Abstract
<jats:p>The ECOSTRESS mission provides high resolution thermal infrared observations that support a wide range of applications ranging from evapotranspiration monitoring and drought assessment to Land Surface Temperature (LST) and emissivity retrieval. Accurate estimation of Precipitable Water Vapour (PWV) is critical for these applications because of strong influence of atmospheric water vapour on thermal infrared radiative transfer that influences emissivity retrieval. The current ECOSTRESS processing chain estimates PWV from GEOS5-FP numerical weather prediction data. In this study, we investigate an alternative approach that retrieves PWV directly from ECOSTRESS thermal brightness temperatures using symbolic Regression (PYSR). Using more than 269,000 spatio-temporally matched ECOSTRESS-GNSS observations, we derived a unified all-season analytical formula capable of estimating PWV without relying on external atmospheric profiles or ancillary emissivity. Instead, the proposed model relies only on readily available and temporally stable ancillary variables, namely digital elevation model (DEM) and Normalized Difference Vegetation Index (NDVI) data. The resulting analytical formulation (All-season PySR) derived using PySR symbolic retrieval achieved a Root Mean Square Error (RMSE) of 7.22 mm and an R2 of 0.624 when evaluated against GNSS-derived precipitable water vapor observations. In order to improve the results, season and regime specific PySR formulas were first developed to provide interpretable PWV estimates for various conditions. These formulas form the initial retrieval component of the Climate-Adaptive Ensemble formula. The Climate Adaptive Ensemble (CAE) combines PySR formulas, ECOSTRESS inputs and historical ERA5 water-vapour profiles to perform a global analytical ridge regression. The resulting output achieved an RMSE of 5.35mm and R2 of 0.781 on test GNSS observations. The CAE formula was further validated on external radiosonde dataset and independent GNSS observations to evaluate its robustness. The methodology proposed here will be extremely useful for future satellite missions using thermal sensors such as TRISHNA (Thermal Infra-Red Imaging Satellite for High-resolution Natural resource Assessment) and LSTM (Land Surface Temperature Radiometer) as the methodology could help in retrieving PWV instantaneously for atmospheric correction instead of depending on external products.</jats:p>