ANNOUNCEMENTS
High resolution thermal imagery is continuously required for various applications like estimation of surface energy budget, assessment of evapotranspiration and drought prediction, exploring urban heat island effect, and monitoring volcanic eruptive activity. But due to the physical and technological constrains, acquisition of moderate to high resolution thermal data is a very challenging task. At present availability of finer spatial resolution thermal data (<200 m) is a limitation and its temporal resolution is low. While coarser spatial resolution (1000 m) and high temporal resolution (~1 day) thermal data is freely available, but not suitable for many applications due to their coarse spatial information content. A coarse resolution thermal data receive mixed signals from various land cover / material located within a pixel, called as thermal mixture effect. Hence, it is important to build a link between spatial and temporal resolution and reduce thermal mixture effect by developing a technique to monitor the daily basis long-term environmental phenomena. In this context downscaling of spatial resolution of thermal imagery is essential.
The present study is an attempt to downscale coarse spatial resolution thermal image to finer spatial resolution using relationship between vegetation indices (NDVI) and land surface temperature (LST) over a heterogeneous landscape of India. The objective of the present research is multi-fold with development of models, validation in heterogeneous landscapes and checking applicability in environmental monitoring purposes.
The first objective aims to develop a robust LST downscaling algorithm. Eleven models are applied to downscale the LST of Landsat and MODIS data and results are compared. All models are first evaluated on Landsat thermal data aggregated to 960 m resolution and downscaled to 480 m and 240 m resolution. The models are then applied on MODIS thermal data to downscale from 1000 m to 250 m resolution with less than 1° C accuracy. The Least median square regression (LMSDS) and Regression-Kriging (RKDS) models are found to be most suitable in this landscape. The LMSDS model is not sensitive to outliers in the heterogeneous landscape and provides higher accuracy compare to other models. The performance of LST downscaling technique may not be consistent over seasons, LULC classes and different soil moisture conditions. It is a physically based downscaling model and could be dependent on various environmental factors. Therefore spatio-temporal analysis based on the above mentioned factors are performed. It is observed that the downscaling is suitable for agricultural and vegetated landscapes and is not applicable to water body, river bed and sandy open areas.
The second objective aims to apply the downscaled LST on mapping of agricultural drought, soil moisture status and urban applications such as detection of urban centers / built up areas and surface urban heat island (SUHI). The downscaling of MODIS LST is feasible up to a spatial resolution of around 250 m. Since the MODIS thermal data of 1000 m and NDVI of 250 m are readily available on a daily basis, a downscaling of thermal dataset at 250 m spatial resolution would certainly be a step forward for agricultural, climate and environment related studies.
Shodhganga Link
