By Michelle van der Spuy
Dr Mahlatse Kganyago, an expert in remote sensing and GIS, was raised in Limpopo, where his family relied on subsistence farming. His background inspired his mission to help farmers understand and manage climate-related risks and uncertainties.
Now a senior lecturer at the University of Johannesburg’s Department of Geography, Environmental Management and Energy Studies, Dr Kganyago has been nominated for the prestigious NSTF-South32 Awards, widely regarded as South Africa’s Oscars of science. He is a finalist for the NSTF-Agricultural Research Council Award in recognition of his innovative use of space technology and artificial intelligence to improve insight into agricultural systems and support food security.
1. What is your research about?
My research lies at the intersection of agriculture, climate and biodiversity. We use high-resolution satellite and drone imagery, combined with artificial intelligence (AI), to collect detailed data.
This information is then used to develop data-driven tools – AI models and remote sensing systems that support smarter decision-making. These tools help farmers optimise their management strategies and enable governments to better monitor food production.
The aim is to help farmers detect and address crop stress, as well as water and nutrient deficiencies, more quickly, ultimately making agriculture more resilient to climate shocks like heatwaves and droughts. We also provide governments with insights into where and when support is needed, and what potential threats may exist to food security.
2. What initially sparked your interest in this field?
I grew up in a small town in Limpopo, where most families, including mine, relied on subsistence farming. The arrival of the summer rains signalled the start of the planting season: We would plough the fields, monitor crop growth, hand-pull weeds, check for pests, harvest by hand, shell the maize, sift the grain from the chaff, and finally package and transport it to the local milling company. These activities defined much of my childhood summers.
But each season was a gamble. Drought, insect infestations or unpredictable rainfall could undermine all our efforts.
At the University of Johannesburg, I was introduced to geospatial science and remote sensing, fields that opened my eyes to powerful ways of addressing environmental and social challenges. I was especially drawn to how these technologies could be applied to agriculture, helping farmers manage climate risks and uncertainties more effectively.
I was fortunate to gain early exposure to Earth observation through an internship and, later, a master’s scholarship from the South African National Space Agency. These opportunities allowed me to contribute to meaningful projects focused on agriculture and food security.
3. What is the most important contribution of your research to the agricultural sector?
Our research has led to the development of satellite-based tools for monitoring crops and pastures across several African countries. These tools are locally calibrated, meaning they offer region-specific insights that are often missing from international systems.
We’ve also created operational platforms to help decision-makers assess how climate change affects crop and pasture productivity on a national scale. These decision-support tools allow governments to respond more effectively, directing aid where it’s most needed, making informed decisions about food imports and exports, and ultimately safeguarding food security.
Beyond large-scale monitoring, a key focus of our work is delivering field-level, site-specific information to farmers using satellite imagery and artificial intelligence. This allows farmers to make timely decisions, such as where and when to apply fertiliser or irrigation, to improve crop yields while minimising environmental impact.
One example is the CropWatch project for South Africa, funded by the UK Space Agency. As part of this project, we developed a farm intelligence system that provides detailed insights into crop types, arable land use, crop health, and responses to stressors like pests, drought, heatwaves and nutrient deficiencies. With this real-time information, farmers can take corrective action quickly, helping to protect both yields and livelihoods.
This technology puts farm management in the palm of a farmer’s hand – anytime, anywhere. We’re making advanced tools accessible, affordable and locally relevant. Currently, we’re also working on integrating large language models like ChatGPT to deliver clear, practical messages in local languages.
4. What unexpected issues did your research uncover, and were you able to find solutions?
Many current crop monitoring systems rely on the Normalized Difference Vegetation Index (NDVI), a simple metric derived from satellite imagery to estimate crop health and predict yield. Previous research has, however, shown that NDVI can be unreliable during the early stages of crop growth, largely due to soil interference.
Our work uncovered several additional technical and practical challenges. First, we found that partially cloudy conditions significantly degrade the quality of satellite images, especially in key wavelengths used to assess crop health. Even after removing atmospheric distortions, the resulting NDVI values often remained unreliable.
Second, we confirmed that there was no universal solution: Monitoring approaches must be tailored to specific crops, regions and growth stages.
Third, the fusion of data from different satellite resolutions (particularly Sentinel-2) introduced delays in processing.
Finally, we observed that many machine learning models used in crop monitoring were not easily transferable to other regions without performance loss.
To address these issues, we developed a number of practical solutions. We found that using crop structure variables such as leaf area density can bypass complex atmospheric corrections entirely. We also discovered that analysing individual Sentinel-2 bands – rather than fusing data from multiple sources – enables faster and more efficient crop information retrieval.
Incorporating sun and sensor angles into our models further improved accuracy. And by introducing just a small amount of local data (as little as 25%), we were able to significantly improve the adaptability of machine learning models to new geographic regions.
Lastly, we developed a new method called Spectral Triads Feature Selection, which identifies the most useful combinations of satellite data features. This method is as accurate as more complex techniques, but up to 80% faster.
In summary, our research has produced a suite of advanced, reliable crop monitoring tools – powered by machine learning and light-crop interaction models – that overcome the limitations of traditional approaches.
5. What do you do to stay curious and creative as a researcher, especially when things don’t go according to plan or progress slowly?
I’ve learnt to see unexpected outcomes not as failures, but as essential parts of the scientific process. When things don’t go as planned, which is often, I focus on understanding why they happened. What lessons can we take from them? That shift in mindset helps keep my curiosity alive.
I also make a point of collaborating with researchers from diverse fields, including agronomy, data science and ecology, which often leads to fresh insights and turns challenges into innovative solutions.
Most importantly, I involve end users in every project. Whether it’s farmers, agricultural extension officers or government officials, their input from the ground ensures that our research remains relevant and responsive to real-world needs. Engaging with the people we aim to support is one of my greatest sources of inspiration and creativity.
* Dr Mahlatse Kganyago holds a PhD in geography, specialising in earth observation, and an MSc (cum laude) in applied remote sensing and geographic information systems (GIS). He is a researcher and lecturer in remote sensing and GIS at the University of Johannesburg, and the principal investigator on several locally and internationally funded projects.
