Förbättra uppskattningar av solbestrålning på ytan i Arktis:utvärdering och biaskorrigering av CARRA-reanalysenmed hjälp av maskininlärning
Year: 2026
Type: dissertation
Abstract: Accurate estimation of surface solar irradiance is essential for understanding Arctic climate processes and evaluating the potential for renewable energy. In this study, the performance of the Copernicus Arctic Regional Reanalysis (CARRA) in estimating downwelling shortwave irradiance (DSI) over Svalbard was evaluated and improved using statistical and machine-learning bias-correction methods. Observations from 21 pyranometer stations collected between 2006 and 2024 were used to quantify systematic biases and train correction models, including Multiple Linear Regression (MLR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and a Multilayer Perceptron (MLP). Results show that uncorrected CARRA data systematically overestimate irradiance, with a mean bias of +22.8 W m⁻² and a limited ability to reproduce high-irradiance conditions above 630 W m⁻². All bias-correction methods improved accuracy, with XGBoost performing best, reducing Root Mean Square Error (RMSE) by approximately 25–35% and nearly eliminating mean bias. Temporal and solar-geometry predictors dominated model skill, while spatial variables added little. Residual slope bias and limited spatial generalisation remained, particularly in complex terrain. These findings demonstrate that machine learning can enhance reanalysis-based irradiance estimates in Svalbard, offering a framework for improved solar resource assessment and climate monitoring in data-sparse regions.
Author: Libal, Clara
Language: English
Field: Energy
Domain: Physical Sciences
SDG: Climate action
Open Access status: green
OpenAlex ID: W7204676004