This study utilizes a rich UK data set of smart demand metering data, household characteristics, and weather data to develop a demand forecasting methodology that combines the high accuracy of machine learning models with the interpretability of statistical methods. For this reason, a random forest model is used to predict daily demands 1 day ahead for groups of properties (mean of 3.8 households/group) with homogenous characteristics. A variety of interpretable machine learning techniques [variable permutation, accumulated local effects (ALE) plots, and individual conditional expectation (ICE) curves] are used to quantify the influence of these predictors (temporal, weather, and household characteristics) on water consumption. Results show that when past consumption data are available, they are the most important explanatory factor. However, when they are not, a combination of household and temporal characteristics can be used to produce a credible model with similar forecasting accuracy. Weather input has overall a mild to no effect on the model's output, although this effect can become significant under certain conditions.

@en

The moisture storage available to vegetation is a key parameter in the hydrological functioning of ecosystems. This parameter, the root zone storage capacity, determines the partitioning between runoff and transpiration, but is impossible to observe at the catchment scale. In this research, data from the experimental forests of HJ Andrews (Oregon, USA) and Hubbard Brook (New Hampshire, USA) was used to test the hypotheses that: (1) the root zone storage capacity significantly changes after deforestation, (2) changes in the root zone storage capacity can to a large extent explain post-treatment changes to the hydrological regimes and that (3) a time-dynamic formulation of the root zone storage can improve the performance of a hydrological model. At first, root zone storage capacities were estimated based on a simple, water-balance based method. Briefly, the maximum difference between cumulative rainfall and estimated transpiration was determined, which could be considered a proxy for root zone storage capacity. These values were compared with root zone storage capacities obtained from four conceptual models (HYPE, HYMOD, FLEX, TUW), calibrated for consecutive 2-year windows. Both methods showed a sharp decline in root zone storage capacity after deforestation, which was followed by a gradual recovery signal. It was found in a trend analysis that these recovery periods took between 5 and 13 years for the different catchments. Eventually, one of the models was adjusted to allow for a time-dynamic formulation of root zone storage capacity. This adjusted model showed improvements in model performance as evaluated by 28 hydrological signatures, such as rising limb density or peak flows. Thus, this research clearly shows the time-dynamic character of a crucial parameter, which is often considered to remain constant in time. Root zone storage capacities are strongly affected by deforestation, leading to changes in hydrological regimes, and time-dynamic formulations of root zone storage are therefore necessary in systems under change. @en

The core component of many hydrological systems, the moisture storage capacity available to vegetation, is impossible to observe directly at the catchment scale and is typically treated as a calibration parameter or obtained from a priori available soil characteristics combined with estimates of rooting depth. Often this parameter is considered to remain constant in time. Using long-term data (30-40 years) from three experimental catchments that underwent significant land cover change, we tested the hypotheses that: (1) the root-zone storage capacity significantly changes after deforestation, (2) changes in the root-zone storage capacity can to a large extent explain post-treatment changes to the hydrological regimes and that (3) a time-dynamic formulation of the root-zone storage can improve the performance of a hydrological model. A recently introduced method to estimate catchment-scale root-zone storage capacities based on climate data (i.e. observed rainfall and an estimate of transpiration) was used to reproduce the temporal evolution of root-zone storage capacity under change. Briefly, the maximum deficit that arises from the difference between cumulative daily precipitation and transpiration can be considered as a proxy for root-zone storage capacity. This value was compared to the value obtained from four different conceptual hydrological models that were calibrated for consecutive 2-year windows. It was found that water-balance-derived root-zone storage capacities were similar to the values obtained from calibration of the hydrological models. A sharp decline in root-zone storage capacity was observed after deforestation, followed by a gradual recovery, for two of the three catchments. Trend analysis suggested hydrological recovery periods between 5 and 13 years after deforestation. In a proof-of-concept analysis, one of the hydrological models was adapted to allow dynamically changing root-zone storage capacities, following the observed changes due to deforestation. Although the overall performance of the modified model did not considerably change, in 51% of all the evaluated hydrological signatures, considering all three catchments, improvements were observed when adding a time-variant representation of the root-zone storage to the model. In summary, it is shown that root-zone moisture storage capacities can be highly affected by deforestation and climatic influences and that a simple method exclusively based on climate data can not only provide robust, catchment-scale estimates of this critical parameter, but also reflect its timedynamic behaviour after deforestation.

@en