Unveiling the impact of cosmic rays and solar activities on climate through optimized boost algorithms

Abstract

This investigation explores the enhancement of climate anomaly predictions by incorporating Solar Sunspot Number (SSN) and Cosmic Ray (CR) data into climate models. Leveraging XGBoost and CatBoost regression methodologies enhanced by Atom Search Optimization (ASO) and Nuclear Reaction Optimization (NRO) for predictive analysis. Utilizing a dataset spanning from 1965 to 2020, comprising 672 data points per climate parameter, the study delves into the dynamics between CR flux, SSN variability, and climate parameters. The models aimed to forecast variations in total precipitation anomaly (TPA), total cloud cover anomaly (TCCA), and sea surface temperature anomaly (SSTA) based on decadal solar cycle activities and CR data. Our findings reveal the significant impact of integrating SSN and CR data into environmental prediction models for TCCA, TPA, and SSTA, employing CatBoost and XGBoost machine learning (ML) algorithms. Performance evaluation, centered on root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), and Nash-Sutcliffe efficiency (NSE), illuminated the efficacy of ASO and NRO in model optimization, particularly under scenarios with and without SSN/CR data inclusion. The analytical outcomes underscore the enhanced prediction accuracy for TCCA, TPA, and SSTA when incorporating SSN and CR data, with ASO generally outperforming NRO in optimizing model parameters. Our regression models, optimized using ASO and NRO, showed a marked improvement in SSTA forecasts, with an increase in the R² value from 0.73 to 0.76 when SSN/CR data were not included. The CatBoost was superior the XGBoost models with results of four error metrics. These results underscore the critical role of solar activity data and optimized algorithms in enhancing the accuracy and reliability of climate modeling. This study underscores the utility of advanced ML techniques and the importance of strategic variable selection in environmental modeling, offering new insights into the complex interactions between solar activity, CR, and climate dynamics.