Image-to-Image Training for Spatially Seamless Air Temperature Estimation With Satellite Images and Station Data

Air temperature at approximately 2 m above the ground ( $T_{a}$ ) is one of most important environmental and biophysical parameters to study various earth surface processes. measured from meteorological stations inadequate its spatio-temporal patterns since are unevenly sparsely distributed. Satellite-derived land (LST) provides global coverage, generally utilized estimate due close relationship between LST . However, products sensitive cloud contamination, resulting in missing values leading estimated being spatially incomplete. To solve data problem, we propose a deep learning method seamless that contains values. Experimental results on 5-year mainland China illustrate image-to-image training strategy alleviates problem fills gaps implicitly. Plus, strong linear relationships observed daily mean notation="LaTeX">$T_{mean}$ ), minimum notation="LaTeX">$T_{min}$ maximum notation="LaTeX">$T_{max}$ make estimation , simultaneously possible. For China, proposed achieves with notation="LaTeX">$R^{2}$ 0.962, 0.953, 0.944, absolute error (MAE) 1.793, 2.143, 2.125 notation="LaTeX">$^\circ$ C, root square (RMSE) 2.376, 2.808, 2.823 C for respectively. Our new paradigm estimating ground-level satellite products. Code more available https://github.com/cvvsu/LSTa