Fully distributed hydrological models take into account the spatial variability of a catchment, and allow for assessing its hydrological response at virtually any location. However, these models can be time-consuming when it comes to model runtime and calibration, especially for large-scale catchments. Meanwhile, deep learning models have shown great potential in the field of hydrological modelling, but a multivariable, fully distributed hydrological deep learning model is still lacking. To address the aforementioned challenges associated with fully distributed models and deep learning models, we explore the possibility of developing a fully distributed multivariable deep learning model by using Graph Neural Networks (GNN), an extension of deep learning methods to non-Euclidean topologies. We develop a surrogate model of wflow_sbm, a fully distributed, physics-based hydrological model, by exploiting the similarities between its underlying functioning and GNNs. The GNN model uses the same input as wflow_sbm: gridded static parameters based on physical characteristics of the catchment and gridded dynamic meteorological forcings. The GNN model is trained to approximate wflow_sbm outputs, consisting of multiple gridded hydrological variables such as streamflow, actual evapotranspiration, subsurface flow, saturated and unsaturated groundwater storage, snow storage, and runoff. Our results show that the GNN model accurately predicts multiple hydrological variables in unseen catchments (median KGE=0.76), and can serve as an emulator of wflow_sbm with a shorter runtime. We furthermore demonstrate how the GNN model can function up to a prediction horizon of a full year, using physical system states to account for system memory, as well as a curriculum learning strategy combined with a multi-step ahead loss function during training. Overall, this study contributes to the field of fully distributed modelling using a deep learning approach.

In July 2021 severe flooding occurred in the South of Limburg, Belgium, and Germany due to heavy precipitation. Extreme precipitation events like this are expected to occur more often in our changing climate. Urbanization is thought to be another contributing factor to the July 2021 flood event. The Netherlands is expected to increase urbanization as a solution to its housing shortage. To make room for urbanization, while minimizing the effect of climate change, the government wants to make ”water and soil leading”, Water en Bodem Sturend, in the decision-making about the layout of the Netherlands.

Therefore, the goal of this research is to investigate the best suitable subcatchment for the construction of new residential houses within the Geul catchment, in terms of flooding. The July 2021 flood event is used as a reference. The first step was to investigate the hydrological response of the Geul catchment. Secondly, this hydrological response was modelled by the semi-distributed hydrological models HBV coupled to D-RR and by the distributed model Wflow_sbm. HBV and D-RR are set up in this research, while Wflow_sbm is adopted from Klein (2022) and Bouaziz (2022). The hydrological models are coupled to the Geul hydrodynamic model D-HYDRO of Hulsman, Weijers, Verstegen, and Goedbloed (2023). The building plans in the Geul catchment were investigated and scenarios were constructed. These scenarios were simulated in the hydrological models. This method resulted in a workflow that can be found in Idsinga (2024). The workflow can be applied on analyses of different land cover types.

The modelled hydrographs showed differences between the hydrological models. Each model better describes one part of the hydrological response compared to the other. HBV and D-RR better represent the subsurface flow and describe the hydrological response during consecutive precipitation events. Wflow_sbm represents the overland flow flux better and therefore describes the hydrological response during the July 2021 flood event. The modelled flood extents during the July 2021 flood event are also compared to the estimated extent by Slager, de Moel, and de Jong (2021). Wflow_sbm showed better similarity to the measured flood extent than HBV and D-RR. The Province of Limburg wants to build 18,730 new houses in the South of Limburg. This results in an increase of 6 km2 paved area. In this research, this increase is applied to different locations in the Geul catchment. Next, the impact of completely paved subcatchments was investigated. The relatively small 6 km² increase in paved area did not result in different discharge behaviour and the total area of the flood extent showed a small difference. However, it impacted the flooded paved area. Building far from the river on the hills resulted in no increase in the flooded paved area. New houses in the valleys, close to the river, are more exposed to flooding. In the Meerssen subcatchment, the added paved area was responsible for 95% of the total increase in the flooded paved area. This was also the case in the Gulp subcatchment, where about 90% of the increase in flooded paved area came from the added paved area.

The Meerssen subcatchment is the most vulnerable to flooding. This subcatchment contains the most paved area and more runoff will result in a more flooded paved area. A completely paved Gulp subcatchment results in a less flooded paved area than building 6 km2 close to the Geul in the Meerssen subcatchment. When the Belgians build new houses in the Sippenaeken subcatchment, the Netherlands will receive more water during an extreme event such as in July 2021.

The letter Water en Bodem Sturend states that new houses must be built in sensible locations. In this research, the location of new houses is found to be important for the hydrological response. Building close to the river results in a more flooded paved area than building far from the river. The Gulp subcatchment is the least vulnerable to flooding and can be considered the best building location for new houses among the three investigated subcatchments.

Climate change imposes an increasingly big challenge worldwide regarding floods and droughts. The Netherlands is no exception to this, as it has been increasingly hit by these phenomena in recent years, especially in the south of the country. The July 2021 flood in the Geul catchment intensified discussions on climate resilience. It prompts consideration of transforming the Dutch landscape into a more sponge-like system. Urban areas, identified as both problem areas and potential solutions within the catchment, stand out, since these areas are highly vulnerable and amplify climate change effects. Implementing Urban Nature-Based Solutions (UNBSs) emerges as a promising approach to address these challenges, potentially offering a solution to enhance climate resilience and mitigate the vulnerabilities of urban areas.
This research addressed the challenge of flood protection in the Geul catchment. It focuses on studying the impact of UNbSs and developing a methodology to select, model, and assess their performance in the catchment. This has been done by answering the following questions:

Which Urban Nature-based Solutions are suitable for the Geul catchment?
What is the hydrological effect of these Urban Nature-based Solutions locally?
What is the hydrological effect of Urban Nature-based Solutions at catchment scale?

This research followed a three-step workflow in line with the research questions. Firstly, the study focused on an assessment of neighbourhood types in the catchment and selecting suitable UNbSs for the Geul catchment. Based on this assessment, UNbS measures were chosen for their compatibility with these neighbourhoods. Next, this research analyzed the local effects of the implementation of these measures into the chosen neighbourhoods using the Climate Resilient City Tool. The final step involved an assessment of the hydrological impact of UNbSs on catchment scale. This was achieved by converting the previous results to wflow parameters.
The research succeeded in establishing a workflow for modeling UNbS impact at the catchment scale. It involved selecting UNbSs for three neighborhood types, resulting in the selection and implementation of green roofs, water roofs, permeable pavement, retention ponds, removing pavement to plant green and bioswales.
Locally, the study found that permeable pavement and bioswales were most effective for increasing storage capacity, evapotranspiration, and groundwater recharge. However, the overall order of magnitude for all measures remained consistent across neighbourhoods. Considering the total storage capacity increase, the order of magnitude was found to be between 34- and 39-mm storage equivalent over the total surface area of the neighbourhoods.
On a catchment scale, UNbS implementation resulted in a discharge reduction ranging from 1.71% to 3.10%, with a more pronounced effect upstream. Three neighbourhood scenarios exhibited minimal differences, all below 0.1%, which is considered insignificant. Absolute discharge reduction consistently followed patterns across low and high discharge periods, with more substantial reductions during peak discharges. However, the percentage difference between original and altered discharge was found to be similar across all periods, making this trend less evident.
Based on these results, the research proposed the following five recommendations regarding future research and implementation: 1) improve model transparency and sensitivity analysis, 2) consider practical implementation challenges, 3) refine typology mapping and data, 4) evaluate Nature-based Solutions in diverse landscapes and 5) promote the role of UNbSs in climate resilience.