Spatio-temporal mapping of groundwater-related flooding using two methods: piezometric levels and Sentinel-2 based remote sensing

The purpose of this study is to assess the efficacy and validity of using piezometric data and remotely sensed data to spatially and temporally map groundwater-related flooding, using Nouakchott, Mauritania as a case study.

Despite a warm and dry climate, the city of Nouakchott in Mauritania has been experiencing constant flooding for nearly a decade, making portions of the city inhabitable and posing long-term health and socio-economic threats. During the rainy season, a combination of factors has led to the increasing frequency and duration of flooding events, including a shallow groundwater table, limitations of the domestic water system, reduced infiltration caused by rapid urbanization, and climate change.

The goal of the study is to better understand and quantify the extent of flooding in the developed areas of Nouakchott, both in space and in time, and to relate this flooding to seasonal and annual fluctuations in precipitation and hydrogeological conditions. To do this, we estimate the presence of flooding from two different perspectives: (1) by analyzing the piezometric levels from a network of 23 piezometers and comparing the interpolated piezometric surfaces to the topographic elevations, and (2) by using Sentinel-2 multi-spectral satellite imagery and machine learning with in-situ training data to identify pixels that are classified as flooded. Flooded area maps are then developed using these two methods for days with available data within the period of record (since 2015 for both data sources). These results are then used to develop a time series of flooded areas for both methods, allowing for comparison and potential validation of the results with each other and with the available in-situ data and observations. Preliminary results show that the piezometric analysis was sensitive to the topographic information and underestimated the flooded area compared to the remote sensing analysis. The remote sensing analysis showed satisfactory accuracy when compared to validation data but does not provide as detailed of information on the hydrogeological dynamics as the piezometric analysis. These findings demonstrate the complementarity of using both methods in tandem.

This estimation of groundwater-related flooding extents and seasonal variability was useful to better understand the relationships between the flooding dynamics and climatic factors, to identify vulnerable areas and communities, and to calibrate hydrogeological modeling. Additionally, this novel and open-source approach can produce critical data for flood risk assessment and planning in under-monitored and data-poor areas, mitigation scenario development, and urban management strategies. Next steps for the project include further linking the two methods by developing a piezometric record from the flooding information obtained from the remote sensing analysis using the temporal change in flooding extents and known topographic information.