alexandriaberry-ops/mira-madagascar

What can we observe in diet using remotely sensed climate data? # MIRA Madagascar — Vegetation Indices and Household Food Consumption Data and analysis code accompanying the MIRA study of southern Madagascar, relating satellite vegetation indices (EVI and NDVI) to household food consumption at the commune level. The analysis reduces per-item household consumption records to a single dominant dimension of dietary variation via mean-centered singular value decomposition, then tests that component against commune-level vegetation greenness over time. > *Recurrent droughts and food shortages in southern Madagascar underscore the need for real-time food security monitoring. Since 2018, a monthly survey of 602 households has tracked dietary coping strategies in response to food shocks; 547 meeting a 70% response-rate threshold are included in our analysis. Here we assess whether satellite-measured vegetation greenness can track variation in household-level food consumption. Greenness does not predict the four commonly used food security indices in the survey. It is instead significantly associated with the leading mode of dietary variability, which captures substitution between staple or aid-supported foods and lower-preference fallback foods eaten during food stress. Projected onto commune-level monthly patterns, this mode explains an average of 43% of within-commune dietary variability. Greenness represents it skillfully, explaining 32–58% of its variability, and 38–70% once seasonality is included. Higher-order dietary modes show no such relationship, and rotated decompositions, lagged predictors, and alternative vegetation indices do not improve performance. Satellite-measured greenness therefore captures primary dietary shifts, supporting its integration into more-detailed food security monitoring systems. DOI added once available. ## Repository layout ``` data/ commune_food_monthly.csv mean food consumption by commune-month evi_timeseries/ commune-mean EVI series, c1-c8 commune_vecs/ mont …