Climate data plays a crucial role in water resources management, which is becoming an increasingly relevant asset in all types of hydrological analysis not only for climate change studies but for various horizon forecasting. Though the ever-improving accuracy of climate models' spatial and temporal resolution has surged the validity of their outputs, the products of global and regional climate models need to be corrected to be reliably used for local purposes. Here, we propose a comprehensive analysis of statistical univariate and multivariate, as well as machine learning methods for bias correction, which are compared on different temporal scales, ranging from hourly time steps to monthly aggregations, in an environment of complex Alpine orthography, using ERA5-Land reanalysis data. The results reveal different trends in the performance of the bias correction methods for precipitation and temperature across the various time resolutions.

Using the global climate model outputs without any adjustment may bring errors in water resources and climate change investigations. This study tackles the critical issue of bias correction temperature in ERA5-Land reanalysis for 10 ground stations in northern Italy using nine machine learning (ML) techniques. Among standalone ML models, XGBoost regression emerged as the most effective standalone ML model, outperforming others across 6 out of 10 stations, while random forest regression, Gaussian process regression, and support vector regression obtained the second to fourth places. In contrast, AdaBoost regression (ABR) achieved the least favorable performance. Furthermore, nine ensemble ML models are proposed to correct bias of the reanalysis of temperature data. The results indicated that the K-nearest neighbors-based ensemble model excelled and secured the top rank in 7 out of 10 stations, while the multiple linear regression-based ensemble model achieved the highest precision in 4 out of 10 stations. Furthermore, other ML-based ensemble models displayed satisfactory results. On the other hand, the ABR-based ensemble model exhibited the lowest accuracy among ML-based ensemble models. The findings highlight the potential of ML-based ensemble models in effectively addressing bias correction in climate data.

Drinking water demand modelling and forecasting is a crucial task for sustainable management and planning of water supply systems. Despite many short-term investigations, the medium-term problem needs better exploration, particularly the analysis and assessment of meteorological data for forecasting drinking water demand. This work proposes to analyse the suitability of ERA5-Land reanalysis data as weather input in water demand modelling. A multivariate deep learning model based on the long short-term memory architecture is used in this study over a prediction horizon ranging from seven days to two months. The performance of the model, fed by ground station data and ERA5-Land data, is compared and analysed. Close-to-operative forecasting is then presented using observed data for training and ERA5-Land dataset for testing. The results highlight the reliability of the proposed architecture fed by ERA5-Land data for different time horizons. In particular, the ERA5-Land shows promising performance as input of the multivariate machine learning forecasting model, although some meteorological biases are present, which can be improved, especially in close-to-operative application with bias correction techniques. The proposed study leads to practical implications in the use of regional climate model outputs to support drinking water forecasting for sustainable and efficient management of water distribution systems.