We proposed a methodology to obtain useful and robust probabilistic predictions from computational flood simulators using satellite-borne flood extent observations. We developed a Bayesian framework to obtain the uncertainty in roughness parameters, in observations errors, and in simulator structural deficiencies. We found that it can yield improvements in predictions relative to current methodologies and can potentially lead to consistent ways of combining data from different sources.

Reanalysis datasets are increasingly used to drive flood models, especially for continental and global analysis. We investigate the impact of using four reanalysis products on simulations of past flood events. All reanalysis products underestimated the number of buildings inundated, compared to a benchmark national dataset. These findings show that while global reanalyses provide a useful resource for flood modelling where no other data are available, they may underestimate impact in some cases.

Streamflow forecasts a day to a month ahead are highly valuable for water resources management. Current practice often develops forecasts for specific lead times and aggregation timescales. In contrast, a single, seamless forecast can serve multiple lead times/timescales. This study shows seamless forecasts can match the performance of forecasts developed specifically at the monthly scale, while maintaining quality at other lead times. Hence, users need not sacrifice capability for performance.

The study compares the uncertainty of discharge curves for vegetated channels, calculated using several methods, including the simplest ones, based on the Manning formula and advanced approaches, providing a detailed physical representation of the channel flow processes. Parameters of each method were identified for the same data sets. The outcomes of the study include the widths of confidence intervals, showing which method was the most successful in explaining observations.