Signatures-based appraisal of global rainfall datasets to capture hydrological trends in a meso-scale catchment

Precipitation data is a critical input for hydrological models that regulates the spatio-temporal variability of other hydrological fluxes. However, in many regions worldwide, obtaining in-situ rainfall data remains a challenge. In such situations, global rainfall products can be valuable, providing global/regional coverage but these products are susceptible to errors from various factors. Previous studies have assessed the performance of different rainfall products to simulate hydrological models, primarily on their ability to reproduce time series of output variables (streamflow, groundwater level, evapotranspiration, or soil moisture), quantified using various error metrics. While the comparison of time series using error metrics provides insights into general model performance, it may not adequately highlight the capability of these products to simulate specific catchment characteristics, such as groundwater contribution to streamflow, catchment behaviour in high/low flows, etc. Utilizing hydrological signatures can offer additional insights into the hydrological behaviour of the modelled catchment. Therefore, this study aims to evaluate the potential of global rainfall datasets to capture catchment’s hydrological characteristics using a range of hydrological signatures for streamflow and groundwater levels, beyond the traditional time series comparisons.

The analysis was conducted on a meso-scale transboundary catchment, Aa of Weerijs, covering an area of 346 km². A fully distributed physically based hydrological model coupled with a hydrodynamic model was setup using the MIKE-SHE and MIKE-11 modelling tools of DHI, Denmark. The base model had a grid size of 500 by 500 m and fed with rainfall data from three local gauge stations (2010-2019). Four rainfall products (MSWEP, IMERG, ERA5 land and E-OBS) were shortlisted based on their comparative fine spatial resolution. To achieve the objective, firstly, a direct comparison of rainfall data from these products was conducted against rainfall data from the gauge stations using metrics such as probability of detection, false alarm ratio, equitable threat score and frequency bias. Secondly, the model was run with each dataset, and the performance assessment of the simulated outputs was done using hydrological signatures. The selected signatures included the flow duration curve's (FDC) high-flow segment volume, FDC's mid segment slope, groundwater duration curve, base flow index, runoff ratio, rising limb density, autocorrelation

The findings indicate that the performance of a rainfall product in direct comparison with a gauge station may not consistently align with its effectiveness in simulating model variables. Furthermore, the quantification of a product’s ability to simulate output variables varies depending on the evaluation criteria or metrics used. We advocate for the use of a range of hydrological signatures in the assessment criteria, as it provides additional insights into the capability of global datasets to simulate hydrological responses.