” J. Appl. Meteorol. 37(9), 888–911 (1998). [CrossRef]
3. C. André, C. Ottlé, A. Royer, and F. Maignan, “Land surface temperature retrieval over circumpolar Arctic using SSM/I–SSMIS and MODIS data,” Remote Sens. Environ. 162, 1–10 (2015). [CrossRef]
4. Y.-G. Qian, N. Wang, L.-L. Ma, Y.-K. Liu, H. Wu, B.-H. Tang, L.-L. Tang, and C.-R. Li, “Land surface temperature retrieved from airborne multispectral scanner mid-infrared and thermal-infrared data,” Opt. Express 24(2), A257–A269 (2016). [CrossRef] [PubMed]
5. F. N. Kogan, “Operational space technology for global vegetation assessment,” Bull. Am. Meteorol. Soc. 82(9), 1949–1964 (2001). [CrossRef]
6. Z. Su, “The Surface Energy Balance System (SEBS) for estimation of turbulent heat fluxes,” Hydrol. Earth Syst. Sci. Discuss. 6(1), 85–100 (2002). [CrossRef]
7. R. Zhang, J. Tian, H. Su, X. Sun, S. Chen, and J. Xia, “Two improvements of an operational two-layer model for terrestrial surface heat flux retrieval,” Sensors (Basel) 8(10), 6165–6187 (2008). [CrossRef] [PubMed]
8. Y. Imada, S. Maeda, M. Watanabe, H. Shiogama, R. Mizuta, M. Ishii, and M. Kimoto, “Recent Enhanced Seasonal Temperature Contrast in Japan from Large Ensemble High-Resolution Climate Simulations,” Atmosphere 8(3), 57 (2017). [CrossRef]
9. D. A. Efthymiadis and P. D. Jones, “Assessment of maximum possible urbanization influences on land temperature data by comparison of land and marine data around coasts,” Atmosphere 1(1), 51–61 (2010). [CrossRef]
10. J. Townshend, C. Justice, D. Skole, J.-P. Malingreau, J. Cihlar, P. Teillet, F. Sadowski, and S. Ruttenberg, “The 1 km resolution global data set: needs of the International Geosphere Biosphere Programme,” Int. J. Remote Sens. 15(17), 3417–3441 (1994). [CrossRef]
11. F. Becker and Z.-L. Li, “Temperature-independent spectral indices in thermal infrared bands,” Remote Sens. Environ. 32(1), 17–33 (1990). [CrossRef]
12. P. Dash, F.-M. Göttsche, F.-S. Olesen, and H. Fischer, “Land surface temperature and emissivity estimation from passive sensor data: theory and practice-current trends,” Int. J. Remote Sens. 23(13), 2563–2594 (2002). [CrossRef]
13. S.-B. Duan, Z.-L. Li, and P. Leng, “A framework for the retrieval of all-weather land surface temperature at a high spatial resolution from polar-orbiting thermal infrared and passive microwave data,” Remote Sens. Environ. 195, 107–117 (2017). [CrossRef]
14. Z.-L. Li, B.-H. Tang, H. Wu, H. Ren, G. Yan, Z. Wan, I. F. Trigo, and J. A. Sobrino, “Satellite-derived land surface temperature: Current status and perspectives,” Remote Sens. Environ. 131, 14–37 (2013). [CrossRef]
15. F. Becker and Z.-L. Li, “Towards a local split window method over land surfaces,” Remote Sens. 11(3), 369–393 (1990). [CrossRef]
16. Z. Wan and J. Dozier, “A generalized split-window algorithm for retrieving land-surface temperature from space,” IEEE Trans. Geosci. Remote Sens. 34(4), 892–905 (1996). [CrossRef]
17. X. Zhang and L. Li, “A method to estimate land surface temperature from Meteosat Second Generation data using multi-temporal data,” Opt. Express 21(26), 31907–31918 (2013). [CrossRef] [PubMed]
18. A. Gillespie, S. Rokugawa, T. Matsunaga, J. S. Cothern, S. Hook, and A. B. Kahle, “A temperature and emissivity separation algorithm for Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images,” IEEE Trans. Geosci. Remote Sens. 36(4), 1113–1126 (1998). [CrossRef]
19. C. Coll, V. García-Santos, R. Niclòs, and V. Caselles, “Test of the MODIS Land Surface Temperature and emissivity separation algorithm with ground measurements over a rice paddy,” IEEE Trans. Geosci. Remote Sens. 54(5), 3061–3069 (2016). [CrossRef]
20. X. OuYang, N. Wang, H. Wu, and Z. L. Li, “Errors analysis on temperature and emissivity determination from hyperspectral thermal infrared data,” Opt. Express 18(2), 544–550 (2010). [CrossRef] [PubMed]
21. J. C. Jimenez-Muñoz and J. A. Sobrino, “A generalized single‐channel method for retrieving land surface temperature from remote sensing data,” J. Geophys. Res., D Atmospheres 108, 4688 (2003). [CrossRef]
22. Z. Qin, A. Karnieli, and P. Berliner, “A mono-window algorithm for retrieving land surface temperature from Landsat TM data and its application to the Israel-Egypt border region,” Int. J. Remote Sens. 22(18), 3719–3746 (2001). [CrossRef]
23. Z.-L. Liu, H. Wu, B.-H. Tang, S. Qiu, and Z.-L. Li, “Atmospheric corrections of passive microwave data for estimating land surface temperature,” Opt. Express 21(13), 15654–15663 (2013). [CrossRef] [PubMed]
24. F. Weng and N. C. Grody, “Physical retrieval of land surface temperature using the special sensor microwave imager,” J. Geophys. Res., D, Atmospheres 103(D8), 8839–8848 (1998). [CrossRef]
25. M. Fily, A. Royer, K. Goıta, and C. Prigent, “A simple retrieval method for land surface temperature and fraction of water surface determination from satellite microwave brightness temperatures in sub-arctic areas,” Remote Sens. Environ. 85(3), 328–338 (2003). [CrossRef]
26. A. Royer and S. Poirier, “Surface temperature spatial and temporal variations in North America from homogenized satellite SMMR‐SSM/I microwave measurements and reanalysis for 1979–2008,” J. Geophys. Res., D, Atmospheres 115, D08110 (2010). [CrossRef]
27. H. Gao, R. Fu, R. E. Dickinson, and R. I. Negron Juare, “A practical method for retrieving land surface temperature from AMSR-E over the amazon forest,” IEEE Trans. Geosci. Remote Sens. 46(1), 193–199 (2008). [CrossRef]
28. M. J. McFarland, R. L. Miller, and C. M. Neale, “Land surface temperature derived from the SSM/I passive microwave brightness temperatures,” IEEE Trans. Geosci. Remote Sens. 28(5), 839–845 (1990). [CrossRef]
29. M. Owe and A. Van De Griend, “On the relationship between thermodynamic surface temperature and high-frequency (37 GHz) vertically polarized brightness temperature under semi-arid conditions,” Int. J. Remote Sens. 22(17), 3521–3532 (2001). [CrossRef]
30. T. Holmes, R. De Jeu, M. Owe, and A. Dolman, “Land surface temperature from Ka band (37 GHz) passive microwave observations,” J. Geophys. Res., D, Atmospheres 114, D04113 (2009). [CrossRef]
31. S. Chen, X. Chen, W. Chen, Y. Su, and D. Li, “A simple retrieval method of land surface temperature from AMSR-E passive microwave data—A case study over Southern China during the strong snow disaster of 2008,” Int. J. Appl. Earth Obs. Geoinf. 13(1), 140–151 (2011). [CrossRef]
32. M. S. Salama, R. Van der Velde, L. Zhong, Y. Ma, M. Ofwono, and Z. Su, “Decadal variations of land surface temperature anomalies observed over the Tibetan Plateau by the Special Sensor Microwave Imager (SSM/I) from 1987 to 2008,” Clim. Change 114(3-4), 769–781 (2012). [CrossRef]
33. J. Zhou, F. Dai, X. Zhang, S. Zhao, and M. Li, “Developing a temporally land cover-based look-up table (TL-LUT) method for estimating land surface temperature based on AMSR-E data over the Chinese landmass,” Int. J. Appl. Earth Obs. Geoinf. 34, 35–50 (2015). [CrossRef]
34. L. M. Zurk, D. Davis, E. G. Njoku, L. Tsang, and J.-N. Hwang, “Inversion of parameters for semiarid regions by a neural network,” NASA Tech. Rep. 19930063847 (1992). [CrossRef]
35. F. Aires, C. Prigent, W. Rossow, and M. Rothstein, “A new neural network approach including first guess for retrieval of atmospheric water vapor, cloud liquid water path, surface temperature, and emissivities over land from satellite microwave observations,” J. Geophys. Res., D, Atmospheres 106(D14), 14887–14907 (2001). [CrossRef]
36. C. Gao, X. Jiang, Y. Qian, S. Qiu, L. Ma, and Z.-l. Li, “A neural network based method for land surface temperature retrieval from AMSR-E passive microwave data,” in Geoscience and Remote Sensing Symposium (IGARSS),2013IEEE International, (IEEE, 2013), 469–472. [CrossRef]
37. Z. Liu, B. Tang, and Z. Li, “Calculation of land surface temperature based on AMSR-E data,” Science & Technology Review, 24–27 (2009).
38. J. Gomes, “Late-quaternary landscape dynamics in the Iberian Peninsula and Balearic Islands,” (Dissertation, Universidade do Porto, Portugal, 2007).
39. “AMSR-E/Aqua Daily Gridded Brightness Temperatures”, retrieved http://nsidc.org/data/docs/daac/nsidc0301_amsre_gridded_tb.gd.html.
40. Z. Wan, “MYD11B1 MODIS/Aqua Land Surface Temperature/Emissivity Daily L3 Global 6km SIN Grid V006. NASA EOSDIS Land Processes DAAC.” (2015), retrieved https://doi.org/10.5067/MODIS/MYD11B1.006.
41. K. Didan, MYD13A3 MODIS/Aqua Vegetation Indices Monthly L3 Global 1km SIN Grid V006. NASA EOSDIS Land Processes DAAC. ” (2015), retrieved https://doi.org/10.5067/MODIS/MYD13A3.006.
42. E. G. Njoku and L. Li, “Retrieval of land surface parameters using passive microwave measurements at 6-18 GHz,” IEEE Trans. Geosci. Remote Sens. 37(1), 79–93 (1999). [CrossRef]
43. F. T. Ulaby, R. K. Moore, and A. K. Fung, “Microwave remote sensing active and passive-volume III: from theory to applications,” (1986).
44. F.-C. Zhou, X. Song, P. Leng, and Z.-L. Li, “An Effective Emission Depth Model for Passive Microwave Remote Sensing,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 9(4), 1752–1760 (2016). [CrossRef]
45. K. Mao, “The study of algorithm for retrieving land surface temperature and soil moisture from thermal and microwave data,” MS thesis, Inst. Remote Sens. Appl., Chin. Acad. Sci., Univ. Chin. Acad. Sci., Beijing, China, Tech. Rep, 216 (2007).
