Combining satellite imagery and machine learning to predict poverty

Abstract

Reliable data on economic livelihoods remain scarce in the developing world, hampering efforts to study these outcomes and to design policies that improve them. Here we demonstrate an accurate, inexpensive, and scalable method for estimating consumption expenditure and asset wealth from high-resolution satellite imagery. Using survey and satellite data from five African countries--Nigeria, Tanzania, Uganda, Malawi, and Rwanda--we show how a convolutional neural network can be trained to identify image features that can explain up to 75% of the variation in local-level economic outcomes. Our method, which requires only publicly available data, could transform efforts to track and target poverty in developing countries. It also demonstrates how powerful machine learning techniques can be applied in a setting with limited training data, suggesting broad potential application across many scientific domains.

A Crystal Ball for Forests?: Analyzing the Social-Ecological Impacts of Forest Conservation and Management over the Long Term

“…Many of these data sets cover comparatively long time periods and are now widely available for analysis. Technological changes have had less infl uence on the collection of socioeconomic data, though recent studies (e.g., Jean et al 2016) suggest this situation is changing rapidly. A second reason for the preponderance of ecologically focused studies is the higher cost of collecting household data in far-fl ung areas of developing countries.…”

The Material Stock–Flow–Service Nexus: A New Approach for Tackling the Decoupling Conundrum

“…Indicators of material stocks as well as material stock maps could help elucidating the importance of specific qualities of material stocks respectively their spatial patterns for resource demand, waste production and closing material loops. Mapping of human activities is so far largely restricted to population density maps and other datasets mainly derived from proxies such as nighttime lights [97,127,150]. Maps of material stocks would be useful to characterize spatial patterns of material stocks and thereby analyze stock-flow-service relations and the spatial distribution of human activities on earth much more consistently than is possible today.…”

“…Population maps [122,123] are widely used to downscale national or regional data to maps, following the assumption that human drivers or impacts scale with population [124,125]. Such maps are constructed by combining census statistics with proxies from remote sensing, e.g., nighttime light [126,127]. Similar approaches allowed mapping the global distribution and environmental impacts of livestock [128][129][130].…”

Priorities to Advance Monitoring of Ecosystem Services Using Earth Observation

“…While being governed by multiple, interacting factors, demand for both subsistence and commercial NTFPs is generally a function of population density and distribution, household characteristics, accessibility to NTFP harvest areas and markets, institutions and regulations, and NTFP preferences and values [47,54]. Population density and economic well-being [40,41] as well as roads and fluvial networks [55] can be extracted from satellite imagery in some cases, but information on the other factors fully relies on household surveys as well as other economic and socio-cultural data.…”

“…Second, satellite data has been used to parameterize (as input, initial conditions or variables) or to validate spatially-explicit, process-based models of ecosystem service supply (e.g., using MODIS-Leaf Area Index to simulate plant growth in the Soil and Water Assessment Tool, [39]). Third, though much more rarely done, a few studies have used Earth observation data to estimate the location, size and economic well-being of communities as potential beneficiaries (e.g., using satellite night lights, high-resolution optical or radar data; [40,41]) or to map the demand for specific ecosystem services (e.g., pollination-dependent crops, [42]). Finally, by monitoring land use change activities Earth observation has been applied to evaluate the effectiveness of ecosystem service intervention or incentives programs (e.g., Payments for Ecosystem Services -PES, [43]).…”