Land Surface Temperature and Emissivity Retrieval ... - IEEE Xplore

2 downloads 0 Views 2MB Size Report
Abstract—This work addressed the retrieval of land surface emissivity (LSE) and land surface temperature (LST) by using. Middle Infra-Red (MIR) and Thermal ...
1552

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013

Land Surface Temperature and Emissivity Retrieval From Time-Series Mid-Infrared and Thermal Infrared Data of SVISSR/FY-2C Yonggang Qian, Shi Qiu, Ning Wang, Xiangsheng Kong, Hua Wu, and Lingling Ma

Abstract—This work addressed the retrieval of land surface emissivity (LSE) and land surface temperature (LST) by using Middle Infra-Red (MIR) and Thermal Infra-Red (TIR) channels from the data acquired by the Stretched Visible and Infrared Spin Scan Radiometer (SVISSR) onboard Chinese geostationary meteorological satellite FengYun 2C (FY-2C). SVISSR/FY-2C sensor acquires image covering the full disk with a temporal resolution of 30 minutes. The LST and LSE retrieval procedures can be shown as follows. Firstly, taking into the fact that land surface is non-lambertian characteristics, the time-series bi-directional reflectances in SVISSR/FY-2C MIR channel 4 (3.8 ) were estimated from the combined MIR and TIR channels with day-night SVISSR/FY-2C data. A diurnal temperature cycle (DTC) model was used to correct for the atmospheric effects. The atmospheric profile data provided by European Centre for Medium-Range Weather Forecasts (ECMWF) were adopted with the aid of the radiative transfer code (MODTRAN 4.0). Secondly, a Bidirectional Reflectance Distribution Function (BRDF) model named as RossThick-LiSparse-R model was used to estimate the hemispherical directional reflectance in MIR channel from the time-series bi-directional reflectance data. Then, the LSE in MIR channel can be retrieved according to Kirchhoff’s law. The LSEs in TIR channels can be estimated based on the Temperature Independent Spectral Indices (TISI) concept. And the LST can be retrieved using the split-window algorithm. Finally, a cross-validation method was used to evaluate the retrieval accuracies with the Moderate-resolution Imaging Spectroradiometer (MODIS) MOD11B1 LST/LSE V5 product. The results demonstrated that the emissivities in 11 and 12 were underestimated approximately 0.003 and 0.004 compared with MOD11B1 LSE product over the study area. The FY-2C LST were overestimated approximately 1.65 K and 2.87 K during the night-time and day-time, respectively, compared with MOD11B1 LST product over the study area. Index Terms—Atmospheric correction, diurnal temperature cycle model, land surface emissivity, land surface temperature, SVISSR/FY-2C, TISI concept. Manuscript received December 16, 2012; revised February 26, 2013; accepted March 28, 2013. Date of publication May 13, 2013; date of current version June 17, 2013. This work was supported in part by the National Natural Science Foundation of China under Grants 41101330, 41271342, and 41201367. (Corresponding author: H. Wu.) Y. Qian is with the Key Laboratory of Quantitative Remote Sensing Information Technology, Academy of Opto-Electronics, Chinese Academy of Sciences, Beijing 100094, China (e-mail: [email protected]). S. Qiu is with the LSIIT, UdS, CNRS, 67412 Illkirch, France. N. Wang and L. Ma are with Academy of Opto-Electronics, CAS, Beijing 100094, China (e-mail: [email protected]; [email protected]). X. Kong is with College of Geography and Planning, Ludong University, Yantai 264025, China (e-mail: [email protected]). H. Wu is with the State Key Lab of Resources and Environmental Information System, Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China (e-mail: [email protected]) Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTARS.2013.2259146

I. INTRODUCTION

L

AND SURFACE TEMPERATURE (LST) is a key variable for studying global or regional land surface processes, energy and water cycle. Land surface emissivity (LSE), as an intrinsic property of the natural materials, is often regarded as an indicator of material composition, especially for the silicate minerals, which depends on the composition, roughness, and moisture content of the surface and on the observation conditions [1], [2]. Direct estimation of LST/LSE from passive satellite measurements is very difficult due to the coupling of the LST, LSE and the atmospheric contamination [3]–[5]. The top-of-atmosphere (TOA) radiance is influenced by surface properties (LST and LSEs) and atmosphere. Besides the complications due to atmospheric absorption/emission and surface reflection, a separation of temperature and emissivity from passive measurements at ground level is still a great challenge because the problem is mathematical underdetermined. For a sensor with N observation channels, there will always be (N channel emissivities and one surface temperature) unknowns even that other parameters are known. In order to solve this ill-posed problem, different assumptions and constraints have been proposed either by increasing the number of equations or by reducing the number of unknowns [1], [6]. To date, the retrieval method based on Temperature Independent Spectral Indices (TISI) concept proposed by Becker and Li [6] has been applied to simultaneously retrieve both LST and LSE for several years because of its physics based nature and no rigorous assumption except for the constant TISI between day and night. Jiang et al. [4] developed the diurnal temperature cycle (DTC) model based on the DTC model proposed by Schädlich et al. [7] to correct for the atmospheric effects and used this TISI method between mid-infrared and thermal infrared data of Spinning Enhanced Visible and Infra-Red Imager (SEVIRI) data on board Meteosat Second Generation (MSG) to retrieve MSG-SEVIRI LSE. FengYun 2C (FY-2C) is the first operational geostationary meteorological satellite of China, which was successfully above launched on October 19, 2004 and located at 105 the equator. The payload of FY-2C is the Stretched Visible Infrared Spin Scan Radiometer (SVISSR). It measures the TOA radiances in 1 Middle Infra-Red (MIR) channels, 3 Thermal Infra-Red (TIR) channels, and 1 visible channel, with a spatial resolution of 5 km, 5 km and 1.25 km at the nadir, respectively, as shown in Table I. FY-2C has a high temporal resolution and

1939-1404/$31.00 © 2013 IEEE

QIAN et al.: LAND SURFACE TEMPERATURE AND EMISSIVITY RETRIEVAL FROM TIME-SERIES MID-INFRARED AND THERMAL INFRARED DATA

1553

TABLE I SPECTRAL CHANNEL CHARACTERISTICS OF SVISSR/FY-2C

: Noise Equivalent Temperature Difference; SNR: Signal to Noise Ratio,

can scan the full disk in every 30 minutes during approximately July to August for the requirement of weather forecast and in every one hour during other times. Because the same observation angle for a fixed pixel on the disk, which means there is no angle variation of LSEs for a short time, it is very suitable to study the retrieval method of LST and LSEs by using the time-series data of FY-2C. In this paper, we mainly focus on the retrieval of both LST and LSEs with the MIR channel of IR4 (3.8 ) and TIR channels of IR1 (10.9 ) and IR2 (11.9 ) in SVISSR/FY-2C sensor. Section II presents the theoretical methods, including the DTC model for atmospheric correction, TISI concept, RossThick-LiSparse-R model, and the retrieval method of LSEs and LST. Section III introduces the data and study area. Section IV gives the results and discussions. Finally, the conclusions are drawn in Section V.

.

is the solar irradiance at the TOA, and are solar zenith and azimuth angles, and is the total atmospheric transmittance along the sun to target path. For the measurements in the TIR channel and the night-time measurements in the MIR channel, (1) and (2) can be simplified as: (3) (4) The TISIs can be constructed in the following expression [5] using the daytime and nighttime data in FY-2C MIR and TIR channels according to (1)–(4). (5) (6)

II. METHODOLOGY A. Radiative Transfer Equation and TISI Concept at the Based on radiative transfer theory, the radiance TOA in infrared channel of the sensor is written as follows for a cloud-free atmosphere in thermodynamic equilibrium (Li et al. [3]). (1) (2) is the TOA brightness temperature in channel , where is the land surface emissivity, is the channel atmospheric transmittance, and are the observed channel radiance and the brightness temperature at ground level, respectively. is the upwelling atmospheric channel radiance, is the thermal path upward radiance resulting from the scattering of solar radiation, is the channel downward atmospheric radiance, defined as times the total downward atmospheric radiance, is the channel downward solar diffusion radiation over a hemisphere divided by . is the surface bidirectional reflectance in channel , is the solar irradiance at ground level and equals to .

Where, and denote the MIR and TIR channels. is the land surface emissivity. is the radiance at ground level. The superscripts and in TISI denote that the acquisition times of FY-2C data are nighttime and daytime, respectively. is the fitting coefficient between radiance and temperature for a black body, i.e. . The coefficients and depend on the spectral characteristics of the sensor [6] and have been determined by the least squares fit method (shown in Table II). is the surface bidirectional reflectance in channel , is the solar irradiance at ground level. The correction factors , , , and will be estimated by the method proposed by Nerry et al. [8], who revealed that the uncertainties in the estimation of those correction factors have a minor impact on final results. Hence, the bi-directional reflectance in MIR channel can be retrieved directly from the radiances at ground level by assuming . (7) Where, reflectance in FY-2C MIR channel.

,

is the bi-directional

1554

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013

TABLE II AND AND FITTING ERRORS IN FY-2C VALUES OF THE COEFFICIENTS IR CHANNELS

et al. [4] was used to perform the atmospheric correction. The DTC model with six parameters is shown as follows.

(8) , and . The parameters and are the unknown coefficients. is the angular frequency. is the time at which temperature reaches the maximum. is the starting time of attenuation, and is the decay coefficient. and are the daytime and nighttime LST, respectively. The DTC model has six parameters, , , , , , and , in two parts. The evolution of the daytime LST was modeled by a cosine function based on the thermal diffusion equation, and the decay of the LST at night was described in conjunction with an exponential function assuming that the natural surface follows Newton’s law of cooling [8]. However, there are six parameters in DTC model and ECMWF only provides atmospheric profiles at four times every day. Two assumptions are made for the DTC model to correct for the atmospheric effects in the TIR images: the DTC at the TOA and that at ground level have the same and . This assumption is different to that of Jiang et al. [4]. Actually, all parameters in DTC at TOA and at ground level always have some discrepancies. Jiang et al. [4] assumed the DTC at TOA and at ground level have the same angular frequency and ts, but no reason can be found. The dissertation by Zhang Xiaoyu [5] in Section 6.3 gave the detailed analysis. To fit the DTC model using TOA temperature and ground brightness temperature, small discrepancies of and have been got from four sites and even more advantageous for is found. Based on above analysis, the assumption that the DTC at TOA and at ground level have the same and has been used in this work. The and can be determined through fitting DTC model with the brightness temperatures at TOA, and then they could be regarded as known in the fitting procedure at ground level. Taking into account the difference in the spatial resolutions between FY-2C data with 5 km and ECMWF data with 0.5 , the atmospheric parameters estimated using ECMWF data with MODTRAN 4 had to be interpolated to match the FY-2C data. Over the study area, the atmospheric correction of the FY-2C images in TIR channels IR1, IR2 and MIR channel IR4 at UTC times 6:00, 12:00, 18:00 and 0:00 were performed by using MODTRAN 4.0 with the synchronous ECMWF atmospheric profiles, respectively. For FY-2C MIR channel IR4, time-nearest atmospheric data were used for the images acquired from other times. The direct solar irradiances at ground level were calculated by the use of MODTRAN 4.0 with the approximate atmospheric profiles. For the images in FY-2C TIR channels IR1 and IR2, the DTC model was used to correct for the atmospheric effects as described above. Where,

The fitting coefficients are obtained under a temperature range of 270 K 330 K, step: 0.1 K

The bi-directional reflectances in FY-2C MIR channel were calculated from local time 7:00 to 17:00 for each clear-sky pixel with an interval of 1 hour from (7). Those observations with solar zenith angle greater than 60 were discarded (Jiang, et al. [4]) and the images acquired at UTC time 18:00, i.e., the local time of about 02:00 (next day) in China, was used as the nighttime data to construct the , because the atmospheric profile at UTC time 18:00 can be extracted from European Centre for Medium-Range Weather Forecasts (ECMWF). The images acquired in the daytime of MIR and TIR channels were used to construct the . Finally, the time-series bi-directional reflectances in MIR channel can be well estimated based on the above analysis. B. Atmospheric Correction Because of the coupling of land surface and atmosphere, the atmospheric effects are required to be corrected for before constructing TISI and separating LST and LSEs. Generally, the atmospheric effects are difficult to be removed due to the lack of synchronous atmospheric profiles, especially for a large study area. Fortunately, the ECMWF provided the latest global atmospheric reanalysis from the period 1979 to present, which has the details on atmospheric vertical distribution and can be used to correct for atmospheric effects with the radiative transfer code. The ECMWF reanalysis atmospheric profile data provide atmospheric profiles of pressure, temperature, relative humidity, and geo-potential at 0.5 latitude/longitude spatial resolutions for 4 daily UTC times: 0:00, 6:00, 12:00, 18:00 h (ECMWF report, 1995). The vertical 21 levels of altitude from sea level to 48 km were selected from ECMWF reanalysis data. The atmospheric parameters, including downwelling radiance, upwelling radiance, transmittance, et al., can be got with the aid of MODTRAN 4.0 using ECMWF atmospheric profiles. Because the ECMWF data are only available at four daily UTC times, the direct atmospheric correction can only be performed at these specific times. For other times when atmospheric status is unavailable, the atmospheric correction in MIR and TIR channel will use the procedures as follows. For the MIR channel, the time-nearest ECMWF data was used directly for the FY-2C data because the atmospheric correction is less sensitive to the change of water vapor content in MIR channel. For the TIR channel, a modified DTC model with six unknown parameters similar to the method proposed by Jiang

C. Retrieval of Emissivities in MIR and TIR Channels Assuming that the land surface behavior is lambertian, the LSEs in MIR and TIR channels can be estimated from the data of MIR and TIR channles based on TISI concept [6]. As we know, land surface does not scatter the solar irradiance in equal

QIAN et al.: LAND SURFACE TEMPERATURE AND EMISSIVITY RETRIEVAL FROM TIME-SERIES MID-INFRARED AND THERMAL INFRARED DATA

quantities in all directions and shows a behavior far from being a lambertian reflector. It is obvious that the assumption of lambertian characteristics might produce a great estimation error of LSEs. In this paper, the emissivity in MIR channel will be retrieved from the hemispherical directional reflectance in MIR channel related to the time-series bi-directional reflectances based on RossThick-LiSparse-R model [10]–[12]. Then, the emissivity in TIR channel can be retrieved from emissivity in MIR channel based on TISI concept. The descriptions in detail have been shown as follows. 1) Hemispherical Directional Reflectance and BRDF Model: For an opaque medium in thermal equilibrium, the hemispherical directional reflectance is related to the bi-directional reflectance according to [12] by:

1555

The LiSparse geometric kernel was derived by Wanner et al. [16]. The original form of the kernel is not reciprocal in and , and then was modified into a reciprocal form under the assumption that the sunlit component simply varies as [17]. Hereafter the reciprocal LiSparse kernel is called LiSparse-R shown as follows.

(9) is the view zenith angle (VZA), is the solar where, zenith angle (SZA), and is relative azimuth angle (RAA), is the bi-directional reflectance of land surface derived from TISI in (7), and is hemispherical directional reflectance of land surface. The RossThick-LiSparse-R model [10]–[12] was a linear kernel-driven BRDF model and based on the theory that land surface reflectance typically consists of three components: the isotropic scattering, the volumetric scattering and the geometric-optical surface scattering [11]. The combination of the Roujean’s volumetric model and the LiSparse-R model is the so-called RossThick-LiSparse-R model, which has been used in the MODerate-resolution Imaging Spectroradiometer (MODIS) at-launch BRDF/Albedo algorithm (MODIS BRDF/Albedo Product: ATBD version 5). This model has been widely validated with its modeling ability, its performance with sparse angular sampling and less sensitivity to noise, and the retrievals are generally reliable [13]. Tang et al. [14] have successfully estimated the hemispherical reflectivity in mid-infrared channel from MODIS data using this model. (10) where is the isotropic scattering term, is the coefficient of the Roujean’s volumetric kernel . is the coefficient of the LiSparse-R geometric kernel . The Roujean’s volume model proposed by Roujean et al. [11] was developed for the correction of satellite data over a wide variety of surface types, and this kernel is a single scattering solution to the classic canopy radiative transfer equation by Ross [15] for plane-parallel dense vegetation canopy with uniform leaf angle distribution, and equal leaf reflectance and transmittance.

(11) Where,

is the phase angle,

(12) where O is the overlap area between the view and solar shadows. h/b and b/r are the dimensionless crown relative height and shape parameters, respectively. In MODIS BRDF/Albedo products, and , i.e., the spherical crows are separated from the ground by half their diameter. These settings will be used in the modeling of the bi-directional reflectivities in FY-2C MIR channel. From (9) and (10), the hemispherical directional reflectance in FY-2C MIR channel can be expressed as follows. (13) where, , the subscription represents or . The parameters and have been expressed as the function of VZA. It is obvious that there is always the double integral problem associated with the SZA and RAA to solve the and in (13). To simplify the calculation, the following approximations proposed by Jiang and Li [18] are used in this paper. (14) (15) are unknown parameters. where, , , , , , , and The Levenberg-Marquardt minimization method was applied to estimate those unknown parameters. The integrals and the fitting results are shown in Table III and Fig. 1, and the numerical integrals and the fitting values using those two functions with the view zenith angle (Table III) are in well agreements and good fitting accuracies, which will be used in the calculation of the directional emissivities in FY-2C MIR channel. It is more convenient in land surface modeling than look-up tables of the kernel integrals.

1556

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013

TABLE III FITTING COEFFICIENTS IN (14) AND (15)

Fig. 1. Fitting results of the volumetric kernel (a) and the geometric kernel (b) of the RossThick-LiSparse-R model.

(10). It is obvious that at least three bi-directional reflectances with different view directions estimated from FY-2C data are required to get the three unknown parameters in RossThickLiSparse-R model. As we know, the FY-2C sensor is onboard a geostationary satellite, and it observes a large area of solar angular measurements at least every one hour at fixed view zenith angles but various sets of relative azimuth angles. However, due to the impact of cloud, wind, among others, more observation data need to be gotten for a good result. Otherwise, there will be a biased estimation of the BRDF for a region covered by the continued cloud or affected by wind. In practice, In order to invert the BRDF model correctly, the minimum number of measurements has been set to five. In addition, if the absolute difference between measured and modeled bi-directional reflectiances is greater than two times the rms error, this measurement will be discarded and the hemispherical directional reflectance will not be got. Then, the hemispherical directional reflectance of FY-2C MIR channel can be estimated from (13)–(15). Theoretically, RossThick-LiSparse-R model with three unknown parameters only needs the FY-2C data of three times. However, the FY-2C data, especially in China area, are usually contaminated by much cloud. In practice, In order to invert the BRDF model correctly, the minimum number of measurements has been set to five. In addition, if the absolute difference between measured and modeled bi-directional reflectance is greater than two times the rms error, this measurement will be discarded and the hemispheric directional reflectance will not be got. Fig. 2 shows the histograms of the rms errors of the bidirectional reflectance for the study area. More than 90% of the rms errors between the measured and modeled bi-directional reflectance are less than 0.01. 2) Retrieval of Emissivities in MIR and TIR Channels: For an opaque medium in thermal equilibrium, directional emissivity is related to hemispherical directional reflectance by the Kirchhoff’s law: (16)

In this work, the RossThick-LiSparse-R model was used to fit the atmospherically corrected time-series bi-directional reflectances at ground level in FY-2C MIR channel to estimate the parameters of the BRDF model according to the

is the hemispherwhere, is directional emissivity, and ical directional reflectance. The LSE in MIR channel can be estimated from the above equation, and the hemispherical directional reflectance

QIAN et al.: LAND SURFACE TEMPERATURE AND EMISSIVITY RETRIEVAL FROM TIME-SERIES MID-INFRARED AND THERMAL INFRARED DATA

1557

estimated by the GSW algorithm [25] is less than 1 K for the sub-ranges with the View Zenith Angle (VZA) less than 30 or for the sub-rangs with VZA less than 60 and the WVC less than 3.5 provided that the LSEs are known. E. Cross-Validation

Fig. 2. Histogram of the rms error between measured and modeled bidirectional reflectance in MIR channel for study area.

can be retrieved from (13). To determine the emissivities in FY-2C TIR channels, the two-channel TISI indices are introduced [6] and assumed LSEs cannot change in the daytime and nighttime. The equation is shown as follows: (17) Where, and denote MIR and TIR channels, respectively. When the directional emissivity in channel is known, the directional emissivity in channel is easily derived from (17). D. Retrieval of Land Surface Temperature Split-window method is used to retrieve the LST based on the differential water vapor absorption in two adjacent infrared channels. This method was firstly proposed by McMillin [19] to estimate sea surface temperature from satellite measurements. The split-window method has been one of the best-validated methods [20]–[24], which has been reported to reach a good retrieval accuracy of LST. A variety of split-window algorithms have been developed and modified to retrieve LST, and successfully applied to MODIS, AVHRR, and MSG/SEVIRI. Generalized Split-window (GSW) algorithm used to estimate LST is a linear function of the brightness temperatures measured by satellite at TOA in atmospheric window channels centered at 10.9 (IR1 channel) and 11.9 (IR2 channel) of FY-2C according to Wan et al. [23].

(18) where the parameters , , , , , and are the coefficients. and are the TOA brightness temperatures of FY-2C data. The coefficients – of the GSW were often estimated from the simulated data. The GSW algorithm by Tang et al. [25] has been used to retrieve the LST from FY-2C data in this study. The algorithm divided LST, average emissivity and column water vapor content (WVC) into several overlap sub-ranges to improve LST retrieval accuracy. The Root Mean Square Error (RMSE) of LST

The LST retrieved from satellite data can be validated using temperature-based, radiance-based, or cross-validation methods in coincidence with a satellite overpass. In principle, LST and LSE derived from satellite measurements can be validated with the ground-truth measurements concurrently with the satellite overpass, which is referred to as the temperature-based method [24]. However, it is a complex and hard task to perform such field measurements, because satellite pixels usually cover several kilometers where the corresponding LST and LSE are quite heterogeneous. Cross-validation involves cross-validating the LST and LSE values retrieved by the method under test with well documented and validated LST and LSE values retrieved from other satellite data. This technique represents an alternative method for LST and LSE validation if there are no atmospheric profiles or ground LST and LSE measurements available or if the T- and R-based validations cannot be conducted. The inter-comparison between LST products estimated from different satellites has been complemented [26]–[28] A cross-validation method proposed by Qian, et al. [28] was successfully used to intercompare the SEVIRI LST product with the LST extracted from MODIS LST product by taking into account many factors, such as the time matching, coordinate matching, view angle, areaweighted aggregation algorithm, the quality control criterion. This method will also be adopted to cross-validate the LST and LSE products retrieved from FY-2C data with those extracted from MODIS MOD11B1 V5 product. To carry out the cross-validation, the LST and LSE derived from the FY-2C data and extracted from the MODIS/Terra MOD11B1 product over the study area were aggregated to the same spatial resolution using area-weighted algorithm [28]. In addition, time matching is another requirement for LST/LSE cross-validation. This cross-validation can be completed with a high accuracy only if the view time is accurately determined. The MOD11B1 product is a composite of several neighbouring orbits with a difference in acquisition time of about 90 minutes according to the tracks of the Terra’s orbit. The view times (local time) stored in MOD11B1 are divided into 12 minute-wide strips. The UTC time as a function of the longitude of each pixel can be calculated. Because the FY-2C sensor provides an image every one hour in this work and the nearest FY-2C acquisition time to MODIS acquisition time is used to evaluate the FY-2C LST product, the maximum difference of the acquisition time between these two products is 12 minutes. III. APPLICATION A. Data The FY-2 series of geo-stationary meteorological satellites are operated by China Meteorological Administration (CMA). Besides four IR channels and one visible channel, the cloud

1558

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013

Fig. 3. Cloud mask product of the FY-2C on December 8–9, 2008. Colorbar denotes the number of cloud free images during the day.

classification data, satellite zenith angle, sun zenith angle, and azimuth have been supplied by FY-2C. In this paper, to estimate the land surface temperature and emissivities of FY-2C, the MIR channel of IR4 (3.8 ), TIR channels of IR1 (10.9 ) and IR2 (11.9 ) were used. Because the DTC model cannot be performed well obviously in the region covered by much of cloud in a diurnal cycle, the satellite data should be first elaborately chosen. Therefore, the data acquired on December 8 and 9, 2008 are selected. There are 24 images for one day, one image per hour. The cloud-contaminated situation is shown in Fig. 3. Because the retrieval accuracies are directly related to the quality of measurements, the calibration is prerequisite. Jiang et al. [29] proposed an inter-calibration of the infrared channels IR1, IR2, and IR4 of the SVISSR/FY-2C against MODIS/Terra and AIRS/Aqua channels. The inter-calibration coefficients were directly adopted as described by Jiang et al. [29] to calibrate the MIR and TIR channels in SVISSR/FY-2C. The MODIS/Terra+Aqua LST products have been validated with in situ measurements in more than 50 clear-sky cases in the temperature range from to 58 and the column water vapor range of 0.4–4 [24], [30], [31]). In this paper, the Terra LST product has been used to validate the retrieved LST of FY-2C. The MODIS/Terra MOD11B1 V5 product, tile-based and girded in the Sinusoidal projection at 6 km spatial resolution, is produced by the day/night LST algorithm from pairs of day-time and night-time observations in seven MODIS TIR bands (http://nsidc.org/data/modis/data_summaries/landgrid_v5.html). This method yields the accuracies of 1 K for LST and 0.01 for LSE in channels 31 and 32 for most of the situations. MOD11B1 product also contains the QC file for LST/LSE. The view zenith angle (VZA) and solar zenith angle (SZA) are stored in MOD11B1 product. These parameters are also needed for the cross-validation. In order to compare the LSE between MODIS and FY-2C data, the spectral variation of emissivity between TIR channels of FY-2C and MODIS should be considered. The spectral response functions of SVISSR/FY-2C and MODIS/Terra instruments are presented in Fig. 4. A total of 110 emissivity spectra are extracted from the MODIS UCSB emissivity library (http://www.icess.ucsb.edu/ modis/EMIS/html/em.html), in which the number of water, ice, snow, vegetation, soil and mineral is 7, 3, 2, 70, 28, respec-

Fig. 4. Spectral response functions of IR1 and IR2 channels in FY-2C and those of channel 31 and 32 in MODIS.

tively, to account for the spectral differences in TIR channels between FY-2C and MODIS. The channel emissivities in SVISSR/FY-2C and MODIS TIR bands were firstly calculated according to MODIS UCSB emissivity library and then the emissivity relationship between FY-2C and MODIS channels was built by using a linear fitting procedure (Fig. 5). The relationships are represented as: (19) (20) and are the emissivities of FY-2C channel 1 where and 2, respectively. and are the emissivities of MODIS channel 31 and 32, respectively. The coefficients of determination are greater than 0.98 and the Root Mean Square Errors (RMSEs) are less than 0.003. B. Study Area The study area of the whole China was selected for the retrieval of LST/LSE from FY-2C data. China’s landscape is vast and diverse covering approximately 9.6 million square kilometers, the world’s third largest country in total area behind Russia and Canada, and very similar to the United States, with forest steppes and the Gobi and Taklamakan deserts occupying the arid

QIAN et al.: LAND SURFACE TEMPERATURE AND EMISSIVITY RETRIEVAL FROM TIME-SERIES MID-INFRARED AND THERMAL INFRARED DATA

Fig. 5. Relationships between the emissivities in the two split-window channels of FY-2C and MODIS using the MODIS UCSB emissivity library.

north and northwest near Mongolia and Central Asia, and subtropical forests being prevalent in the wetter south near Southeast Asia. The terrain of western China is rugged and elevated, with the Himalaya, Karakoram, Pamir and Tian Shan mountain ranges separating China from South and Central Asia. IV. RESULTS AND ANALYSIS According to the method above, the results of LSEs and LST estimated from FY-2C MIR and TIR channels have been shown in Fig. 6 and Fig. 7. Fig. 6 shows the LSEs estimated by the combination of the MIR and TIR channel with day-night FY-2C data. The top and bottom columns denote the LSEs of FY-2C on December 8, 2008, and on December 9, 2008, respectively. There are twenty-four LST imageries for one day retrieved from the SVISSR/FY-2C time-series data, considering the length of paper, only the LSTs at UTC times 06:00 on December 8 and 9, 2008 are shown in Fig. 7. From Fig. 6 and Fig. 7, it can be found that there are large areas with no LSE/LST values. The probable reason might be that the retrieved directional reflectance is based on the time series satel-

1559

lite data in a day, and the covered or contaminated by clouds during a long time period may prevent the retrievals. If the data was contaminated by cloud, the accuracy of bi-directional reflectance in MIR channel could be affected greatly, and then the LSEs might not be fitted well. Table IV gives the bias and the RMSE between FY-2C LSE/LST and MODIS/Terra MOD11B1 Product LST over the study area. Compared with the MOD11B1 LSE, the bias of emissivity difference is low, and the FY-2C LSEs at 11 and 12 underestimate the emissivities by about 0.003 and 0.004. However, the RMSEs of LSEs are very large, those are 0.045 and 0.048 over the study area (Table IV, Fig. 8(a) and (b)). The FY-2C LST over the study area is large within 1.65 K compared with the LST extracted from the MOD11B1 during the night-time. However, the LST difference may reach up to 2.87 K during the day-time. The bias and RMSE of the temperature differences at daytime are larger than those at night-time over the study area (Table IV, Fig. 8(c) and (d). However, the relatively large RMSE of LST differences during the daytime are somewhat concerning. The comparison results of LST at night-time are better than those during the daytime because the surface temperature at night-time is more homogeneous. To validate the stability of the proposed method, the retrieved results of the two consecutive days have been used for comparison. Table V gives the bias and the RMSE in FY-2C LSE difference of the two consecutive days over the study area, on December 8 and 9, 2008. The retrieved results have been shown that the bias of FY-2C LSE differences at 11 and 12 are approximately 0.002 and 0.003, respectively, and the RMSEs of FY-2C LSE differences are approximately 0.008 and 0.014. It can be found that the bias of the retrieved FY-2C LSE difference of the two consecutive days has a slightly discrepancy. It should be pointed out that the LST estimated from the FY-2C satellite data has not been validated with in situ measurements since there are no in situ measurements available. Taking into account the fact that, due to the extreme difficulty or impossibility to get the LST and LSE at ground level representative at 5 km 5 km, only the cross-validation method was applied to validate the LST derived from FY-2C data by using the LST product provided by MODIS data. From Tables IV and 5, the results show that cross-validation method can be well performed to evaluate the accuracies of the retrieval LST and LSE and the retrieval method can be well used to describe in stably the LSE. The retrieval accuracies of LST & LSE depend on many factors, such as view angle, BRDF fitting accuracy, modeling accuracy, et al. Meanwhile, the cross-validation algorithm depends on the consistency of view time. Because the FY-2C sensor provides an image every one hour in this work and the nearest FY-2C acquisition time to MODIS acquisition time is used to evaluate the FY-2C LST product, the maximum difference of the acquisition time between these two products is 12 minutes, which might be one reason that the larger discrepancy might be produced. Meanwhile, the VZAs in China area are almost larger than 30 , the accuracy of LST retrieval method will decrease in large VZA. Another reason might be that the fitted BRDFs can affect the accuracy of LSE retrieved, which also can influence the accuracy of LST retrieval.

1560

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013

Fig. 6. LSEs in MIR, IR1 and IR2 retrieved from FY-2C data on December 8–9, 2008.

Fig. 7. LSTs retrieved from FY-2C data on December 8–9, 2008.

V. CONCLUSIONS In this paper, we proposed a method to retrieve the LSEs and LST from combined MIR and TIR data of FY-2C. As we know from the radiative transfer equations, the main difficulties are necessary to correct for atmospheric perturbations and variable surface situations. The spatial interpolation scheme for ECMWF atmospheric profiles and the DTC model were adopted for the atmospheric correction. The bi-reflectance of MIR channel using the TISI concept was retrieved combining

with the mid-infrared and thermal infrared data of FY-2C. Consequently, the hemispherical directional reflectance and land surface emissivity in FY-2C MIR channel was also retrieved using the RossThick-LiSparse-R model. Based on TISI model, the LSE in TIR channel was estimated. Finally, the LST can be retrieved from the two TIR channels of FY-2C and the cross-validation method was used to validate the retrieval accuracies of LSE and LST with MODIS MOD11B1 product. The results have been demonstrated that the proposed method has been successfully used to estimate the LSE and LST of

QIAN et al.: LAND SURFACE TEMPERATURE AND EMISSIVITY RETRIEVAL FROM TIME-SERIES MID-INFRARED AND THERMAL INFRARED DATA

1561

Fig. 8. Histogram of cross-validation of LSE and LST between FY-2C and MOD11B1 products. TABLE IV CROSS-VALIDATION OF LSE AND LST BETWEEN FY-2C AND MOD11B1 PRODUCTS

in retrieved LST. Compared with the MOD11B1 products, the FY-2C LSTs are 2.87 and 1.65 K larger during the daytime and night-time, respectively. The comparison results of LST during the night-time are better than those during the daytime because of the relative homogeneity of the surface temperature at night-time. Although the retrieved FY-2C LST error is within 1 K with simulated data, the large FY-2C view angle over the study area also impacts the retrieval accuracy of the LST from satellite data. Another reason is that the retrieval accuracy of the FY-2C LSE will directly degrade the retrieval accuracy of the FY-2C LST. ACKNOWLEDGMENT

TABLE V THE RETRIEVED FY-2C LSES DIFFERENCE OF THE TWO CONSECUTIVE DAYS, ON DECEMBER 8 AND 9, 2008

The authors thank anonymous referees for their comments and suggestions that have significantly improved the article. The authors would also like to sincerely thank the University of California, Santa Barbara, for the MODIS UCSB spectral library, the European Centre for Medium-Range Weather Forecasts for the ECMWF atmospheric profile data, and the National Satellite Meteorological Center for the FY-2C data. REFERENCES

SVISSR/FY-2C data. Compared with the LSEs extracted from MOD11B1 product, the FY-2C LSEs at 11 and 12 underestimate the emissivities by about 0.003 and 0.004 over the China. Although the bias between FY-2C and MOD11B1 LSE is low, the LSE RMSEs in 11 and 12 are very large, that is 0.045 and 0.048. The large LSE RMSEs might cause the large errors

[1] J. A. Sobrino, N. Raissouni, and Z.-L. A. Li, “Comparative study of land surface emissivity retrieval from NOAA data,” Remote Sens. Environ., vol. 75, no. 2, pp. 256–266, 2001. [2] J. A. Sobrino and J. C. Jiménez-Muñoz, “Land surface temperature retrieval from thermal infrared data: An assessment in the context of the Surface Processes and Ecosystem Changes Through Response Analysis (SPECTRA) mission,” J. Geophys. Res., 2005, 10.1029/2004JD005588. [3] Z.-L. Li, F. Petitcolin, and R. H. Zhang, “A physically based algorithm for land surface emissivity retrieval from combined mid-infrared and thermal infrared data,” Science in China (Series E), vol. 43, no. 1, pp. 23–33, 2000.

1562

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013

[4] G. M. Jiang, Z.-L. Li, and F. Nerry, “Land surface emissivity retrieval from combined mid-infrared and thermal infrared data of MSG-SEVIRI,” Remote Sens. Environ., vol. 105, no. 4, pp. 326–340, 2006. [5] Z. Xiaoyu, “Estimation of Thermal Inertia and Ground Heat Flux Using Stationary Meteorological Satellite Data,” Ph.D Dissertation, Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, , 2008. [6] F. Becker and Z.-L. Li, “Temperature independent spectral indices in thermal infrared bands,” Remote Sens. Environ., vol. 32, no. 1, pp. 17–33, 1990a. [7] S. G. Schädlich, F.-M. Göttsche, and F.-S. Olesen, “Influence of land surface parameters and atmosphere on METEOSAT brightness temperatures and interpolation of atmospheric correction,” Remote Sens. Environ., vol. 75, no. 1, pp. 39–46, 2001. [8] F. Nerry, F. Petitcolin, and M. P. Stoll, “Bidirectional reflectivity in AVHRR channel 3: Application to a region in North Africa,” Remote Sens. Environ., vol. 66, no. 3, pp. 298–316, 1998. [9] F.-M. Göttsche and F.-S. Olesen, “Modelling of diurnal cycles of brightness temperature extracted from METEOSAT data,” Remote Sens. Environ., vol. 76, no. 3, pp. 337–348, 2001. [10] X. Li and A. H. Strahler, “Geometric-optical bidirectional reflectance modeling of the discrete crown vegetation canopy: Effect of crown shape and mutual shadowing,” IEEE Trans. Geosci. Remote Sens., vol. 33, no. 2, pp. 276–292, 1992. [11] J. L. Roujean, M. Leroy, and P. Y. Deschamps, “A bidirectional reflectance model of the Earth’s surface for the correction of remote sensing data,” J. Geophys. Res., vol. 97, no. 20, pp. 455–468, 1992. [12] A. H. Strahler and J.-P. Muller, MODIS Science Team Members, MODIS BRDF/Albedo Product: Algorithm Theoretical Basis Document—Version 5.0 1999. [13] J. L. Privette, T. F. Eck, and D. W. Deering, “Estimating spectral albedo and nadir reflectance through inversion of simple BRDF models with AVHRR/MODIS-like data,” J. Geophys. Res., vol. 102, no. D24, pp. 29529–29542, 1997. [14] B.-H. Tang, Z.-L. Li, and Y. Bi, “Estimation of land surface directional from MODIS data,” emissivity in mid-infrared channel around 4.0 Opt. Express, vol. 17, no. 5, pp. 3173–3182, 2009. [15] J. K. Ross, The Radiation Regime and Architecture of Plant Stands, W. Junk, Ed. , Netherland: The Hague, 1981. [16] W. Wanner, X. Li, and A. H. Strahler, “On the derivation of kernels for kernel-driven modelsof bidirectional reflectance,” J. Geophys. Res., vol. 100, no. D10, pp. 21077–21089, 1985. [17] W. Lucht and P. Louis, “Theoretical noise sensitivity of BRDF and albedo retrieval from EOS-MODIS and MISR sensors with respect to angular sampling,” Int. J. Remote Sens., vol. 21, no. 1, pp. 81–98, 2000. [18] G. M. Jiang and Z.-L. Li, “Intercomparison of two BRDF models in the estimation of the directional emissivity in MIR channel from MSG1SEVIRI data,” Opt. Express, vol. 16, no. 23, pp. 19320–19321, 2008. [19] L. M. McMillin, “Estimation of sea surface temperature from two infrared window measurements with different absorption,” J. Geophys. Res., vol. 80, pp. 5113–5117, 1975. [20] F. Becker and Z.-L. Li, “Toward a local split window method over land surface,” Int. J. Remote Sens., vol. 11, no. 13, pp. 369–393, 1990b. [21] J. Sobrino, V. Caselles, and C. Coll, “Theoretical split window algorithms for determining the actual surface temperature,” II Nuovo Cim., vol. 16, pp. 219–236, 1993. [22] V. Caselles, C. Coll, and E. Valor, “Land surface emissivity and temperature determination in the whole HAPEX-Sahel area from AVHRR data,” Int. J. Remote Sens., vol. 18, no. 5, pp. 1009–1027, 1997. [23] Z. Wan and J. Dozier, “A generalized split-window algorithm for retrieving land-surface temperature from space,” IEEE Trans. Geosci. Remote Sens., vol. 34, no. 4, pp. 892–965, 1996. [24] Z. Wan and Z.-L. Li, “Radiance-based validation for the V5 MODIS land-surface temperature product,” Int. J. Remote Sens., vol. 29, no. 17–18, pp. 5373–5395, 2008. [25] B.-H. Tang, Y. Y. Bi, Z.-L. Li, and J. Xia, “Generalized split-window algorithm for estimate of land surface temperature from Chinese geostationary FengYun meteorological satellite (Fy-2C) data,” Sensors, vol. 8, no. 2, pp. 933–951, 2008. [26] E. J. Noyes, G. K. Corlett, X. Kong, J. J. Remedios, and D. T. Llewellyn-Jones, “The AATSR Land Surface Temperature (LST) product: Comparison with LSTs from SEVIRI,” in SAF 2nd User Training Workshop, March 8–10, 2006. Lisbon: Instituto de Meteorologia, 2006. [27] I. F. Trigo, I. F. Monteiro, F. Olesen, and E. Kabsch, “An assessment of remotely sensed land surface temperature,” J. Geophys. Res., vol. 113, pp. 1–12, 2008.

[28] Y.-G. Qian, Z.-L. Li, and F. Nerry, “Evaluation of land surface temperature and emissivities retrieved from MSG-SEVIRI data with MODIS land surface temperature and emissivity products,” Int. J. Remote Sens., vol. 34, no. 9–10, pp. 3140–3152, 2013. [29] G.-M. Jiang, H. Yan, and L.-L. Ma, “Intercalibration of SVISSR/FY-2C infrared channels against MODIS/Terra and AIRS/Aqua channels,” IEEE Trans. Geosci. Remote Sens., vol. 47, no. 5, pp. 1548–1558, 2009. [30] Z. Wan, Y. L. Zhang, Q. Zhang, and Z.-L. Li, “Quality assessment and validation of the MODIS global land surface temperature,” Int. J. Remote Sens., vol. 25, no. 1, pp. 261–274, 2004. [31] C. Coll, V. Caselles, J. M. Galve, E. Valor, R. Niclòs, J. M. Sánchez, and R. Rivas, “Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS data,” Remote Sens. Environ., vol. 97, no. 13, pp. 288–300, 2005.

Yonggang Qian received the B.S. degree in applied mathematics from Yantai University, Yantai, China, in 2003, the M.S. degree in geographical information system from Taiyuan University of Technology, Taiyuan, China, in 2006, the Ph.D. degree in cartography and geographical information system from Beijing Normal University, Beijing, China, in 2009, and the Post Doctor in cartography and geographical information system from the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, in 2009–2011. He is currently an Associate Professor with the Academy of Opto-Electronics, Chinese Academy of Sciences. His research mainly includes the retrieval and validation of surface temperature and emissivity from remotely sensed data.

Shi Qiu received the B.E. degree in Engineering from Huazhong University of Science and Technology, Wuhan, China, in 2007; the Ph.D. degree in Physical from University of Strasbourg, Strasbourg, France, from 2008 to now (expected June 2013). Her research interests mainly include the retrieval and validation of surface temperature and emissivity from remotely sensed data.

Ning Wang received the B.S. and M.S. in Geophysical Science and Cartography and Geographic Information System from Beijing Normal University, China, in 2004 and 2007, respectively, and the Ph.D. degree in cartography and geographical information system from Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China, in 2011. He is currently a Research Assistant with the Academy of Opto-Electronics, Chinese Academy of Sciences. His research interests include thermal infrared remote sensing and hyperspectral remote sensing.

Xiangsheng Kong received the B.E. degree in coal geology and exploration from Shanxi Mining College, Taiyuan, China, in 1995, the M.E. degree in coal, oil gas geology and exploration from Taiyuan University of Technology, Taiyuan, China, in 1998, the Ph.D. degree in geodetection and information technology from Chengdu University of Technology, Chengdu, China, in 2007. He is currently an Associate Professor with the College of Geography and Planning, Ludong University. His research area mainly focuses on the identification of high temperature target from remotely sensed data.

QIAN et al.: LAND SURFACE TEMPERATURE AND EMISSIVITY RETRIEVAL FROM TIME-SERIES MID-INFRARED AND THERMAL INFRARED DATA

Hua Wu received the B.S. degree in photogrammetric engineering and remote sensing from Wuhan University, Wuhan, China, in 2003, the M.S. degree in cartography and geographical information system from Beijing Normal University, Beijing, China, in 2006, and the Ph.D. degree in cartography and geographical information system from the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, in 2010. He is currently a Research Assistant with the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences. His research mainly includes the retrieval and validation of surface temperature and emissivities, and scaling of remotely sensed products.

1563

Lingling Ma received the M.S. and Ph.D. degrees in geography and geographical information system from the Chinese Academy of Sciences, Beijing, China, in 2005 and 2008, respectively. She is currently an Associate Professor with the Academy of Opto-Electronics, Chinese Academy of Sciences. Her research interests include calibration and validation techniques of remote sensing data and products, and inflight performance assessment of optical sensors.