When one is using climate simulation outputs, one critical issue to consider is the systematic bias affecting the modelled data. The bias correction of modelled data is often used when one is using impact models to assess the effect of climate events on human activities. However, the efficacy of most of the currently available methods is reduced in the case of extreme events because of the limited number of data for these low probability and high impact events. In this study, a novel bias correction methodology is proposed, which corrects the bias of extreme events. To do so, we extended one of the most popular bias correction techniques, i.e., quantile mapping (QM), by improving the description of extremes through a generalised extreme value distribution (GEV) fitting. The technique was applied to the daily mean temperature and total precipitation data from three seasonal forecasting systems: SEAS5, System7 and GCFS2.1. The bias correction efficiency was tested over the Southern African Development Community (SADC) region, which includes 15 Southern African countries. The performance was verified by comparing each of the three models with a reference dataset, the ECMWF reanalysis ERA5. The results reveal that this novel technique significantly reduces the systematic biases in the forecasting models, yielding further improvements over the classic QM. For both the mean temperature and total precipitation, the bias correction produces a decrease in the Root Mean Squared Error (RMSE) and in the bias between the simulated and the reference data. After bias correcting the data, the ensemble forecasts members that correctly predict the temperature extreme increases. On the other hand, the number of members identifying precipitation extremes decreases after the bias correction.
We compare three machine learning models—artificial neural network (ANN), random forest (RF), and convolutional neural network (CNN)—for spatial downscaling of temperature at 2 m above ground (T2M) from a 9 km ERA5-Land reanalysis to 1 km in a complex terrain area, including the Non Valley and the Adige Valley in the Italian Alps. The results suggest that CNN performs better than the other methods across all seasons. RF performs similar to CNN, particularly in spring and summer, but its performance is reduced in winter and autumn. The best performance was observed in summer for CNN (R2 = 0.94, RMSE = 1 °C, MAE = 0.78 °C) and the lowest in winter for ANN (R2 = 0.79, RMSE = 1.6 °C, MAE = 1.3 °C). Elevation is an important predictor for ANN and RF, whereas it does not play a significant role for CNN. Additionally, CNN outperforms others even without elevation as an additional feature. Furthermore, MAE increases with higher elevation for ANN across all seasons. Conversely, MAE decreases with increased elevation for RF and CNN, particularly for summer, and remains mostly stable for other seasons.
20 February 2025 in Bulletin of Atmospheric Science and Technology
This study presents a novel bias correction methodology for enhancing the reliability of climate projections, particularly in regions vulnerable to climate change impacts, such as the Southern African Development Community (SADC). This methodology is here applied jointly with a downscaling method to correct systematic distributional biases and refine the spatial resolution of climate model outputs, with the aim of improving the identification of extreme weather events. Our work builds upon a unique bias correction approach that enhances the quantile mapping (QM) technique by fitting a generalised extreme value distribution (GEV) to the tails of the distribution. This approach addresses the limitations of traditional methods when dealing with extreme events, which are often tied to the scarcity of data for such high-impact, yet low-probability occurrences. Moreover, we extend the applicability of our methodology to climate projections by integrating an approach designed to preserve relative trends in climate data. The technique is then applied to the daily mean temperature data of long-term climate projections (until 2100) from the Coupled Model Intercomparison Project Phase 6 (CMIP6). The bias correction of historical simulations produces an increase in the ensemble forecast members predicting temperature extremes in agreement with the reference dataset (i.e., the reanalysis ERA5-Land). When analysing future scenarios, there is an expected increase in both the intensity and frequency of extreme events, especially in more pessimistic scenarios, such as SSP585. However, these results vary depending on the specific climate model used. Our results suggest that considering multiple climate models of varying resolutions and comparing their outputs against observations and other climate datasets might be key to a comprehensive understanding of extreme event trends in climate scenarios. This approach accounts for each model’s strengths and limitations, providing a robust analysis of changes in extreme weather phenomena.