How do communities in lower-income countries adapt to wildfire smoke? We combine GPS data from 13 million smartphones with modeled pollution exposure across 10,000 Indonesian villages during the 2019 fire season to study behavioral responses to wildfire-attributable PM2.5. Smoke reduces out-of-home activity overall, but responses diverge sharply by income: wealthier areas show patterns consistent with self-protective avoidance — including increased hotel stays and long-distance travel — while in poorer villages, home visits rise without corresponding declines elsewhere, suggesting involuntary displacement from informal and outdoor work rather than active adaptation. Respiratory healthcare visits increase with exposure, with the burden falling disproportionately on rural and poorer populations.

Agricultural commodity booms can cause violent social conflict by creating incentives to transform land use, a channel distinct from the income shocks emphasized in the existing literature. Studying Indonesia's oil palm boom, we use agricultural suitability and available land to measure local incentives for plantation expansion. This expansion pressure causes conflicts, particularly over resources and during local elections. These effects are amplified where land is scarce and more contestable. Separately, economic shocks increase conflict on existing plantations. Our findings show that agricultural booms can destabilize communities through competition over future rents and increased vulnerability to economic shocks.

Evaluating reforestation programs at the plot level is difficult because sub-meter satellite imagery is expensive and sparsely available, while cheaper imagery is too coarse. We develop a two-stage machine learning framework that trains on limited very-high-resolution imagery and transfers the learned mapping to widely available 5 m basemaps, producing reliable plot-level tree-cover estimates at a fraction of the data cost. Validation and sensitivity analyses indicate that the framework yields reliable plot-level estimates while outperforming common methods for reforestation monitoring. Applying this to Madagascar's PLAE afforestation program (2014–2019) with a matching design, we find that tree cover on planted plots is 4.7–12.5 percentage points higher than on matched controls 5–8 years after planting.

Does land rights formalization make reforestation programs more effective? Studying Madagascar's PLAE smallholder restoration initiative (2014–2019), we combine primary household surveys with geospatial data and exploit variation in overlapping colonial and post-colonial land titles that still impede land formalization today. Successful land certification significantly increases tree growth on reforested plots, in both self-reported and remotely sensed measures. It also raises reported tree theft but leaves perceived tenure security, charcoal income, and broader welfare outcomes unchanged — highlighting both the potential and the limits of land formalization in weak institutional and market environments.