Comparison of surface roughness derived from different remote sensing sources for flood simulations

The occurrence of floods and the movement of water heavily influenced by the roughness of the terrain surface. This is the most important parameter affecting overland flow patterns after topography. The uncertainty of the roughness parameter affects the hydrographic properties, causing errors in water level estimation. Therefore, Hydraulic models should parameterize the effect of roughness using hydraulic friction coefficients such as Manning or Chézy's C, which define the resistance of channel and the flood water flow. Using remote sensing methods, creating an effective roughness map requires some knowledge of the land cover in the floodplain and this depends on the availability of spatial data with sufficient detail. Land cover / land use maps are used to estimate the roughness values to show the most realistic flood propagation patterns. High-resolution land cover data can help to predict more accurate roughness, making it the most realistic hydraulic modelling. This study aims to determine suitability for use in flood modelling and how the roughness maps obtained from different remote sensing images show a variation pattern, compare these maps with each other and finally determine the most suitable remote sensing image which can be used for roughness maps which is one of the parameters of flood modelling. In this study; land cover/land use maps were obtained from Landsat 8 OLI and Sentinel-2 remote sensing images using Random Forest algorithm. It was mapped surface roughness values using these data's and CORINE-2018 land cover data. These roughness maps were compared with the roughness map obtained from the DigitalGlobe high resolution image which was used as reference data. According to the results, spectral resolution as well as spatial resolution of data type, have also been effective in producing a more realistic roughness map. While roughness maps produced using Sentinel-2 10 meters bands are expected to be more realistic, it is observed that the roughness maps produced from Sentinel-2 20 meters and Landsat-8 OLI 30 meters resolution are closer to the high resolution DigitalGlobe data. This result; it is closely related with the bandwidth and the number of bands used. The greater the spacing and number of bands used affects more accurate; land cover/land use, which in turn affects the accuracy of roughness maps.