Drought modelling
GAD supported a project to develop a standardised approach to estimating the poverty and nutritional impacts of droughts in Sub-Saharan Africa.
We worked with the Centre for Disaster Protection, to develop a model that simulates soil moisture in Malawi based on historic data. GAD’s model provided a methodology which can be adapted for simulating other drought measures in different countries.
Drought indicators
The Centre for Disaster Protection collaborated with the World Bank to examine various drought indicators and welfare measures.
They were looking to find a link between droughts and the impact on households, using historical weather event series as indicators of drought. GAD’s role was to develop a model to simulate the drought indicators.
Data science
GAD carried out a detailed exploration of historical data to ensure we understood the features and dynamics which we needed to capture in our modelling. We used this to develop a model which produced 10,000 simulations, over a 1-year time horizon of soil moisture at 5-kilometer points throughout Malawi.
We used data science techniques to build a model to reduce the dimensionality of the data and fit an appropriate time-series model. We then created simulated future projections of soil moisture for each of the data points.
Our analysts had to overcome the challenge of capturing the following behaviours of the underlying drought indicators:
- temporal: how to capture the seasonal trends and varying levels of soil moisture during the rainy and dry seasons
- spatial: how to incorporate correlations between spatial points (where we had to look at the relationships between 32,000 data points)
- scalability: how to make the model flexible and scalable so it can be applied to other countries and drought indicators
It is hoped that our model could be used to improve how we plan for and take action to avoid or mitigate the impact of droughts in Sub-Saharan Africa.