Comparing Object Based and Random Forest classification models to estimate the rural human footprint in southern Africa

Authors: Ariel Weaver*, University of Louisville
Topics: Remote Sensing, Applied Geography, Africa
Keywords: africa, southern africa, conservation, development, human footprint, object based image analysis, remote sensing, random forest, classification models, image processing
Session Type: Paper
Day: 4/12/2018
Start / End Time: 1:20 PM / 3:00 PM
Room: Lafayette, Marriott, River Tower Elevators, 41st Floor
Presentation File: No File Uploaded

Rapid population growth and the resulting expansion of human footprint are impacting biogeochemical processes occurring at varying spatial and temporal scales across the globe. Remote sensing provides a tool to quantify some characterization of the human footprint appropriate for various applications for environmental research. For this study, we model the rural human footprint (RHF) in the context of rural conservation and development initiatives in a southern African savanna, characterized by a highly heterogeneous mixture of trees, grasses, and shrubs that share overlapping spectral signatures with built settlement areas and agriculture. The spectral overlap makes traditional pixel based methods less effective at separating land cover classes. We look particularly at the eastern Zambezi Region of Namibia, a region that is currently balancing rural population growth and settlement/agricultural expansion with conservation and development initiatives. The extent of the RHF in the eastern Zambezi Region has implications in the connectivity of wildlife corridors and human-wildlife conflicts as it central to the largest terrestrial conservation area in the world, the Kavango-Zambezi Trans-Frontier Conservation Area. We present an analysis of the RHF for the eastern Zambezi Region, utilizing three methods of image analysis that move beyond pixel based methods, including: 1) Object Based Image Analysis, 2) Random Forest, and 3) a mixed Object-based /Random Forest approach to classify the RHF in this primarily rural region, using Landsat 8 imagery from the 2017 dry season. We compare these three methods and discuss the relevant application of the output for conservation and development initiatives in the region.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login