This blog is part of the GSMA AgriTech AI Blog Series, which highlights AI innovations in agriculture. Throughout the series, we will explore various use cases where AI is transforming agritech solutions and farming practices.
Since 2011, Ignitia has been offering weather and climate intelligence, addressing the shortcomings of traditional forecasting methods that fail millions of climate vulnerable populations, including smallholder farmers in low-and-middle income countries (LMICs). By leveraging machine learning, the company has developed highly accurate, localised weather models specifically tailored to regions most vulnerable to climate disasters.
This blog presents how Ignitia is using machine learning, a subset of Artificial Intelligence (AI), to provide climate vulnerable populations, including smallholder farmers, with the weather information they need to thrive in an increasingly unpredictable climate. We’ll also explore the benefits delivered by Ignitia’s innovative AI-driven approach, some considerations for using AI in weather intelligence, and what lies ahead for the company.
Poor accuracy of traditional weather forecasting in the tropical belt
For many of the world’s 500 million smallholder farmers, reliable weather information has long been out of reach. Complex and unique atmospheric dynamics in the tropical regions where they live make traditional global weather forecasts largely ineffective. Tropical thunderstorms, which only last for a couple of hours, are particularly challenging to predict with conventional forecasting models. Besides their low accuracy, traditional weather forecasts—typically broadcasted on radio—are often location-agnostic, making it difficult for farmers to know what actions to take for their specific location.
Consequently, these farmers often rely on their own weather observations to make agricultural decisions. For example, they predict the start of the rainy season by looking for signs of rising humidity or clouds. However, climate change has made weather patterns increasingly erratic. It is not uncommon to experience prolonged dry spells even within the rainy season, making these empirical assumptions risky for timing agricultural activities.
Getting access to accurate and localised weather forecasts is critical in agriculture, particularly for smallholder farmers who rely on rain to grow their crops. Accurate forecasts enable farmers to make informed decisions about planting, fertilizing, and harvesting. The consequences of mistimed planting can be severe, potentially resulting in crop failure and financial loss for smallholder farmers who often operate on thin margins. For instance, a study conducted in Nigeria found that delaying maize planting by just 30 days beyond the optimal date resulted in a 52% reduction in yield.
Harnessing AI for tropical weather forecasting
Founded in 2011, Ignitia set itself on a mission to solve the tropical weather forecasting conundrum. The company spent years researching, and testing, how to capture the types of weather events, particularly storms, that drive weather conditions in West Africa. This led co-founder and Chief Science Officer Andreas Vallgren to develop an ensemble forecasting system, where multiple forecasting models collaborate to estimate the probability of weather events. These models quickly outperformed global models in the region.
Ignitia took a significant leap forward by incorporating AI into its forecasting models in 2018. The purpose of AI was to replicate the expertise and decision-making processes of Ignitia’s seasoned tropical weather forecasters, enhancing the accuracy of predictions, particularly for short-lived weather events like thunderstorms.
With nearly eight years of historical data at its disposal by 2018, Ignitia used machine learning to recognise patterns and produce more accurate forecasts based on a vast array of data inputs, including rainfall forecasts from Ignitia’s own supercomputer cluster, meteorological data covering various parameters such as wind and humidity, and satellite imagery for rainfall estimates. Ignitia’s AI-powered system uses deep learning-based modelling to look at the whole global tropical and subtropical belt at once, searching for patterns and producing forecasts based on these data inputs. The machine learning model is retrained at least once a year with more recent data to adjust for climate change impacts.
The resulting weather forecasts, provided to end-users through various channels, including SMS, WhatsApp, and web dashboards, include short-term forecasts (2-day and 7-day), real-time storm alerts, and longer-term monthly and seasonal outlooks that help farmers plan for the entire growing season. Using a farmer’s Cell ID, i.e. the unique identifier for a cell tower in a wireless network, Ignitia is able to provide these forecasts in a location-specific manner.
Key benefits of using AI in weather forecasting
Incorporating AI into digital weather and climate services has the potential to greatly enhance their accuracy, efficiency, and overall impact to farmers.
Since the incorporation of AI in its modelling, Ignitia has seen notable improvements in forecasting accuracy and speed, while also benefiting from more scalable and cost-effective operations.
Improved accuracy: The accuracy and reliability of AI-powered models was even better than the ensemble forecasting model Ignitia had previously developed. The absence of human biases, which can sometimes lead to conservative predictions, allowed AI-powered models to make more accurate assessments, especially in extreme weather scenarios such as flooding or flash floods. This is crucial as the frequency of these events is increasing due to climate change.
Faster forecasting: While the ensemble model could take 4 to 5 hours to produce a forecast, following the roll-out of AI, the model can produce a forecast in a matter of seconds. This is critical in the case of real-time weather alerts.
Scalability: AI has enabled Ignitia to expand its geographic coverage rapidly. It took Ignitia 5 to 6 years to develop the ensemble forecasting model for West Africa, but with the use of AI, the company was able to expand across the full African continent in just 6 months, and across the entire tropical belt in a further 3 months. That would have taken a decade using Ignitia’s pre-AI ensemble forecasting system, according to Vallgren.
Cost-effectiveness: The use of AI has replaced the need for massive, expensive supercomputer systems making it possible for Ignitia to deliver high-quality forecasts at a lower cost. This has also allowed the company to scale its operations without significant additional investments in hardware.
Reallocation of human resources: The use of AI prompted a restructuring of expertise within the team. With weather forecasters no longer needing to analyse input data, their attention shifted to other value-added activities. Now, these experts focus on translating weather data into actionable insights for farmers, offering practical agronomic advice on topics such as when to plant or apply fertiliser.
Considerations for using AI in weather forecasting
While using AI can improve the efficiency and impact of weather forecasting, its implementation can be challenging. Like many other agritech companies leveraging AI, Ignitia faced difficulties in accessing datasets, selecting data inputs and defining metrics that are both accurate and valuable for farmers.
Accessibility of weather datasets: Getting access to the necessary weather data to train the AI model using machine learning remains a challenge. This is particularly important for predicting rare events, for which there is limited historical data. While some organisations, like the European Centre for Medium-Range Weather Forecasts, have made strides in making data more accessible, in many cases, data remains largely unavailable.
Selecting data inputs: Determining the appropriate data inputs to train the AI model can be a difficult task. As Vallgren explains, decisions must be made regarding the length of the time series and the interval required to ensure that no crucial information is lost.
Defining the right metrics for the AI model: Ignitia’s machine learning specialists struggled to define metrics for the AI model that resulted in not only scientifically accurate but also valuable forecasts for farmers. Close collaboration between meteorologists, machine learning experts, sales and customer service teams has been key to identify key metrics and align the model with user needs. For example, understanding the relative importance of different types of errors (e.g., false alarms vs. missed events) has helped optimise the AI model. As Vallgren puts it, “AI in itself can only take you so far if you don’t understand the problem that you’re trying to solve.”
Future outlook for the use of AI at Ignitia
Looking ahead, Ignitia plans to expand the use of AI to enhance its real-time weather alerts for early warning systems, develop climate risk scoring to expand access to insurance and finance, and provide agronomic advisory services that strengthen farmers’ resilience to climate change.
One of the key focus areas in the future will be using machine learning to assess crop suitability in specific locations. By building datasets that consider both current and future climate conditions, Ignitia aims to help farmers choose the most appropriate crops for their regions, thereby increasing farm profitability and resilience to climate change.
Ignitia is also exploring the use of AI to generate more tailored advisory content. While the input data for their agronomic advisory services already comes from AI-enhanced datasets, the next step is to use AI to generate the advisories themselves. The company is thinking of integrating large language models, similar to those used in conversational AI systems. This would enable users to interact with Ignitia’s systems more intuitively, asking questions and receiving personalised advice based on real-time data.
Finally, while Ignitia’s current AI models focus on short to medium-term forecasts, the company is exploring ways to apply machine learning to seasonal forecasting. Vallgren refers to ongoing R&D to use AI for predicting the start date of the rainy season based on factors like sea surface temperature. That creates opportunities for AI to be used to analyse longer-term climate trends and provide insights into future climate risks for tropical regions.
Realising this vision holds potential to empower farmers make informed decisions, potentially leading to improved agricultural outcomes and enhanced food security in the region.
Stay tuned for more insights into how AI is shaping the future of agriculture.
