NN-XGBoost: Nearest-Neighbor eXtreme Gradient Boosting for Temporal Downscaling and Prediction

Accurate forecasting of Downward Shortwave Radiation (DSWR) and other climate data is crucial for applications in renewable energy, agriculture, and environmental research. There are many products available providing accurate estimates for DSWR at a coarse temporal resolution, however global datasets often do not provide an accurate localized fine temporal resolution DSWR estimate required for proper power system planning and operation. This study proposes to use a combination of the nearest-neighbor (NN) method and the eXtreme Gradient Boosting algorithm (XGBoost) to provide accurate temporal downscaling and prediction via integrating localized smoothing with the global predictive power of XG-Boost. Proposed nearest neighbor eXtreme Gradient Boosting algorithm (NN-XGBoost) model is employed to use downscale the global DSWR data obtained from Open-Meteo and to match the higher-resolution local meteorological observations obtained from Ambient Weather. The data integration involves temporal downscaling of global datasets to match the higher-resolution local data. Two comparison approaches are explored: (1) using a single local variable and (2) using multiple local variables as predictors for DSWR. The results showed that the NN-XGBoost model consistently outperforms XGBoost and ARIMAX in terms of root mean square error (RMSE) and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, in both single-variable and multiple-feature cases. This study demonstrates the effectiveness of the NN-XGBoost model for predicting DSWR, offering a scalable methodology for improving the precision of renewable energy forecasting and environmental monitoring.