E. V. Kadantsev, A. I. Chavro
Reconstruction of small-scale fields of climatic variables (like temperature, precipitation, pressure, etc.) on large-scale values predicted by General Circulation Models (GCMs), is an important link for understanding the effects of global climate change on smaller regional scales.
In this paper, we used a neural network based approach to model the relationship between regional and global fields. The paper discusses how the architecture of neural network was built, and the results of its application for solving the inverse task of downscaling average daily values of surface temperature on meteorological stations of Moscow region on the basis of large-scale fields of temperature, predicted using a global short-term forecasting model.
Preliminary analysis of results for Moscow region has shown that the nonlinear neural network makes it possible to significantly improve the accuracy of the reconstruction, as compared with the linear model used by us earlier.