The recent WeatherBench 2 provides a versatile framework for the verification of deterministic and ensemble forecasts. In this paper, we present an explicit extension to binary forecasts of hydroclimatic extremes. Seventeen verification metrics for binary forecasts are employed, and scorecards are generated to showcase the predictive performance. The extension facilitates more comprehensive comparisons of hydroclimatic forecasts and provides useful information for forecast applications.

The classic logistic function characterizes the stationary relationship between drought loss and intensity. This paper accounts for time in the magnitude, shape and location parameters of the logistic function and derives nonstationary intensity loss functions. A case study is designed to test the functions for drought-affected populations by province in mainland China from 2006 to 2023. Overall, the nonstationary intensity loss functions are shown to be a useful tool for drought management.

The local performance plays a critical part in practical applications of global streamflow reanalysis. This paper develops a decomposition approach to evaluating streamflow analysis at different timescales. The reanalysis is observed to be more effective in characterizing seasonal, annual and multi-annual features than daily, weekly and monthly features. Also, the local performance is shown to be primarily influenced by precipitation seasonality, longitude, mean precipitation and mean slope.

This paper focuses on the effect of the water balance constraint on the robustness of the long short-term memory (LSTM) network in learning rainfall-runoff relationships. Through large-sample tests, it is found that incorporating this constraint into the LSTM improves the robustness, while the improvement tends to decrease as the amount of training data increases. The results point to the compensation effects between training data and process knowledge on the LSTM’s performance.

The lack of physical mechanism is a critical issue for the use of popular deep learning models. This paper presents an in-depth investigation of the fundamental mass balance constraint for deep learning-based rainfall-runoff prediction. The robustness against data sparsity, random parameters initialization and contrasting climate conditions are detailed. The results highlight that the water balance constraint evidently improves the robustness in particular when there is limited training data.