Land cover mapping for Continental Africa

Creators: Midekisa, Alemayehu and Holl, Felix and Savory, David J. and Andrade-Pacheco, Ricardo and Gething, Peter and Bennett, Adam and Sturrock, Hugh J.W.
Title: Land cover mapping for Continental Africa
Item Type: Conference or Workshop Item
Event Title: ASTMH 65th Annual Meeting
Event Location: Georgia, Atlanta, USA
Event Dates: November 13-17, 2016
Date: November 2016
Divisions: Gesundheitsmanagement
Abstract: Land cover type influences transmission of a number of diseases, including vector-borne diseases such as malaria. However, high spatial resolution land cover data through time are lacking for continental Africa, hindering the ability to model and test hypotheses. The objective of this study was to develop a high spatial resolution (30 meter) land cover dataset for continental Africa for the years 2000 and 2015. To generate gold standard model data, high resolution satellite imagery was visually inspected and used to identify (7212 sample points) Landsat pixels that were entirely made up of 1 of 7 classes (water, impervious surface, high biomass, low biomass, rock, sand and bare soil). For model validation purposes, 80% of points from each class were used as training data, with 20% withheld as a validation dataset. Cloud free Landsat 7 and 8 annual composites for 2000 and 2015 were generated. Spectral bands from the Landsat image were then extracted for each of the training and validation points and a random forest model using the full dataset was used to classify the 2000 and 2015 Landsat images into each of the 7 classes. In addition to the Landsat spectral bands, spectral indices such as normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) were used as covariates in the model. Additionally, calibrated night time light imageries from the National Oceanic and Atmospheric Administration (NOAA) were included as a covariate. Using the validation dataset, classification accuracy including omission error and commission error were computed for each land cover class. Model results showed that overall accuracy of classification was over 90 percent. This high resolution land cover product developed for the continental Africa will be available for public use and can potentially enhance the ability to test models and hypotheses.
Citation:

Midekisa, Alemayehu and Holl, Felix and Savory, David J. and Andrade-Pacheco, Ricardo and Gething, Peter and Bennett, Adam and Sturrock, Hugh J.W. (2016) Land cover mapping for Continental Africa. In: ASTMH 65th Annual Meeting, November 13-17, 2016, Georgia, Atlanta, USA.

Actions (login required)

View Item View Item