Predicting poverty with vegetation index

Abstract

Accurate and timely predictions of the poverty status of communities in developing countries are critical to policymakers. Previous work has applied convolutional neural networks (CNNs) to high‐resolution satellite imagery to perform community‐level poverty prediction. Although promising, such imagery has limitations in predicting poverty among poor communities. We provide the first evidence that a publicly available, moderate‐resolution vegetation index (the normalized difference vegetation index [NDVI]), can be used with CNNs to produce accurate poverty predictions contemporaneously among poor communities heavily dependent on agriculture. We also show that the NDVI can effectively detect consumption variation over time. To our knowledge, this is the first attempt to use remote sensing data to predict future‐period consumption expenditure at the community level.

Predicting poverty and malnutrition for targeting, mapping, monitoring, and early warning

“…While Jean et al (2016) find success in predicting asset wealth, they find less success predicting consumption expenditures. This finding is echoed by findings in Barriga Cabanillas et al (2021) and Tang et al (2021); overall, stock measures appear easier to predict than flow measures.…”

“…Recent advances in both data and methods have enabled production of high‐quality subnational maps with tolerably accurate estimates of current and probable poverty and malnutrition conditions (Jean et al, 2016; Yeh et al, 2020). Likewise, poverty and malnutrition forecasting for the purpose of early warning is making incremental gains (Browne et al, 2021; Tang et al, 2021; Yeh et al, 2020). While development and humanitarian programming are enjoying a machine learning and big data revolution, there exist real risks that these new data series and methods underdeliver on the promise of improving targeting, mapping, monitoring, and early warning.…”

How can new technology support better measurement of extreme poverty?

“…Box 2: The 'transfer learning' process The performance of this methodology has been assessed using publicly available images from Google Static Maps for Uganda (Xie et al, 2016) and five African countries (Jean et al, 2016), as well as for the whole of Africa using public, and more frequently updated Landsat 7 satellite imagery (Perez et al, 2017). Tang et al (2018) show that features of similar predictive power can also be extracted from publicly available moderate-resolution maps of vegetation index (NDVI), with additional advantages for the dynamic updating of poverty indicators in rural areas, thanks to the higher sensitivity of NDVI for the measurement of change, as well as in countries with a heavy dependence on agriculture.…”

“…Although ground truth estimates would be required to formally test this assumption, it is expected that some outcomes and geographical levels are more robust than others to this kind of misspecification. For initial assessments of the use of repeated remote sensing data for continuous monitoring see Bansal et al (2020), Tang et al (2018), and Bennett and Smith (2017).…”

A review of explainable AI in the satellite data, deep machine learning, and human poverty domain

“… Tang et al. 62 2022 First, a VGG-16 CNN, pre-trained on ImageNet and fine-tuned on night time light intensities. Second, a RF model built on the fine-tuned features.…”