Geological controls of discharge variability in the Thames Basin, UK from cross-spectral analyses: Observations versus modelling

• Published in: Journal of hydrology (Amsterdam) 2023-10, Vol.625, p.130104, Article 130104
• Main Authors: Weedon, Graham P., Robinson, Emma L., Bloomfield, John P., Turner, Stephen, Crane, Emily J., Best, Martin J.
• Format: Article
• Language: English
• Subjects: BFI; Discharge variability; Geological controls; Lag1 autocorrelation; Spectral analysis; Thames Basin; Transfer function
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2023.130104

•In catchments 300 km2 variability mostly from mixing upstream- and downstream-flow.•JULES land surface model performs poorly in small, high-permeability catchments.•Diverting some rainfall as bypass flow to bedrock improves JULES simulated flow. Geological factors controlling daily- to multi-year discharge variability in 48 sub-catchments spanning 10–1000 km2 in the Thames Basin were investigated using cross-spectral analysis. The analyses represent a ‘transfer function approach’ applied to daily observed streamflow (output) versus catchment-wide precipitation (input) for data spanning 1990–2014. Catchments dominated by high-permeability bedrock have significant attenuation of high-frequency precipitation variability and large delays at all frequencies with streamflow dominated by baseflow (high lag1 autocorrelation and high Base Flow Index, BFI). Catchments dominated by low-permeability rocks have little high-frequency attenuation and small delays and consequently ‘flashy’ behaviour. For all sub-catchments >300 km2 in the Thames Basin, attenuation of the highest frequency precipitation variability caused by mixing of flow from upstream plus groundwater flow (representing ‘older’ variability) with direct surface flow (‘younger’ variability) constitutes real-world moving averaging as indicated by a roll-off in power at the highest frequencies. The success of the JULES land surface model in simulating discharge (i.e. surface and sub-surface runoff routed between grid boxes) is also linked to the underlying geology. Larger catchments (>300 km2) are modelled well because routing between numerous grid boxes leads to moving averaging that is a good analogue for the observations. Modelling was least successful (e.g. lowest Kling-Gupta Efficiency) for small catchments (