Forecasting Precipitation and Temperature Evolution Patterns Under Climate Change Using a Random Forest Approach With Seasonal Bias Correction

Abstract

The accurate prediction of climate patterns is crucial for effective environmental management and strategic planning, particularly in the face of increasing climate variability and extreme weather events. This study focuses on enhancing the predictive accuracy of global climate models by integrating a random forest algorithm with seasonal bias correction, which is aimed at forecasting changes in precipitation and temperature across various climate scenarios. Our approach significantly improves the precision of climate forecasts by using a comprehensive dataset that includes historical climate data and future projections under Representative Concentration Pathways. For precipitation, after implementing seasonal bias correction, the Correlation Coefficient improved from 0.826 to 0.860, and the Nash–Sutcliffe Efficiency increased from 0.633 to 0.735. In addition, the Mean Absolute Error and Root Mean Square Error decreased, thereby enhancing the model's reliability in predicting extreme precipitation events. Temperature forecasts also benefited from the Random Forest model, achieving a Correlation Coefficient of up to 0.987, indicating a strong predictive performance. These improvements highlight the potential of machine learning methods in refining climate models, thus providing more accurate tools for policymakers and planners to manage climate risks. Furthermore, the successful application of these advanced statistical techniques underscores the importance of continuous innovation in climate science, ensuring that climate projections remain relevant and reliable in informing global climate resilience and adaptation strategies.