Concern vehicle in the rainy season in Kimiti, Chad. Photo by Anastasia Marshak, Tufts University
Droughts are frequent events in Chad. Early warning is key to ensure a timely response to drought to protect lives and livelihoods before they are threatened. To that end, since 2012, the Feinstein International Center has been working with Concern Worldwide to develop and test a model that uses local and historical rainfall in Chad’s Kimiti department to predict future crop production and hence food security.
Our modeling approach borrows from the field of machine learning using historical remote sensing rainfall data obtained from satellites. We combine this with crop productivity data obtained from the Kimiti government agriculture services for the planting period (June-September) to predict future millet productivity.
We use millet as the indicator crop in the model because it is the main cereal crop in Kimiti. In our 2012 household survey of 1,400 households in 69 villages, 71 percent of all households reported having grown millet in the past year. Furthermore, millet is one of the cereals most commonly grown in areas with average yearly rainfall between 400 and 800mm because it is drought-resistant.
The model looks at the amount and distribution of rainfall throughout the five main growth phases of millet, because the requirements for water, nutrients and sun vary during each phase. In addition, it uses a moving start date to account for the large variation in sowing time based on the timing of the first rains. Remote sensing data is particularly valuable because it is available in real time and hence can provide an initial prediction of harvest quality a few months in advance of official national predictions.
The millet productivity data, in turn, can serve as a proxy of food security because most households depend primarily on rain-fed agriculture and have limited access to agricultural inputs.
In a context where, according to our survey results, 89 percent of the population are directly or indirectly reliant on agriculture for food or income, and 75 percent consume what they grow and have limited alternative employment opportunities, cereal crop production can serve as an appropriate proxy for food availability and potentially access.
While we have seen some success with the model in terms of accuracy, some limitations remain:
· When developing models, the more data you have the more accurate your model is. Forecasting models therefore require hundreds or thousands of data points. In this case, the more years of rainfall and crop production data we have, the more accurate the model can be. Currently we have only 14 years of data. To address this we used a statistical technique called repeated cross-validation, which allowed us to create “new” data points and increase our total number of observations. As we continue to add data annually, the information should become more accurate and precise.
· The model does not apply to all households. For example, households who have access to market gardens or cereal plots near the seasonal rivers would be less affected by a lack of rainfall. Pastoralists, who mostly rely on markets for cereals, are less impacted by local production shocks, although poor production would affect supply and prices.
As we continue to improve the model’s predictions, we are working with Concern Worldwide to consider how best to integrate it into existing local systems. The model must be useful to a wide range of potential users, including local communities and decision-makers at the departmental, regional, and national level in order to elicit an appropriate response to a potential emergency.
This model was registered as an official source for informing the ‘Cadre Harmonisé’ in Chad, which is a regional framework aimed at preventing food crises by quickly identifying affected populations and taking appropriate measures to improve their food and nutrition security. The Cadre Harmonisé uses food and nutrition security outcome indicators, corroborated by relevant contributing factors, to identify the food and nutrition insecure areas within the Sahel and West African Countries.
Anastasia Marshak is a researcher at the Feinstein International Center, Tufts University.