Improving Droughts Monitoring in East Africa

The last few decades have seen devastating droughts affect millions of people in East Africa. Most recently, the 2015 Ethiopian drought left more than 80 million people requiring emergency food assistance. A theme running throughout this project is the need to deliver research that is useful to decision makers, thus meeting a fundamental goal of climate science; namely to reduce the impact of natural hazards.
I focus on answering 4 related questions:
1) Can we combine satellite derived datasets to produce a holistic drought metric?
2) Using these datasets and relevant climatological information can we use machine learning (neural networks) to improve standard statistical approaches to predicting vegetation health and drought extent?
3) Can we more accurately estimate climatic and ecosystem thresholds and the risk of reaching these thresholds?
4) What have been the impacts of historical droughts on the Ethiopian and Kenyan society and economy?
I have used satellite derived estimation of precipitation, evapotranspiration and soil moisture to quantify the magnitude and spatial extent of droughts over the last 30 years. These datasets have been used to create a number of indices, covering the entire hydrological cycle in order to more properly quantify droughts in East Africa. Moving beyond meteorological drought is important for decision makers who need to understand how drought propagates through the terrestrial part of the hydrological cycle. Satellite data was used because of its low latency, high spatial resolution and relatively long period of record. This work has been carried out with the CASE Industrial Partner, Airbus and their remote sensing division.
We developed a novel algorithm for estimating vegetation health, a useful proxy for agricultural drought, and sought to quantify and understand the relationships between vegetation health and climatological variables. This is essential for better forecasting of hydrological droughts and is a key space for improvement, constantly called for by farmers, insurers and policy makers in both Kenya and Ethiopia.
We used the vegetation health data to measure the impacts of droughts in different regions of Ethiopia and Kenya. Combining this information with land cover classifications, we had a statistical approach for estimating thresholds of certain areas and crop-types. This was then followed up by collaborating with plant ecologists and farmers in Kenya. Together we constrained the possible value of thresholds and can therefore more accurately measure the risk of certain thresholds being met within different return periods.
Finally, through working with another industrial partner, Vivid Economics, we set about quantifying the impact of droughts on economic indicators.
This research project has engaged directly with industrial partners and end-users in order to meet the demand of managing and minimising risk of drought related disasters. Through a spatial focus on East Africa, specifically Ethiopia and Kenya, it has brought information to an area that has been devastated by drought events over the last 30 years. It has used and developed new drought indicators and datasets for further scientific research. Furthermore, it has demonstrated the usefulness of machine learning in improving our understanding of drought and aiding our decisions in response to it. It is my hope that these data and approaches can be used elsewhere.