DEEP NEURAL NETWORKS FOR FORECASTING ENVIRONMENTAL CHANGES: APPLICATIONS TO CLIMATE MODELING

Abstract

This research focuses on portraying temperature and precipitation predictions by using deep neural networks (DNNs) in climate modeling. Real world climate systems now possess increasing complex structures which makes it challenging for traditional forecasting methods to track intricate patterns and relationships. Our study employed a CNN-RNN hybrid architecture to exploit the use of spatial and temporal data for this purpose. Assessment of the models involved a 5-year dataset which demonstrated superior accuracy levels for the DNN models when compared to conventional methods. When predicting temperatures the hybrid combination of CNN-RNN achieved an optimal performance through a Mean Absolute Error (MAE) result of 0.75°C and Root Mean Squared Error (RMSE) of 0.94°C while presenting the highest correlation value of 0.94. Result data from the sensitivity study demonstrated that temperature elements led to the greatest changes in model operational capacity. The newly developed forecasting model confirmed its superior capabilities through testing against standard numerical weather prediction models as well as statistical approaches resulting in better RMSE and reduced correlation coefficients. The research results demonstrate that DNNs can make substantial improvements to climate forecasting by delivering better accuracy than conventional forecasting models. More research will become essential to develop these models because they currently face challenges with data integration and interpretability problems. The presented work establishes a viable method that leads to better and more practical climate forecasting capabilities within environmental research domains.