Analysis of the added value of remote sensing soil moisture products for near-future inundation risk assessment in North Brabant, the Netherlands.

Abstract

Information products that are based on remote sensing data are increasingly being used for operational and strategic water management in the Netherlands. The Dutch regional water authority Brabantse Delta (Waterschap Brabantse Delta) is interested in using data products that map the current state of soil moisture content based on remote sensing data. This study aimed to list the currently available soil moisture products for the Netherlands and make analyses on the reliability and the accuracy of those products for the study area in Noord-Brabant, the Netherlands. Furthermore, it was analyzed what the relationships was between the amount of precipitation, soil moisture content, groundwater levels, and discharges, in order to find the most important factor in inundation risk assessment. In order to incorporate remote sensing soil moisture products in operational water management, there should be a clear understanding of how representative those products are. Two soil moisture data products have been analyzed in detail, the OWASIS product by HydroLogic and VanderSat soil moisture (by VanderSat), at 250m and 100m resolution per cell, respectively. The products were statistically compared to in-situ measurements at 15-18 sites and at depths varying between 5 and 20 cm below surface level. This was done at various spatial scales of the data product: one cell, 3x3 cells and at the scale of the full Raam catchment. OWASIS showed a very weak Pearson correlation of R=0.29 (single cell site average), 0.31 (3x3-grid of cells), 0.18 (catchment scale). It was found that OWASIS models the soil moisture in the rootzone rather than the full unsaturated zone. OWASIS does not report the (dynamic) depth of the rootzone, rendering the data unusable for inundation risk assessment. The VanderSat product showed a moderate to strong correlation with in-situ measurements: R=0.66 (single cell site average) and R=0.82 (catchment scale). The mean absolute error for VanderSat compared to in-situ measurements was 0.068 [m³/m³] (single cell) and 0.031 [m³/m³] (catchment scale). The spatial and temporal variability in soil moisture content among individual sites was much higher in-situ compared to the VanderSat product, the VanderSat product is particularly smooth in spatial distribution compared to the much more dynamic observations at the in-situ sites. The basic rainfall-runoff analysis showed a clear gradient from low to high discharge when transitioning from (relatively) deep to medium to shallow groundwater levels; while for the soil moisture content the transition from medium to wet conditions showed more of a mixed nature. This indicates that for precipitation events in the study area the discharge is generally stronger correlated to groundwater depth than to soil moisture content. Though, in more rare cases soil moisture conditions were more important than groundwater levels, as was shown by a small number of events in which the groundwater level was relatively close to the surface, while discharge remained low due to dry soil moisture conditions. The study concluded that the investigated soil moisture products are currently an unfeasible option to improve the assessment of near-future inundation risk for the province of Noord-Brabant, because the available products are either unreliable or too smooth to provide accurate soil moisture content estimates at cell scale (100 - 250 m). The VanderSat product showed some potential to be used at catchment scale (approx. 100 km²), but it depends on the application whether this is sufficient for operational use. Due to the fact that soil moisture conditions were generally shown to be less important in (high) runoff generation than the level of the groundwater, it would be advisable to focus on other options than remote sensing soil moisture content for inundation risk assessment.