Estimation of Flood Water Levels by Merging DTM, Satellite Imagery and Hydraulics Laws through AI
Puech, C1; Hostache, R2; Raclot, D3
1Cemagref; 2CRGL; 3IRD
Merging DTM and flood imagery can be used for water levels estimation. A simple process generates independent estimations, spatially non-uniform over the plain and often of poor quality . To obtain an accuracy acceptable for hydraulic modelling, we propose a methodology in two main steps: (1) a remote sensing step : definition of a merging procedure for DTMs and flood imagery to obtain a confidence interval of independent estimates. Relevant locations are selected, and provide an adequate interval of water levels (Min, Max). (2) a dependence step to reduce the uncertainties using hydraulics laws through an AI constraining procedure. The constraining procedure relates independent estimates applying a flow scheme over the flood plain. Successive water levels are assumed to decrease along the flow direction giving a system of numerical constraints solved by AI techniques. The results show both a strong decrease in the Min Max interval width and a complementation of the flood depth estimates.
Whis this procedure the independent initial Min Max intervals range from 0.80 m for aerial photography to 1.60 m for satellite radar images. After application of the constraining procedure, the mean interval decreases to respectively 0.40 and 0.80 m. In fact this interval represents the envelop that includes the ‘true’ value of water levels. The real accuracy is better : a validation conducted on the Alzette River (Luxemburg) with field-recorded flood levels provides a RMS of 0.13 m.