This commentary discusses a framework for the benchmarking of hydrological models for different purposes when the datasets for different catchments might involve epistemic uncertainties. The approach might be expected to result in an ensemble of models that might be used in prediction (including models of different types) but also provides for model rejection to be the start of a learning process to improve understanding.

Flow recession analysis, relating discharge Q and its time rate of change -dQ/dt, has been widely used to understand catchment scale flow dynamics. However, data points in the plot of -dQ/dt versus Q typically form a wide point cloud due to noise and hysteresis, and it is still unclear what information we can extract from the data points and how to understand the information. In this study, we utilize a machine learning tool to capture the point cloud using the past trajectory of discharge. Our results show that most of the data points can be captured using 5 days of past discharge. While analyzing the machine learning model structure and the trained parameters is a daunting task, we show that we can learn the catchment scale flow recession dynamics from what the machine learned. We analyze patterns learned by the machine and explain and hypothesize why the machine learned those characteristics. The hysteresis in the plot mainly occurs during the early time dynamics, and the flow recession dynamics eventually converge to an attractor in the plot, which represents the master recession curve. We also illustrate that a hysteretic storage-discharge relationship can be estimated based on the attractor.

Flow recession analysis, relating discharge Q and its time rate of change -dQ/dt, has been widely used to understand catchment scale flow dynamics. However, data points in the recession plot, the plot of -dQ/dt versus Q, typically form a wide point cloud due to noise and hysteresis in the storage-discharge relationship, and it is still unclear what information we can extract from the plot and how to understand the information. There seem to be two contrasting approaches to interpret the plot. One emphasizes the importance of the ensembles of many recessions (i.e., the lower envelope or a measure of central tendency), and the other highlights the importance of the event scale analysis and questions the meaning of the ensemble characteristics. In this study, we examine if those approaches can be reconciled. We utilize a machine learning tool to capture the point cloud using the past trajectory of discharge. Our results show that most of the data points can be captured using 5 days of past discharge. We show that we can learn the catchment scale flow recession dynamics from what the machine learned. We analyze patterns learned by the machine and explain and hypothesize why the machine learned those characteristics. The hysteresis in the plot mainly occurs during the early time dynamics, and the flow recession dynamics eventually converge to an attractor in the plot, which represents the master recession curve. We also illustrate that a hysteretic storage-discharge relationship can be estimated based on the attractor.

The representation of rainfall is important for hydrological modelling, particularly for spatially distributed models. Accurate estimation of rainfall is particularly challenging in mountainous regions where observations are often sparse relative to the spatial variability of rainfall, making interpolation challenging. In these regions, orographic processes lead to complex patterns of rainfall enhancement and rain shadow depletion. This study tests one deterministic method, Natural Neighbour Interpolation (NNI), and two geostatistical methods, ordinary kriging (OK) and ordinary cokriging (CK), to determine if CK improves rainfall interpolation during three extreme rainfall events that occurred in the north west of England. Preliminary analysis using long-term annual average rainfall totals, including additional high elevation rainfall observations, showed that CK with an effective elevation index as a secondary variable performed better than NNI and OK with an overall improvement of around 40%. Using rainfall totals for long-term wind direction and wind speed rainfall classes, CK performance was variable across classes but provided an improvement of approximately 15% for wind direction classes without an easterly wind component. For 15-minute timesteps during extreme rainfall events, there were comparatively small differences between interpolation methods, attributed to having only relatively low elevation rainfall observations for cross-validation, providing weak constraint. Importantly, cross-variogram estimation (that controls the strength of the correlation between rainfall magnitude and the secondary variable) provided differing cross-validation results when estimated for different rainfall total periods: 15-minutes, hourly, daily and long-term. Variograms and cross variograms estimated at a 15-minute timestep frequency were robust for many timesteps, but were difficult to fit automatically for others. Variograms estimated from longer periods were more reliably estimated, but tended to have lower variance and cross-variance and longer correlation ranges producing a smoother interpolated rainfall field. Given the weak cross-validation constraint, care must be taken in identifying the most appropriate method and variogram estimation period.

There is increased interest in the potential of tree planting to help mitigate flooding using nature-based solutions or natural flood management. However, many publications based upon catchment studies conclude that, as flood magnitude increases, benefit from forest cover declines and is insignificant for extreme flood events. These conclusions conflict with estimates of evaporation loss from forest plot observations of gross rainfall, throughfall and stem flow. This study explores data from existing studies to assess the magnitudes of evaporation and attempts to identify the meteorological conditions under which they would be supported. This is achieved using rainfall event data collated from publications and data archives from studies undertaken in temperate environments around the world. The meteorological conditions required to drive the observed evaporation losses are explored theoretically using the Penman-Monteith equation. The results of this theoretical analysis are compared with the prevailing meteorological conditions during large and extreme rainfall events in mountainous regions of the UK to assess the likely significance of wet canopy evaporation loss. The collated dataset showed that Ewc losses between approximately 2 and 38% of gross rainfall (1.5 to 39.4 mm d-1) have been observed during large rainfall events (up to 118 mm d-1) and limited data for extreme events (> 150 mm d-1). Event data greater than 150 mm, where duration was not reported, showed similarly high percentage evaporation losses. Theoretical estimates of wet-canopy evaporation indicated that, to reproduce these high losses, relative humidity and the aerodynamic resistance for vapour transport needed to be within an envelope of approximately 90 to 97.5% and 0.5 to 2 s m-1 respectively. Surface meteorological data during large and extreme rainfall events in the UK suggest that conditions favourable for high wet-canopy evaporation are not uncommon and indicate that significant evaporation losses during large and extreme events are possible but not for all events and not at all locations. Thus the disparity with the results from catchment studies remains.

A document by Keith Beven. Click on the document to view its contents.