Integrating GIS and machine learning for pollution vulnerability mapping in South Africa’s Kruger National Park

The air we breathe is rarely still. Even in the most remote corners of the world, invisible clouds of smoke, dust, and gases drift across borders, carried by wind and weather. While we often think of air pollution as a problem confined to busy cities and industrial zones, these pollutants travel far, settling in wild places where they can harm plants, animals, and the delicate balance of nature. For the managers of vast protected lands, knowing where these invisible threats are strongest is difficult. They cannot simply place a sensor on every hill or river, and the weather changes too quickly for old-fashioned guessing games to work. To solve this, scientists are turning to a powerful combination of tools: satellite eyes that see the whole planet, digital maps that organize that view, and smart computer programs that learn from patterns to predict the future.

In a recent study focused on South Africa's Kruger National Park, researchers used this modern toolkit to create a detailed map of pollution vulnerability. Kruger is one of Africa's largest game reserves, a sprawling wilderness of nearly 20,000 square kilometers where elephants, lions, and rhinos roam freely. The park is not isolated from the world; it sits near human settlements and industrial areas, meaning it is subject to pollutants drifting in from outside. The researchers wanted to understand which parts of this massive park are most at risk from air pollution and what factors make them vulnerable. They did not rely on a single method but instead built a system that combined satellite data on vegetation, rainfall, and temperature with measurements of specific gases like nitrogen dioxide and sulfur dioxide. They then fed this information into three different types of computer learning models to see which one could best predict where pollution would be a problem.

The team gathered data from space, using satellites to track everything from the density of the green vegetation to the concentration of harmful gases in the atmosphere. They looked at how much rain fell, how hot it got, and how much methane and carbon monoxide were present. By layering all this information together, they created a digital picture of the park's environment. They then trained their computer models to recognize the signs of pollution. One model, called Random Forest, works by building many small decision trees and asking them to vote on the answer. Another, the Gradient Boosting Machine, learns by correcting its own mistakes one step at a time. The third, a Support Vector Machine, tries to find the best possible line to separate clean areas from polluted ones. The goal was to see which of these approaches could most accurately identify the areas where the ecosystem is most likely to suffer from poor air quality.

The results showed that the computer models were quite successful at mapping the invisible threats. The Gradient Boosting Machine model proved to be the most accurate at correctly classifying the data, getting the right answer about 76 percent of the time. The Random Forest model was slightly less accurate in that specific measure but was exceptionally good at distinguishing between different levels of risk, a quality that is crucial for making fine-grained decisions. The third model, the Support Vector Machine, performed well but was not as strong as the other two. All three models agreed on the general pattern: the southern and south-central parts of the park, particularly near the Crocodile and Sabie rivers, are the most vulnerable. These areas showed higher concentrations of nitrogen dioxide and sulfur dioxide, gases often linked to burning fuel and industrial activity. The models also revealed that these two gases were the most important factors in predicting where pollution would be a problem, more so than the amount of rain or the density of the trees.

What makes this finding significant is that it confirms pollution is not just a city problem. The study suggests that even in a wild place like Kruger, the air quality is heavily influenced by human activity far away, carried in by the wind. The researchers found that the areas with the highest risk are not random; they cluster around specific river valleys and the southern edges of the park, where the wind often carries pollutants from nearby towns and roads. The computer models also highlighted that vegetation plays a complex role. While thick forests can sometimes absorb pollution, the data showed that areas with dense greenery were often correlated with higher levels of certain gases, likely because these areas are also closer to the sources of the pollution. This nuance is something that simple observation might miss, but the computer models caught it clearly.

The final output of this work is a set of high-resolution maps that divide the park into five categories, ranging from very low vulnerability to very high vulnerability. These maps show that about one-fifth of the park falls into the highest risk category, while another fifth is in the very low category. The middle ground, where the risk is moderate, covers the largest portion of the land. These maps are not just pictures; they are practical tools for park managers. By knowing exactly where the air is most likely to be toxic, rangers and conservationists can focus their monitoring efforts, plan fires more carefully, and protect sensitive wildlife in the areas that need it most. The study suggests that this approach of combining satellite data with smart computer learning can be a game-changer for protecting large wilderness areas, offering a way to see the invisible and act before damage is done.

The research also connects to broader goals for the planet. By protecting the land and its air, the study supports efforts to maintain healthy ecosystems, which is a key part of global sustainability goals. It also touches on climate action, as the same weather patterns that move pollution also drive climate change. While the study did not measure the health of people living nearby, it provides the environmental data needed to understand how pollution moves and settles, which is the first step toward protecting both wildlife and human communities. The work demonstrates that even in a place as vast and wild as Kruger, science can provide a clear view of the challenges facing nature, turning complex data into a simple, actionable guide for conservation. The maps produced are a testament to the idea that with the right tools, we can see the threats to our natural world clearly enough to stop them.