Spatial Interpolation of meteorology monitoring data for western China using back-propagation artificial neural networks

Abstract

Spatial interpolation algorithms are vital to environmental monitoring systems, especially for the real-time monitoring systems of critical variables in converting the point measurements to spatial continuous surfaces. This paper describes the spatial interpolation of meteorological observations (air temperature as an example) using a feed-forward back-propagation neural network based on the environment-affecting factors. These model independent estimators were (1) meteorological stations' longitude, latitude, altitude; (2) Normalized Difference Vegetation Index; (3) slope and aspect. This is a first to consider all the factors for are temperature spatial interpolation when interpolating using a neural network. Especially the study area covers large region of complex terrain, which includes only 241 national meteorological stations over almost half-total area of China. However, the simulated results show that the model could provide reliable spatial estimations of monthly mean air temperature. Goodness of fit of model was very high (R>0.95) and efficient.