Hydrological models are important tools for making predictions of river flows to be used in water resources management strategies, and for hypotheses testing in hydrological research. Hydrological modelling studies have been constrained by data availability and computational efficiency in the past; however, the development of national and global open-source data products and significant gains in computational power has allowed hydrological models to be implemented on many spatial scales. DECIPHeR (Dynamic fluxEs and ConnectIvity for Predictions of Hydrology) is a new flexible hydrological modelling framework that simulates streamflow from spatial scales of small headwaters catchments to entire continents. DECIPHeR can be adapted by the user to specific hydrological systems and data availability, and can be modified to represent different hydrological processes. Here, DECIPHeR is applied to the Upper Niger River in West Africa – a large and data scarce basin. This basin is characterised by highly variable climatic and physiographic features from its upstream headwaters to the downstream outlet. The initial DECIPHeR model structure was applied across the Upper Niger basin, forced with one global precipitation data product (MSWEP), and three global PET data products (GLEAM, ECMWF EartH2Observe, and temperature-based estimates). Model performance was evaluated with three performance metrics (NSE, PBIAS, and low-flow volume bias). Initial simulations were able to reproduce the shape and timing of the flow peaks, but were overpredicting flow volumes by a factor of four in all seven sub-basins. The model structure was modified to include a module to represent evaporation from the saturated zone and applied across the basin. This improved model performance significantly in all sub-basins. However, ‘behavioural’ (i.e. NSE>0.5, PBIAS<10, low-flow volume bias<10) model simulations could only be identified in three downstream sub-basins. This is due to large water balance issues in the headwater catchments, likely caused by large errors in the global input data products and a lack of information about the processes occurring in these sub-basins. However, this study has shown that DECIPHeR is particularly valuable for modelling studies in domains where there are little or no ground observations to inform our understanding of catchment functioning, and allows for multiple hypotheses of the dominant processes to be tested.
|Date of Award||19 Mar 2019|
- The University of Bristol
|Supervisor||Jeff Neal (Supervisor) & Jim Freer (Supervisor)|