A LOCAL - LEVEL SPATIAL MODEL OF MALARIA TRANSMISSION SITES IN RURAL ZAMBIA
Embargo until
Date
2018-12-19
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
The Johns Hopkins Malaria Research Institute (JHMRI) is working with local partners to develop strategies to eliminate malaria transmission around the town of Macha in Southern Province, Zambia. There are several studies being conducted to model and evaluate effective interventions to combat transmission and infection. Modeling local transmission and conducting geostatistical analysis of malaria transmission dynamics require location data for several types of sites: aquatic habitats potentially hosting Anopheles mosquito larvae and households where human hosts reside are both too numerous and expensive to manually geolocate. Thus, we sought to mine satellite imagery of Macha and the surrounding area for these sites of interest with GIS techniques and software.
We utilized ArcGIS software to conduct our analyses and develop our model and for this required two different approaches to harvest spatial coordinates for aquatic habitats and potential households because of the high-resolution/low-bands necessary to resolve small features. The classification stage of the Aquatic Habitats model consists of a Principle Components Analysis (PCA) followed by a binary supervised Maximum Likelihood Classification (MLC) using manually generated training samples. The classification stage of the Host Household model applied pansharpening of the multiband raster image with the panchromatic raster followed by a 25-class unsupervised ISODATA classifier, which reclassified on a binary raster based on analyst class selection of appropriate generated classes. Both models shared techniques for feature extraction once we had a binary raster, which consisted of a transformation from raster to polygon, and then filtering the polygonal shape based on empirical data. Following an accuracy assessment, another transformation of point features was done before geolocation of X and Y coordinates.
Our models resulted in enumeration and coordinates for 1,702 aquatic habitat locations and 27,548 host household locations. An accuracy assessment using at 350 km2 test region gave the Aquatic Habitat geolocation model an F1 score of 0.9181 and the Household geolocation model an F - score of 0.8547. We were able to map the observed household and aquatic habitat distribution and densities across the 2400 km2 study area as well as infer spatial relationship to potential malaria transmission sites from these data. These findings support the use of GIS software and spatial modeling to inform or evaluate interventions but external testing of this model on different areas with climate and population variation is necessary to validate our results.
Description
Keywords
Epidemiology, Malaria, Modeling, GIS, Zambia, Spatial