AUTOMATING THE ESTIMATION OF UNOBSERVED DATA FROM WEATHER STATIONS IN ARGENTINA

Abstract

In Argentina, the National Meteorological Service elaborates daily reports of atmospheric data captured at weather stations throughout the country. The Ministry of Agriculture uses that data to do risk analysis, a task in which it is preferred to estimate unobserved or corrupted values, rather than just ignore them. The selection and application of interpolation methods used to be done manually. In this work we present the decisions made to fully automate the process. Under the assumption that no interpolation method is universal, our method automatically selects the most appropriate technique in each case. The application of the methods is also done without the need of human intervention or parameter tuning. The proposed solution was tested using data acquired within a four-year period. The error rates computed were tagged as highly satisfactory by the meteorologists. Moreover, the idea of selecting the best method each time outperforms the search for a universal method for each indicator, validating our approach.