Conference Agenda

Automatic mapping of land cover types on flooded Riparian landscapes is one of the most challenging problems in remote sensing. Although Object based approach has been embraced in Land Use Land cover studies few studies have applied it on riparian vegetation discrimination. In order to improve on the accuracy and efficacy of any classification, new approach to data collection and extraction has become increasingly necessary. The study investigated an Object based classification based on Sentinel 1 Single Look Complex (Synthetic Aperture Radar) data using three classifiers namely Naıve Bayes, Decision Tree and Random Forest. Four land cover types i.e. Acacia Forest, Built-up, Grasslands, water and others were successfully retrieved. Change detection was noted on the North western strands having reduced from an area of 212.7 hectares in 2008 before the floods to an area of 64.64 hectares in 2019 after floods. ALOS-1 Level 1.1 was used as reference image captured in 2008 before floods and Sentinel 1 SLC data captured in 2015, 2017 and 2019 after the floods, Acacia Forest strands were captured to have been degraded especially on the North western part of Lake Nakuru.Classification Results based on Machine learning and Polarimetric Matrix generated bands C11, C22 and their ratio were used to obtain the results as follows; Naive Bayes 91.1% , Decision Tree 94.1% while Random Forest took the lead by 94.4%. One-way Analysis of Variance (ANOVA) was used to compare variations among the group algorithms. However, there was no significant difference between and amongst the Classifier performance since F,2,15=0.529 with a p-value of 0.60 was achieved

Index Terms—Synthetic Aperture Radar, Object Detection, Machine Learning, Change Detection, flooding