An evaluation of the Height Above Nearest Drainage (HAND) flood mapping methodology with the implementation of multivariant linear regression algorithm

Abstract

An evaluation for the Height Above Nearest Drainage (HAND) flood mapping methodology is presented for three main rivers in Massachusetts and their tributaries. The main rivers are Charles River, Neponset River and Mystic River. We implemented a Multivariant Linear Regression algorithm to predict discharges for some of the rivers' tributaries due to the lack of discharge data from the Flood Insurance Study (FIS) reports for these reaches.The HAND methodology is a low complexity, terrain based flood mapping method that uses a Digital Elevation Model (DEM), discharge-stage relationship (Rating Curves) and stream flow data. The 100-year discharge data was extracted from FEMA's FIS reports for each of the three main rivers and some of their tributaries. The rating curves were used for all reaches to determine the water levels (stages). Based on the DEM for the region, we were able to determine the inundated (wet) cell from the non-inundated cells (dry) for the areas around all reaches using ArcGIS and a Python script to generate the flood maps. FEMA's 100-year flood maps were used in this study as our reference to evaluate the accuracy and the efficacy of the HAND method. The results show that the HAND method tends to underpredict some flooded areas around low order reaches and overpredicts the flooded area around very high order reaches that have high discharges. For medium order reaches, the flooded areas were almost identical to FEMA's flood maps. The implementation of the Multivariant Linear Regression algorithm resulted in a HAND flood map with an accuracy close to the accuracy of the original HAND flood map. That implies that the algorithm is an effective and accurate tool to use for predicting river discharge.--Author's abstract