What Aspects of Food Security in Southern Madagascar can be Informed by Satellite-based Remote Sensing Data?

Ongoing drought and famine in Southern Madagascar makes up-to-date monitoring of food security critical. An existing monitoring strategy, initiated in 2018, involves surveying ~600 Southern Madagascar households on a monthly basis. Created by Catholic Relief Services, the Monthly Interval Resilience Analysis (MIRA) platform consists of a questionnaire administered to the same households each month to improve understanding of coping mechanisms in response to shocks and food security at the household level. Collecting this data at high frequency, rather than annually or even less frequently as is typically done, permits better understanding of how seasonality may impact food security. MIRA incorporates the standardized questions for food security indices and coping mechanisms such as the Household Hunger Scale (HHS) , Food Consumption Score (FCS), and the Reduced Coping Strategies Index (RCSI). Additionally, each household reports on relevant economic factors including income sources, types of livestock owned, types of crops harvested, sources of food in the last 24 hours, and shocks. Owing to the need to compile and transmit results, these are generally available within a few days after each survey.

In this study we explore the potential to use remotely sensed observations of crop and pasture lands to provide more frequent and advance indication of food security for the 602 surveyed households. Specifically, we develop a linear mixed effects model that uses remotely sensed Normalized Difference Vegetation Index (NDVI) to predict household food security indices (e.g., the Food Consumption Score). Households are grouped according to k-means clustering on geographic location. NDVI over the growing season is shown to provide moderate predictive skill for predicting household FCS in the ensuing lean season. Residuals, however, show a heteroscedastic distribution, suggesting that further model development is required, and we explore the utility of distinguishing between cropland and pastures, as well as including additional socioeconomic indicators.