Exploring Deep Learning Techniques for Seasonal Prediction of Autumn Precipitation in South America
Year: 2025
Type: conference-paper
Abstract: Seasonal precipitation forecasting is essencial for climate planning and disaster risk reduction. This study used Convolutional Neural Network 1D (CNN 1D), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Graph Convolutional Long Short-Term Memory (GConvLSTM) to forecast autumn 2023 precipitation. CNN 1D had the best performance (MSE: 1.8827, Correlation Coefficient: 0.9137), excelling in accuracy but losing spatial details. GConvLSTM captured spatial patterns better, while LSTM and GRU underperformed. Despite CNN 1D’s superiority, integrating spatial components, as in GConvLSTM, could improve predictions.
Authors: Matheus Corrêa Domingos, Valdivino Alexandre de Santiago Júnior, Juliana Aparecida Anochi, Elcio Hideiti Shiguemori
Institution: Instituto Nacional de Pesquisas Espaciais
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: gold
OpenAlex ID: W4412605705