Recovering Historical Climate Records Using Parallel Artificial Neural Networks in GPU

Abstract

This article presents a parallel implementation of Artificial Neural Networks over Graphic Processing Units, and its application for recovering historical climate records from the Digi-Clima project. Several strategies are introduced to handle large volumes of historical pluviometer records, and the parallel deployment is described. The experimental evaluation demonstrates that the proposed approach is useful for recovering the climate information, achieving classification rates up to 76 % for a set of real images from the Digi-Clima project. The parallel algorithm allows reducing the execution times, with an acceleration factor of up to 2.15 ×.