I have been invited to provide a keynote speech during the 2023 International Conference of Learning Representation's workshop about tackling climate change with machine learning. This post is the speech script and has been modified with some copyediting. The video version is available here.
Good morning, afternoon, or evening dear esteemed participants of the ICLR 2023 workshop on "Tackling Climate Change with Machine Learning." I am Racine Ly, and I lead the work related to Artificial Intelligence at AKADEMIYA2063 to support Africa Union member states to implement the continental agendas in the agricultural sector effectively. We see ourselves as mobilizers of African expertise and provider of data and analytics to support African countries in effectively implementing the Continental Agenda 2063 of the African Union.
I have been asked to provide a keynote speech on this crucial yet critical topic of tackling climate change with machine learning. I am in Morocco to attend a similar event with a different audience: the African Agriculture Adaptation (AAA) Initiative's annual Ministerial meeting. I could not make it in person and benefit from all your essential and insightful discussions.
Climate change is causing a range of impacts in Africa, including more frequent and severe droughts, floods, and other extreme weather events, as well as rising temperatures and sea levels. These changes affect food security, water availability, human health, and biodiversity and exacerbate social, economic, and political challenges. African countries are particularly vulnerable to climate change due to limited access to resources and technology and dependence on agriculture as a primary source of livelihood. Moreover, many African countries lack the infrastructure and resources to adapt to these changes, making them more vulnerable to the impacts of climate change. The effects of climate change in Africa are expected to intensify in the coming years, highlighting the urgent need for action to mitigate greenhouse gas emissions and build resilience in vulnerable communities.
Data and analytics are critical tools in tackling climate change in Africa and globally. They provide insights into the complex environmental systems that govern the continent and help us develop practical solutions to our challenges. By analyzing data from various sources, such as satellite imagery, weather stations, and on-the-ground sensors, we can track changes in climate patterns, identify areas of vulnerability, and predict the impacts of climate change on different sectors and communities. This information can inform policy decisions, guide investments in sustainable infrastructure, and support climate adaptation and resilience efforts. Additionally, data and analytics can help us monitor progress toward climate goals, measure the effectiveness of interventions, and identify areas where additional action is needed.
The most significant contribution of machine learning in all sectors is its prediction capability, be it classification or regression type. It is helping us to make the invisible become visible or at least blurry to allow navigation in times of uncertainties such as climate change and variabilities. Therefore, it is critical in feeding the decision-making process and investment orientations and monitoring progress by providing a "most likely" map of what would happen soon based on what we've been witnessing.
In the agricultural sector, therefore, on the application side of machine learning, they are several examples where machine learning is being used in various ways to combat climate change:
In food crop production forecasts, machine learning algorithms can analyze various environmental data, such as rainfall patterns, temperature, soil quality, and more, to predict crop production and optimize crop management practices. For example, the Africa Agriculture Watch program at AKADEMIYA2063 uses satellite remote sensing data and machine learning to predict food crop production in 47 African countries and nine crops. The model uses climatic and environmental data to provide its prediction; Therefore, any climate variability could be used to assess the impact on food crop production.
Pest detection and management: Machine learning algorithms can be trained to detect pests and diseases in crops and provide recommendations for management and control. The International Center for Tropical Agriculture (CIAT) uses machine learning to analyze images of cassava leaves and detect signs of disease, which can help farmers act before it spreads.
Climate-smart irrigation: Machine learning can optimize irrigation schedules and reduce water usage in agriculture. For example, the South African Agricultural Research Council (ARC) uses machine learning to develop an automated irrigation system that adjusts watering schedules based on local weather data and soil moisture levels.
Soil mapping and management: Machine learning algorithms can analyze soil data and recommend soil management practices, such as fertilization and tillage. The International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) uses machine learning to analyze soil data and develop maps of soil nutrients, which can help farmers optimize their fertilizer use.
On the modeling side, several architectural and computational innovations are also being proposed to improve machine learning models' accuracy, energy cost, and training time.
I can cite the recent advances in deep long-horizon forecasting with a Time-series Dense Encoder (TiDE), which promises to be ten times faster in training than transformer-based baselines while being more accurate on benchmarks. Such capabilities will allow us to soon adopt an exclusively data-driven approach for climate modeling, for example.
In computer vision, Object detection is getting closer to the GPT-3 moment with the hope of visual prompting to spread in the agricultural sector for crop and land features classification and change detection. The Meta's Segment Anything Model (SAM) is proof of that. Soon we will be able to move from object detection to visual prompting, where only a few clicks will be needed to identify areas of croplands suffering from water stress and tree counting, only to cite those.
The Large Language Models (LLM) are on their way to becoming a game-changer in making information available to different audiences in different languages. This could be used in disseminating information coming from climate services.
Today, it is safe to assert that the future of using machine learning to contribute to solving climate-related issues looks bright. However, beyond the need to further work on the adoption of machine learning in decision-making processes in Africa, for quality data availability and accessibility, and infrastructural investments, when it comes to solving climate-related issues, there is also the need to make sure that the outputs of machine learning-driven solutions reach the most vulnerable communities, those who might, first, bear the cost of climate change in their livelihoods. In the agricultural sector, they are the farmers, and let me be clear, our livelihoods depend on them as well. As we march towards and contribute to those significant machine learning innovations to face our most pressing challenges, such as climate change, let us think about them.
Ladies and Gentlemen, this is a short contribution I wanted to share with you today as you delve into the many exciting discussions during the workshop. I wish you an insightful conference and fruitful deliberations that will trigger new avenues of collaboration for the greater good.
Thank you.