Creative Commons License This work is licensed under a Creative Commons Attribution 4.0 International License.


Accounting for Spatial Non-Stationarity to Estimate Population Distribution Using Land Use/Cover. Case Study: the Lake Naivasha Basin, Kenya

Dawit W. MULATU1, Anne VAN DER VEEN1, Robert BECHT1, Pieter R. VAN OEL1, Desta J. BEKALO1
1 University of Twente, ITC-Faculty of Geo-Information Science and Earth Observation, Enschede, THE NETHERLANDS
Pages: 33-44

Abstract. Remotely-sensed data can be used to overcome deficiencies in data availability in poorly monitored regions. Reliable estimates of human population densities at different spatial levels are often lacking in developing countries. This study explores the applicability of a geographically-weighted regression (GWR) model for estimating population densities in rural Africa using land use/cover data that have been derived from remote-sensing while accounting for spatial non-stationarity. This study was conducted for the Lake Naivasha basin in Kenya where population pressure, intense land utilization in the catchment and informal settlements in Naivasha town due to lucrative economic activities are the major challenges of the basin socio-ecological system. The results of this study show that using a GWR model for taking into account the spatially-varying relationship between specific land use/cover classes and population significantly improves population estimates and handles the spatial non-stationarity that could not be addressed by global ordinary least squares (OLS) model. The result revealed that the parameter estimates (coefficients) for grassland and cropland use/cover have a significant spatially varying relationship with population and exhibit locally different signs, which would have gone undetected by a global model. Consequently, this study indicates that incorporating spatial non-stationarity can significantly improve population density estimates for rural Africa based on remotely-sensed data.

K e y w o r d s: land use/cover, remotely-sensed data, population density, non-stationarity, Lake Naivasha basin, geographically weighted regression (GWR), ordinary least squares (OLS)