Dr. Samuel Jonson Sutanto

In many parts of the Global South, smallholder farmers rely not on weather forecasts based on numerical modeling called scientific forecast, but on something far more rooted in their gene—Indigenous knowledge. From observing the behavior of animals to tracking changes in meteorological and astronomy patterns, these local forecasts have been passed down through generations.

But as the climate becomes increasingly variable, traditional indicators no longer hold the same predictive power they once did. At the same time, scientific weather forecasts often remain inaccessible or are viewed as unreliable by rural communities. This creates a dangerous gap in timely and accurate weather information—particularly as farmers face increasingly unpredictable rainfall.

Our latest research, published in Environmental Research Letters, explores a way to bridge this gap through hybrid precipitation forecasting. The approach combines the local forecasting knowledge of farmers in northern Ghana with scientific forecasts, using machine learning models to improve rainfall predictions.

The Innovation: Hybrid Forecasting with Machine Learning

We designed and tested a set of machine learning models that integrated:

• Indigenous knowledge-based forecasts from local communities
• Scientific precipitation forecasts from numerical weather models
• Observed rainfall data as ground truth

The results were striking:
✅ 23% more accurate than using scientific forecasts alone
✅ 33% more accurate than relying solely on Indigenous forecasts

Among the machine learning techniques tested, the most effective were random forests (RFs) and voting classifiers that combined multiple RFs. These models outperformed traditional statistical methods, demonstrating the potential of modern data science tools to capture both empirical and experiential knowledge.

Why This Matters

More accurate and locally grounded precipitation forecasts could significantly enhance agricultural decision-making for smallholder farmers. Better forecasts mean better choices about when to plant, irrigate, or harvest—ultimately leading to more resilient livelihoods.

Beyond technical accuracy, our work underscores a deeper point: Indigenous knowledge has value. When paired with scientific data and cutting-edge analytics, it can lead to forecasting systems that are both more precise and more trusted by local communities.

Looking Ahead

This research is part of a growing movement to co-develop climate information services with users—not for them as top-down approach. By treating farmers as partners and knowledge holders, rather than passive recipients of information, we can create more meaningful and effective climate adaptation tools.

You can read the full paper here.

Authors: Samuel J. Sutanto, Joep Bosdijk, Imme Benedict, Arnold Moene, Dragan Milosevic, Fulco Ludwig, and Spyridon Paparrizos