No abstract

No abstract

<p>The summers of 2017 to 2020 were characterized by exceptional dry spells throughout Europe. Climate models show that such periods of drought could occur more frequently and become even more extreme in the future. The recent periods of intense droughts lead to significant ecological, economic and even societal damages in Flanders (Belgium). During these summers, receding groundwater levels were observed throughout Flanders that reached historical low levels. To monitor low ground water levels and to support a proactive drought management, the Flemish government developed an operational ground water indicator. This indicator gives an overview of the current phreatic ground water levels combined with a prediction for the next month for a selected number of phreatic wells. To increase the spatial resolution of the indicator, we developed a novel data driven regional ground water model for phreatic aquifers.</p><p>The ML model combines a gradient boosting decision tree model (CatBoost) with a Long Short Term Memory (LSTM) network. CatBoost is used to model the average ground water depth at each location. This value is passed to the LSTM network that predicts the temporal evolution of the groundwater at each location around its average. The training dataset for the CatBoost model contains the average groundwater depth of 5.673 wells spread across Flanders and a large set of explanatory variables related to soil texture, distance to a drainage, geology, topography, meteorology and land use. The model performance is evaluated using cross-validation which showed the model generalizes well with a mean absolute error of 69cm. The most important explanatory variables for the model are the thickness of the phreatic aquifer, the vertical distance to closest drain, the topographic index and the precipitation surplus.</p><p>The training dataset for the LSTM model contains 408 wells that have sufficiently long and reliable observations for training. The input data to the LSTM consists of rainfall and evapotranspiration up to 10 years prior to each observation, combined with the same explanatory variables as the CatBoost model. A single regional LSTM model is trained on all 408 wells simultaneously. The resulting model is accurate with a median RMSE of 20cm for the validation data, outperforming the currently used SWAP models [1]. The ML model is however less performant in simulating the higher ground water depths during summer and shows a consistent bias towards lower ground water depths during long dry spells.</p><p>[1] Kroes, J.G., J.C. van Dam, R.P. Bartholomeus, P. Groenendijk, M. Heinen, R.F.A. Hendriks, H.M. Mulder, I. Supit, P.E.V. van Walsum, 2017. SWAP version 4; Theory description and user manual. Wageningen, Wageningen Environmental Research, Report 2780</p>

<p>Belgium is ranked 23rd out of 164 countries in water scarcity and the third highest in Europe according to the Water Resource Institute. The warm and dry summers of the past few years have made it clear that Flanders has little if any buffer to cope with a sharp increase in water demand or a prolonged period of dry weather. To increase the resilience and preparedness against droughts, we developed the framework named hAIdro: an operational early warning system for low flows that allows to take timely, local and effective measures against water shortages. Data driven rainfall-runoff models are at the core of the forecasting system that allows to forecast droughts up to 30 days ahead.</p><p>The architecture of the data driven hydrological models are inspired by the Multi-Timescale Long Short Term Memory (MTS-LSTM, [1]) that allow to integrate past and future data in one prediction pipeline. The model architecture consists of 3 LTSM’s that are organized in a branched structure. The historical branch processes the historical meteorological data, remote sensing data and static catchment features into encoded state vectors. These are passed through fully connected layers to both a daily and an hourly forecasting branch which are used to make runoff predictions on short (72 hours) and long (30 days) time horizons. The forecasting branches are fed with forecasts of rainfall and temperature, static catchment features and discharge observations. The novelty of the proposed model structure lies in the way discharge observations are incorporated. Only the most recent discharge observations are used in the forecasting branches to minimize the consequences of missing discharge observations in an operational context. The models are trained using a weighted Nash-Sutcliffe Efficiency (NSE) as objective function that puts additional emphasis on low flows. Results show that the newly created data driven models perform well compared to calibrated lumped hydrological PDM models [2] for various performance metrics including Log-NSE and NSE.</p><p>We developed a custom cloud-based operational forecasting system, called hAIdro to bring the data driven hydrological models in production. hAIdro processes large quantities of local meteorological measurements, radar rainfall data and ECMWF extended range forecasts to make probabilistic forecasts up to 30 days ahead. hAIdro has been forecasting the runoff twice a day for 262 locations spread over Flanders since April 2021. A continuous monitoring and evaluation framework provides valuable insights in the online model performance and the informative value of hAIdro.</p><p>[1] M. Gauch, F. Kratzert, D. Klotz, G. Nearing, J. Lin, and S. Hochreiter. “Rainfall–Runoff Prediction at Multiple Timescales with a Single Long Short-Term Memory Network.” Hydrol. Earth Syst. Sci., 25, 2045–2062, 2021 </p><p>[2] Moore, R. J. “The PDM rainfall-runoff model.” Hydrol. Earth Syst. Sci., 11, 483–499,  2007</p>