A geographical sampling strategy for field surveys in an urban area using high-resolution satellite imagery

Abstract

Field evaluations for studying the epidemiology of vector-borne diseases like dengue in urban areas are often restricted to selection of households and buildings for field surveys. Therefore, the resulting sampling frame may exclude specific locations within the urban environment that contain vector habitats and thus may bias the results. A sampling strategy was developed for field surveys in an urban area using high-resolution satellite imagery. The site selected was Puntarenas, a city affected by dengue on the Pacific coast of Costa Rica, for which high-resolution satellite imagery was available from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER, 15 m spatial resolution) and QuickBird (0.6 m and 2.4 m spatial resolution for panchromatic and multispectral bands, respectively). Grids obtained from the ASTER imagery and a cover map generated from the QuickBird multispectral bands were used to determine the optimal grid area of 10 000 m2 , which contain 136 houses. A final grid 42 by 42 pixels (100.8 x 100.8 m) was created using the multispectral Quickbird imagery, and cells that had an area less than 90% within one specific locality of Puntarenas were excluded. The remaining cells were grouped according to locality and a random sample (10%) was selected from each. This sample of cells would be used for field data collection on specific mosquito larval habitats by evaluating the entire area within the geographical limits of each cell. To assess the suitability of the selected grid cells, the proportion of tree area (“tree” class Kappa = 0.91) was extracted for the individual cells from the QuickBird cover map. The mean percentage of tree cover in each locality and total area was compared between the selected sample cells and the total cells of the Puntarenas image. Overall, the sample adequately represented the total area and most of the individual localities in terms of tree cover. In 8 of 10 localities the difference between the estimate (sample) and the real percentage of tree cover was less than 3%. These results show that high-resolution satellite imagery and geographical information systems are useful in evaluating urban areas and randomly selecting sections for field data collection on mosquito larval habitats that are practical, representative, and will reduce bias.