Estimating severe low flows on human-influenced catchments by combining weather generator, analogue spatial disaggregation, and hydrological modelling under historical and future climate.

Understanding and quantifying severe low flows is crucial for the management of hydropower or thermal power plants. Moreover, low flows are strongly related to the climatic regime and will be affected by climate change. Therefore, we propose a modelling chain to estimate severe low flow values for several human-influenced catchments in France, both under current and future climate.

Firstly, a bivariate weather generator (Touron, 2019) of daily temperature and precipitation, representing the average of 28 catchments spread out over France, was trained, and used to generate 1000 meteorological time series over a 30-year period. Average daily precipitation and temperature are then spatially disaggregated to produce 1000 local time series for each of the 28 catchments using an analogue approach. Thirdly, MORDOR-SD a lumped conceptual rainfall-runoff model, developed and used at EDF (Garavaglia et al. 2017), combined with upstream-downstream propagation and water management module was forced by the 1000 local meteorological time series. The resulting 1000 time series of simulated river flows are then used to calculate an empirical rare percentile estimate of low flows across 12 large catchments of interest.

The methodology is applied on historical period (1981-2010) using precipitation and temperature observations to train the weather generator. The robustness of the method is evaluated by comparing return levels of low flows obtained through the proposed method and the ones estimated through river flow observations available. Finally, to assess the impact of climate change, the weather generator is also trained/used with 5 downscaled climate projections from the CMIP5 experiments corresponding to: (1) the historical period (1981-2010) and, (2) 4 storylines representing different levels of warming/drying (2036-2065).

The comparison over the historical period has shown the relative agreement between simulated and observed severe low flows. Furthermore, under future conditions, the climatic differences between the 4 storylines lead to logical differences in the estimation of severe low flows, i.e. warmer/drier storylines lead to lower estimation of severe low flows.