Interview with June Lukyu
June Lukuyu is an incoming Assistant Professor in Electrical and Computer Engineering at the University of Washington. She is currently completing a Ph.D. at the University of Massachusetts Amherst in the Systems Towards Infrastructure Monitoring and Analytics (STIMA) Lab with Professor Jay Taneja.
June's research focuses on designing and evaluating technology-driven strategies that enable socio-economic development in underserved regions through stimulating affordable, reliable, and climate-aware electricity use while considering the opportunities and challenges of electricity systems. She is a researcher on an EEG project focusing on electricity demand forecasting in agriculture in Ethiopia and recently took part in an EEG webinar that explored how remote sensors and technology can contribute to better electricity system planning.
Your work involves building models for electrification planning and energy systems planning. Can you explain more about the process and the methods you use, and how technology like satellite data and machine learning can potentially improve the modelling process?
I use a wide range of data analytics and computing techniques to build models that evaluate how the adoption of end-use electric technologies in major economic sectors like agriculture and transportation could impact existing electricity systems or inform the planning of future electricity systems in sub-Saharan Africa.
Lack of data is one of the significant constraints of developing high-quality models for the continent that can produce reliable data-driven insights for policy-making. Satellites measure an incredible amount of data about the earth's conditions at a very granular scale and with high accuracy. However, this data is only useful if it can be leveraged and translated into actionable insights. That's where machine learning comes in. We use the limited ground-based data that is available to teach machine learning models to extract patterns of information from satellite data that are useful for electrification and energy systems planning. We then apply these trained models to areas where no ground-based data exists.
For the EEG project on electricity demand forecasting in agriculture, you are combining satellite data and machine learning with survey data. How does this all work in practice? Why is it important to gather data from on-the-ground surveys rather than just relying on satellite imagery?
A machine learning model is capable of learning to perform a task from examples, for example, a teacher giving math students several practice problems along with the answers to teach them a concept. In our project, our goal was to teach a machine learning model to identify areas with diesel-powered irrigation activity by learning from features related to diesel pumps and irrigation that can be measured remotely using satellites such as pollution, crop cover, and surface water data.
We used real examples of where diesel irrigation pumps were operating in Ethiopia and their associated features (just like the answers to the math problems). We gathered these real examples from on-the-ground surveys. Ground-truth data, in this case from surveys, are essential to train the model and check how well the model has learned the task. Therefore the more ground-truth data you have, the more examples the model has to learn from and the more confident you can be in your model's performance.
Satellite data enables us to apply the trained model beyond the areas where we have ground-truth data. Once we are confident that the model can perform the task accurately on our ground truth data, we can apply it to the satellite data in other areas, in our case, to predict whether these areas have diesel irrigation activity or not. The key is to select applications where alternative guidance does not exist.
How do you see this sort of research and analysis contributing to improving energy access and its productive use?
Policymakers in the energy space need to ground their decision-making on data-driven insights. Yet, in many parts of Sub-Saharan Africa, high-quality, accessible traditional data sets, such as surveys, energy systems, and economic data, are scarce to analyze and subsequently inform policymakers. These methods could help bridge these data gaps by enabling us to make scalable, generalizable, and transferable models that can inform energy access interventions, pilots, and policies even in areas without data.
Why did the research team decide to focus on Ethiopia's agricultural sector?
Ethiopia seeks to transform its agricultural sector, which currently has low productivity. The government is keen to invest in improvements but sought assistance in enhancing the impact of those investments.
Can you share any results or learnings from the project?
A key motivation for this project is to address demand constraints to ensure that electrification stimulates economic growth. When electricity providers like utilities and mini-grid providers build electricity infrastructure in areas with low electricity demand, they try to recoup the cost of their investments by setting" cost-reflective" tariffs that are unfortunately often unaffordable to rural customers, thus hindering them from using electricity to improve their well-being. We collected real examples of where irrigation is happening in Ethiopia's Amhara and Oromia regions, as well as ground-truth information on the method of irrigation used through survey-based methods. We used part of this data to train a machine learning model to identify patterns in satellite-measured pollution, crop cover, and ground topography that reflect diesel-irrigation activity. We tested the accuracy of our model using the remaining ground-truth dataset and found that the model could identify the areas with diesel-irrigation activity with about 75% accuracy. Locations with diesel-powered irrigation represent areas more likely to adopt electricity and have stable revenue from electricity sales, and create a basis for electricity providers to set cost-reflective and affordable tariffs. Our findings are preliminary but promising. Scaling up ground-truth data collection efforts and incorporating our model and model outputs into decision-making processes is crucial for realizing the potential impacts of the work.
You are working with electricity system planners and other stakeholders in Ethiopia to incorporate the findings of the study into the country's electricity planning processes. Does this involve capacity building, and if so, how is it being implemented and what progress has been made?
We have held several workshops attended by stakeholders from the Ministry of Water and Energy, presenting our study and findings and engaging in discussions on the potential value of our methods and findings in electrification planning. I believe this is a valuable step in making a case for the value-add of local capacity building in predictive data analytics. We are eager to engage with system planners across the agriculture and energy sectors to enhance their joint planning. Still, all this hinges on stakeholders becoming conversant with data-driven decision-making.
You recently took part in an EEG webinar on the contribution that remote sensors and other types of technology can make to electricity system planning. What are your key takeaways from the event?
One of the key takeaways for me as an applied researcher is that we need to actively think about how to effectively translate our predictive and analytics models and model outputs into actionable insights that can be seamlessly integrated into the policy-making process. Second, there is a need for investments in platforms for data sharing from government and private stakeholders with researchers and researchers with each other. Investment in infrastructure to ensure data privacy, security, and de-identification will also be needed.
Can you tell us more about your Ph.D. and also what you hope to achieve in your new role at the University of Washington?
My Ph.D. research has focussed on using computing and data analytics tools to build models for evaluating strategies for stimulating economic growth in Africa through enabling affordable, reliable, and climate-aware electricity use under the constraints of local electricity systems. In addition to building predictive data analytics models in the agricultural sector, I have also worked on building models to evaluate pathways for transport electrification in Africa. At UW, I hope to continue as an applied researcher working on data analytics applications in energy, climate, and development. However, I would like to take a more community-centered approach to designing and applying data analytics models for energy and development policy-making.
We need to be proactive instead of reactive in considering the equity and justice implications of predictive and analytics models. Instead of thinking about how our models impact equity and justice in underserved communities, we need to be able to integrate community preferences, needs, and cultures into how we design and interpret our models, as well as how we present them to policymakers. This will require extensive multidisciplinary partnerships and collaborations, which I hope to forge.