Estimation of air surface temperature using MODIS land surface temperature over data-scarce Banas River basin, Western India

Arid and semi-arid areas characterised by low precipitation and high evaporation rates are highly vulnerable to alterations in precipitation regimes, leading to water deficiency and an increase in dependence on groundwater resources. Flash floods have become more frequent in several semi-arid regions due to changing climatic conditions. Thus, an efficient water management system is needed for these regions to manage flash floods and support groundwater recharge. A coupled surface water-groundwater model is an advanced tool for simulating large-scale hydrologic processes and quantifying factors influencing floods and drought. To accommodate the high variability and heterogeneous spatial distribution of surface and groundwater resources, distributed modelling tools are essential. However, scarce monitoring networks may lead to the unavailability of spatio-temporal input data and limit the applicability of these models. Advances in remote sensing (RS) techniques for monitoring hydrological parameters like precipitation, soil moisture, evapotranspiration, and groundwater depth can mitigate this problem.

This study analyses the remote sensing product MOD11A1.006 of Moderate Resolution Imaging Spectroradiometer (MODIS), which provides daily day and night land surface temperature (LST) at a spatial resolution of 1000 m, facilitating the analysis of surface water-groundwater interactions through distributed hydrological modelling in the semi-arid Banas River basin (~6800 km²). Remotely sensed LST data allowed air surface temperature (T_a), which is crucial for estimating reference evapotranspiration, to be retrieved. While T_a at weather stations 2 m above the ground are more accurate, those data have limited spatial coverage. The Banas River basin contains five weather stations located primarily in the central region. To improve the spatial distribution of reference evapotranspiration, which is a significant input of the hydrological models, a linear regression model using T_a observed at the weather stations of Banas basin, along with LST, elevation, Normalized Difference Vegetation Index (NDVI), latitude, and longitude of the pixels coinciding with the location of weather stations was developed to estimate the air surface temperature for the whole basin.

A multiple linear regression model was built by stepwise linear regression (SLR) method using the OLSRR package of R. Calibration using day LST, latitude and longitude provided the best estimate of maximum T_a, with an adjusted R² value of 0.60, Pearson correlation coefficient (r) of 0.72, and Root Mean Square Error (RMSE) of 3.2^o C. While calibration using night LST and elevation data provided the best estimate of minimum T_a with an adjusted R² value of 0.81, r of 0.84 and RMSE of 3.02^o C. The daily LST data and daily T_a data have shown a good agreement. This research improves the understanding of the spatial distribution of daily day and night air temperature in the Banas River basin. It opens a new methodological perspective for groundwater and surface water management through hydrological modelling with a spatial resolution greater than that of the existing monitoring networks.