Simulating sediment discharge at water treatment plants under different land use scenarios using cascade modelling with an expert-based erosion-runoff model and a deep neural network

Abstract. Excessive sediment discharge in karstic regions can be highly disruptive to water treatment plants. It is essential for catchment stakeholders and drinking suppliers limit the impact of high loads on potable supply, but their strategic choices must based simulations integrating surface groundwater transfers taking into account possible changes land use. Karstic environments are particularly challenging as they face a lack accurate physical descriptions modelling process, complex predict due non-linearity processes generating discharge. The aim study was assess variability at plant according multiple realistic use scenarios. To reach that goal, we developed new cascade approach with an erosion-runoff geographic information system (GIS) model (WaterSed) deep neural network. used Radicatel hydrogeological (106 km2 Normandy, France), where spring extracted plant. simulated five design storms under current compared four scenarios (baseline, ploughing up grassland, eco-engineering, best farming practices, coupling eco-engineering/best practices). Daily rainfall time series WaterSed outputs connected sinkholes (positive dye tracing) were input data network model. structure found by classical trial-and-error procedure, trained 2 significant hydrologic years. Evaluation test set showed good performance (NSE = 0.82), application monthly backward-chaining nested cross-validation revealed able generalize datasets. Simulations made suggested 33 % grasslands would increase 5 average. By contrast, eco-engineering practices will significantly reduce (respectively ranges 10 %–44 24 %–61 %). these two strategies most efficient since it affects hydro-sedimentary production transfer (decreasing from 40 80 this offers interesting opportunities prediction springs or plants also provides robust decision-making tools planning suppliers.