Search

There is increasing interest in leveraging Earth Observation (EO) and geospatial data to predict and map aspects of socioeconomic conditions to support survey and census activities. This is particularly relevant for the frequent monitoring required to assess progress towards the UNs' Sustainable Development Goals (SDGs). The Sundarban Biosphere Reserve (SBR) is a region of international ecological importance, containing the Indian portion of the world's largest mangrove forest. The region is densely populated and home to over 4.4 million people, many living in chronic poverty with a strong dependence on nature-based rural livelihoods. Such livelihoods are vulnerable to frequent natural hazards including cyclone landfall and storm surges. In this study we examine associations between environmental variables derived from EO and geospatial data with a village level multidimensional poverty metric using random forest machine learning, to provide evidence in support of policy formulation in the field of poverty reduction. We find that environmental variables can predict up to 78% of the relative distribution of the poorest villages within the SBR. Exposure to cyclone hazard was the most important variable for prediction of poverty. The poorest villages were associated with relatively small areas of rural settlement (<∼30%), large areas of agricultural land (>∼50%) and moderate to high cyclone hazard. The poorest villages were also associated with less productive agricultural land than the wealthiest. Analysis suggests villages with access to more diverse livelihood options, and a smaller dependence on agriculture may be more resilient to cyclone hazard. This study contributes to the understanding of poverty-environment dynamics within Low-and middle-income countries and the associations found can inform policy linked to socio-environmental scenarios within the SBR and potentially support monitoring of work towards SDG1 (No Poverty) across the region. • Earth Observation and geospatial data could support monitoring of development. • We examine association between poverty and the environment with machine learning. • Cyclone hazard was the most important variable for prediction of poverty in the SBR. • The poorest villages were predicted with up to 78% accuracy. • Results can inform policy linked to socio-environmental scenarios in the SBR. [ABSTRACT FROM AUTHOR]