DiluviumDEM: Enhanced accuracy in global coastal digital elevation models
Dominick Dusseau, Zachary Zobel, Christopher R. Schwalm
Abstract
Accurate representation of global coastal topography is essential for numerous scientific disciplines, coastal management, and disaster risk assessment. Even with recent improvements to existing global digital elevation models (DEMs), high and persistent errors in these DEMs result in significant uncertainty when analyzing coastal processes. This results in low confidence for current sea level rise inundation risk assessments. We present DiluviumDEM, the first global DEM appropriate (i.e., the root mean square error (RMSE) is half the total water elevation in 2100 under a specific scenario) for mapping sea level rise inundation under the IPCC SSP2–4.5 and SSP5–8.5 scenarios, with an estimated RMSE of 1.13 m for coastal areas with elevations less than 2 m above mean sea level. Out of ten countries used for validation, DiluviumDEM has the lowest RMSE compared to three other DEMs analyzed, the lowest mean absolute error (MAE) for eight, and the mean error (ME) closest to zero for six. By reducing the error of the European Space Agency's Copernicus DEM using a gradient boosted decision tree model, we have created a new global coastal DEM with up to twice the accuracy compared to other global DEMs.