A novel approach to mapping flooding extent in the Chobe River Basin from 2014 to 2016 using a training library

Abstract

The Chobe River Basin (CRB) is a flood-dependent ecosystem that relies on seasonal floods from the Zambezi and Linyanti Rivers. These flood pulses provide water for the flood recession agriculture in the region, water for the fishing grounds around Lake Liambezi, and nutrients for the vegetation in the CRB. Recent years have shown an increase in the magnitude of flooding, which could have consequences on the region’s biodiversity and the people living in the CRB. The goal of this study is to develop a classification framework based on a training library and time-windows to use in classifying the extent of flooding in the CRB. MODIS MOD09A1 satellite imagery served as the satellite imagery. Bands one through seven were converted into the tasseled cap transformation to serve as the feature selection. The study period, from February to July, is broken down into three time-windows. The time-windows are used because the land covers in the CRB go through significant spectral changes during the study period and the three time-windows seek to improve the classification accuracy. The classification methods include maximum likelihood classifier (MLC), decision trees (DT), and support vector machines (SVMs). The results show that DT and SVMs provide the highest overall accuracy and kappa values over MLC. Classification using the time-window method was statistically significant when comparing kappa values and visually, images classified using the correct training library for a time-window displayed higher agreement with the reference data. Flooding extent was high for 2014 but low in 2015 and 2016, indicating a decreasing trend. DTs provided better inundation maximums compared to SVMs and therefore is the reason that DT are the best classification technique. The results will provide planners with information regarding the extent of flooding in the CRB and where waterborne diseases occur in the region. A new classification technique is also developed for the remote sensing literature.