Atmospheric water vapor retrieval from Landsat 8 thermal infrared images

Abstract

Atmospheric water vapor (wv) is required for the accurate retrieval of the land surface temperature from remote sensing data and other applications. This work aims to estimate wv from Landsat 8 Thermal InfraRed Sensor (TIRS) images using a new modified split‐window covariance‐variance ratio (MSWCVR) method on the basis of the brightness temperatures of two thermal infrared bands. Results show that the MSWCVR method can theoretically retrieve wv with an accuracy better than 0.3 g/cm2 for dry atmosphere (wv <2 g/cm2) conditions and better than 0.5 g/cm2 for wet atmosphere conditions. The method was applied at different locations with dry and moist atmospheres and was validated at 42 ground sites using AERONET (Aerosol Robotic Network) ground‐measured data and MODIS (Moderate Resolution Imaging Spectroradiometer) products. The results show that the retrieved wv from the TIRS data is highly correlated with the wv of AERONET and MODIS but is generally larger. This difference was probably attributed to the uncertainty of radiometric calibration and stray light coming outside from field of view of TIRS instrument in the current images. Consequently, the data quality and radiometric calibration of the TIRS data should be improved in the future.

A Practical Single‐Channel Algorithm for Land Surface Temperature Retrieval: Application to Landsat Series Data

“…To perform the GSC and PSC algorithms, L sen , T sen , ε , w , and T n (optional) should be obtained. L sen (W·m −2 ·sr −1 ·μm −1 ) and T sen (K) can be calculated from TIR data after radiative calibration; ε can be estimated using normalized difference vegetation index (NDVI) thresholds method (Sobrino et al, ); w (g/cm 2 ) can be obtained from Landsat 8 TIRS data using split‐window covariance‐variance ratio method (Ren et al, ; Wang et al, ), satellite AWV products or meteorological products (e.g., MODIS total column precipitable water vapor [MOD05] and NCEP); T n (°C) can be estimated by means of Geographic Information System and Remote Sensing (Cristóbal et al, ; Wang et al, ).…”

Recovering Land Surface Temperature Under Cloudy Skies Considering the Solar‐Cloud‐Satellite Geometry: Application to MODIS and Landsat‐8 Data

“…For this point, after many tests, two thresholds were selected to simply identify the cloud and shadow based on the radiance of bands 10 and 1, respectively. Then, the clear‐sky LSTs were derived by using a typical split window method (Du et al, ; Ren et al, ). The initial height for each cloud pixel was approximated by using equation based on the lapse rate (Cosgrove et al, ): where Cld _ Hi is the cloud top height for pixel i ; LST is the average temperature of the clear‐sky pixels (Du et al, ; Ren et al, ) around the cloud in the Landsat‐8 image; and is the cloud top brightness temperature for pixel i , which was determined by using the radiance of band 10 of Landsat‐8 (equation ).…”

“…Then, the clear‐sky LSTs were derived by using a typical split window method (Du et al, ; Ren et al, ). The initial height for each cloud pixel was approximated by using equation based on the lapse rate (Cosgrove et al, ): where Cld _ Hi is the cloud top height for pixel i ; LST is the average temperature of the clear‐sky pixels (Du et al, ; Ren et al, ) around the cloud in the Landsat‐8 image; and is the cloud top brightness temperature for pixel i , which was determined by using the radiance of band 10 of Landsat‐8 (equation ). where BT is the TOA brightness temperature; Rad is the radiance of the thermal band (i.e., band 10); and K 1 and K 2 are the constants with values of 774.89 and 1,321.08 for band 10, respectively.…”

Improved Assessment of Atmospheric Water Vapor Content in the Himalayan Regions Around the Kullu Valley in India Using Landsat‐8 Data

“…In equation , the critical values of NDVI for vegetation and soil are represented by NDVI max and NDVI min , respectively, and P v is the vegetation fraction. Ren et al () introduced the modified SWCVR for the estimation of WVC incorporating a second‐order polynomial expression as shown in equation , where the coefficients ( a 1 , a 2 , and a 3 ) are determined using regression with a priori reference such as the Thermodynamic Initial Guess Retrieval (TIGR) atmospheric profiles (Chedin et al, ), meteorological observations, and or other spaceborne products of MODIS or other sensors (Wang et al, ). …”

“…However, these techniques provide scattered point observations of WVC and are thus unable to provide much insight in the regional or global distribution of WVC (Li et al, 2003(Li et al, , 2013. In literature, various methods have been proposed for estimating WVC categorized with respect to the remote sensing data, such as near infrared (Gao & Kaufman, 2003), thermal infrared (TIR; Li et al, 2003;Ren et al, 2015), hyperspectral (Barducci et al, 2004), and microwave data (Padmanabhan et al, 2009). Although the methods utilizing near-infrared channels are widely popular for estimation of WVC such as in Moderate Resolution Imaging Spectroradiometer (MODIS) and Sentinel-3 products, they are constrained by the availability of water vapor absorption channels and are often not applicable to certain multispectral sensors such as in Operating Linear Imager (OLI) sensor covers the thermal spectrum using two TIR channels at 11 and 12 μm, which are used for estimation of WVC using split window techniques.…”