Downscaling, bias correction, and spatial adjustment of extreme tropical cyclone rainfall in ERA5 using deep learning
Ascenso, G., Ficchì, A., Giuliani, M., Scoccimarro, E., Castelletti, A. (2024)
Weather and Climate Extremes, 46, 100724.
Abstract
Hydrological models that are used to analyse flood risk induced by tropical cyclones often input ERA5 reanalysis data. However, ERA5 precipitation has large systematic biases, especially over heavy precipitation events like Tropical Cyclones, compromising its usefulness in such scenarios. Few studies to date have performed bias correction of ERA5 precipitation and none of them for extreme rainfall induced by tropical cyclones. Additionally, most existing works on bias adjustment focus on adjusting pixel-wise metrics of bias, such as the Mean Squared Error (MSE). However, it is equally important to ensure that the rainfall peaks are correctly located within the rainfall maps, especially if these maps are then used as input to hydrological models. In this paper, we describe a novel machine learning model that addresses both gaps, RA-Ucmpd, based on the popular U-Net model. The key novelty of RA-Ucmpd is its loss function, the compound loss, which optimizes both a pixel-wise bias metric (the MSE) and a spatial verification metric (a modified version of the Fractions Skill Score). Our results show how RA-Ucmpd improves ERA5 in almost all metrics by 3-28%—more than the other models we used for comparison which actually worsen the total rainfall bias of ERA5—at the cost of a slightly increased (3%) error on the magnitude of the peak. We analyse the behaviour of RA-Ucmpd by visualizing accumulated maps of four particularly wet tropical cyclones and by dividing our data according to the Saffir-Simpson scale and to whether they made landfall, and we perform an error analysis to understand under what conditions our model performs best.
Reinforcement Learning of Multi-Timescale Forecast Information for Designing Operating Policies of Multi-Purpose Reservoirs
Zanutto, D., Ficchì, A., Giuliani, M., & Castelletti, A. (2025)
Water Resources Research, 61, e2023WR036724.
Abstract
Hydrological forecasts have significantly improved in skill over recent years, encouraging their systematic exploitation in multipurpose reservoir operations to improve reliability and resilience to extreme events. Despite the growing availability of multi-timescale forecasts, there is still a lack of transparent and integrated methods for selecting the most suitable forecast products, variables, and lead times for specific operational challenges. In this work, we propose a holistic approach based on Reinforcement Learning (RL) to design multipurpose dam operating policies informed by available multi-timescale forecast products. Our approach extends the traditional Evolutionary Multi-Objective Direct Policy Search method by parametrizing both the operating policy and the forecast information extraction process. We compare our RL approach with a state-of-the-art two-step procedure in which the forecast selection and processing are performed before the policy optimization. We demonstrate the value of the method for the multipurpose operation of Lake Como (Italy) by considering multi-timescale forecasts from short to seasonal lead times to manage flood- and drought-related operational objectives. Our approach identifies solutions achieving an 18% improvement in hypervolume indicator compared to policies not informed by forecasts and a 6% improvement over those designed using the two-step reference methodology. These improvements are accompanied by increased flexibility in policy design and trade-off analysis by directly extracting forecast information within the multi-objective optimization. This study demonstrates the feasibility and benefits of integrating policy design with forecast information extraction, particularly when multiple operational forecasts are available.
Model Predictive Control of water resources systems: A review and research agenda
Andrea Castelletti, Andrea Ficchì, Andrea Cominola, Pablo Segovia, Matteo Giuliani, Wenyan Wu, Sergio Lucia, Carlos Ocampo-Martinez, Bart De Schutter, José María Maestre (2024)
Annual Reviews in Control, Volume 55, Pages 442-465
Abstract
Model Predictive Control (MPC) has recently gained increasing interest in the adaptive management of water resources systems due to its capability of incorporating disturbance forecasts into real-time optimal control problems. Yet, related literature is scattered with heterogeneous applications, case-specific problem settings, and results that are hardly generalized and transferable across systems. Here, we systematically review 149 peer-reviewed journal articles published over the last 25 years on MPC applied to water reservoirs, open channels, and urban water networks to identify common trends and open challenges in research and practice. The three water systems we consider are inter-connected, multi-purpose and multi-scale dynamical systems affected by multiple hydro-climatic uncertainties and evolving socioeconomic factors. Our review first identifies four main challenges currently limiting most MPC applications in the water domain: (i) lack of systematic benchmarking of MPC with respect to other control methods; (ii) lack of assessment of the impact of uncertainties on the model-based control; (iii) limited analysis of the impact of diverse forecast types, resolutions, and prediction horizons; (iv) under-consideration of the multi-objective nature of most water resources systems. We then argue that future MPC applications in water resources systems should focus on addressing these four challenges as key priorities for future developments.
A decision-led evaluation approach for flood forecasting system developments: An application to the Global Flood Awareness System in Bangladesh
Hossain, S., Cloke, H. L., Ficchì, A., Gupta, H., Speight, L., Hassan, A., Stephens, E.
Journal of Flood Risk Management, e12959.
Abstract
Scientific and technical changes to flood forecasting models are implemented to improve forecasts. However, responses to such changes are complex, particularly in global models, and evaluation of improvements remains focussed on generalised skill assessments and not on the most relevant outcomes for those taking decisions. Recently, the Global Flood Awareness System (GloFAS) flood forecasting model has been upgraded from version 2.1 to 3.1 with a significant change to its hydrological model structure. In the updated version 3.1, a single fully configured hydrological model (LISFLOOD) has been adopted, including ground water and river routing processes, instead of two coupled models, a land surface and a simplified hydrological model, of the previous version 2.1. This study aims to evaluate changes in the simulated behaviour of floods and the forecast skill of the two GloFAS versions based on different decision criteria for early action. We evaluate GloFAS reforecasts for the Brahmaputra and the Ganges Rivers in Bangladesh for the period 1999–2018. For the Brahmaputra River, the old GloFAS 2.1 version performs better than the 3.1 version, especially in predicting low- (90th percentile) and medium-level (95th percentile) floods. For the Ganges, GloFAS 3.1 shows improved probability of detection of low- to medium-level floods compared to version 2.1, especially for lead times longer than 10 days. Both versions show limited skill for more extreme floods (99th percentile) but results are less robust for these less frequent floods given the lower number of events. Using lead-time dependent thresholds improves the false alarm ratio while reducing the probability of detection. The changes in model structures influence the model performance in a complex and varied way and forecast skill needs further investigation across regions and decision-making criteria. Understanding the skill changes between different model versions is important for decision-makers; however, focused case studies such as this should also be used by model developers to guide future changes to the system to ensure that they lead to improvements in decision-making ability.