Informed disaggregation

Parroquias (n = 991) represented in this data set range in size from roughly 2 km² to over 8,000 km². Therefore, we felt it prudent to reduce this high spatial variation prior to analyses. To correct for this extreme variation in the spatial resolution of aggregated presence data, the number of positive LI locations in a given parroquia were reassigned from the centroid of the administrative boundary to the centroids of cities, barrios (neighborhoods), and villages where MSP mosquito surveillance was conducted, ≤ 5km in urban extent. Human settlements were identified via a combination of OpenStreetMap (http://openstreetmap.org) and Google Earth (http://earth.google.com) satellite imagery in ArcMap. While satellite images were used to identify population dense areas, guiding disaggregation of LI data, this imagery was not used in mapping or creation of figures. Given the potential for uncertainty, a conservative approach to disaggregation was taken, where occurrence records were not included in the final dataset in cases of spatial ambiguity (e.g. cities larger than 5km in extent with a single occurrence record, multiple developments in an administrative unit exceeding the number of surveys conducted, etc). This method of informed disaggregation allowed for better spatial representation and improved model performance compared to ENMs built with aggregated data, without compromising de-identification (S1 Table).