Multi-sensor mapping of honey bee habitats andfragmentation in agro-ecological landscapes inEastern Kenya

Abstract

Extensive land transformation leads to habitat loss, which directlyaffects and fragments species habitats. Such land transformationscan adversely affect fodder availability for bees and thus colonystrength with consequences for rural communities that use beekeeping as a livelihood option. Quantification of the landscapestructure is thus critical if the linkages between the landscapeand honey bee colony health are to be well understood. In thisstudy, a random forest algorithm was used on dual-polarizedmulti-season Sentinel-1A (S1) synthetic aperture radar (SAR) andsingle season Sentinel-2A (S2) optical imagery to map honey beehabitats and their degree of fragmentation in a heterogeneousagro-ecological landscape in eastern Kenya. The dry season S2optical imagery was fused with the S1 data and class-wise map-ping accuracies (with and without radar) were compared.Relevant fragmentation indices representing patch sizes, isolationand configuration were thereafter generated using the fusedimagery. The fused imagery recorded an overall accuracy of 86%with a kappa of 0.83 versus the SAR imagery only, which had anoverall accuracy of 76% with a kappa of 0.68. However, the S1imagery had slightly higher user’s and producer’s accuracies forunder-represented but important honey bee habitat classes, thatis, natural grasslands and hedges. The variable importance ana-lysis using the fused imagery showed that the short-wave infraredand the red-edge waveband regions were highly relevant for theclassification model. Our mapping approach showed that fusingdata generated from S1 and S2 with improved spectral resolution,could be effectively used for the spatially explicit mapping ofhoney bee habitats and their degree of fragmentation in semi-arid African agro-ecological landscapes.