Validation of Land surface Temperature (LST) from Landsat-5 and MODIS Images (Case study: Wheat fields of Marvdasht Plain)

Land surface temperature (LST) is a key parameter in estimating energy balance that has determinate role in climate change studies. Various scientists have studied monitoring of LST in recent decades. The land surface temperature, which is measured by means of thermometers for certain points, for large scale basin is not cost effective. Using of satellite images for estimating LST make the estimates easier and more economical than ground measurement. In this study, MODIS land surface temperature (LST) was evaluated. In addition, due to use of correction factors which may not always be available for Iran, land surface temperature estimated by Landsat 5 image,which its spatial resolution is much higher than MODIS, was also evaluated.
For this study, two groups of data were used: satellite data and in-situ data. Ground measurements were collected from 261 points of a wheat farm in Marvdasht plain located in Fars province. Temperature was measured in four height of wheat including: canopy cover, middle, 10 centimeter from floor and soil surface. After statistical tests, acceptable data were selected for the comparison. In this study, twenty eight satellite images were implemented; including 26 MODIS images (MOD02 & MOD11 product) and 2 level-1G Landsat 5 images. Land surface temperature was estimated from thermal band’s of Landsat 5 images by applying the necessary corrections. After providing land surface temperature (LST) maps, land surface temperature was extracted from LST map (Landsat5 & MODIS) based on the measurement points. Afterward, the equation between the observed data and estimated surface temperatures from Landsat 5 (MODIS images) were obtained. Relationship between estimated and in-situ data was analyzed on four different heights of the wheat. Land surface temperatures were also estimated by three different split–window algorithms from Becker and Li (1990), Price (1984) and Ultivertal (1994) and the coefficients were calibrated. Finally, Fisher test was used to determine significant differences between the observed and the estimated data.
It was found that the estimated temperature by satellite has the best correlations with the plant canopy temperature. Estimated data were evaluated against the in-situ data. Results showed that Landsat and MODIS images overestimated the LST by RMSE of 4.4 oC and 7.1 oC respectively. Error of Estimating LST with split–window algorithms was within the range of 3.5–3.7 degree centigrade. Among the three studied algorithms, Becker and Li (1990) approach showed the best performance (the least error). The significant differences between in-situ data and the satellite estimates were examined by Fisher Test. No significant differences were observed in any of the pairs of data.
For meso-scale and large-scale studies, using satellite images is efficient and economic than the point surface measurements. The choice of satellite images (Landsat or MODIS) is depend on the accuracy which is expected from the study.