In order to reduce poverty we must first describe, explain and predict its spatial distribution over large areas with as high a level of local accuracy as possible. Poverty maps are traditionally produced by exploiting links between census (wide area) and survey (smaller area coverage) data. The detailed relationships found within the survey data are extended to the census data that must share some predictor variables in common with the survey data. Both census and survey data tend to be socio-economic in nature; the mapping thus exploits the internal correlations within potentially strongly correlated data sets - one 'measure' of poverty is often correlated with another.
Rather than look at the correlates of poverty, we should like to identify its causes. We suggest that poverty is multi-dimensional and that many of its dimensions are environmentally related; people are poor because they are unhealthy, or under-fed, or without access to fuel and water etc. Each of these is environmental in some way or other, and a correct approach to reducing poverty might be first to identify its (environmental) causes. We have attempted to do this with survey data from Uganda and environmental data derived from multi-temporal satellite imagery that measures land-surface conditions and processes (temperature, rainfall, vegetation growth etc.). The same satellite data have already been used to understand the distribution of farming systems throughout Africa and to predict the distribution and intensity of insect and tick carriers of a variety of diseases, and the incidence and prevalence of the diseases they transmit.
In this analysis therefore we examined to what extent satellite data (as a proxy for environmental conditions) are correlated with household survey data. Whilst correlation obviously does not automatically imply causation, we suggest an environmental approach is more likely to reveal causes than will the traditional approach of small area mapping using census and survey data.
However, it is first necessary to establish the relative predictive accuracies of the traditional and environmental approaches. The initial results from the environmental approach, described here, are promising, though we have not yet compared them directly to small area methods.
PPLPI, FAO, Rome, Italy, vi+60 pp.