Coupling advanced earth observation, Machine learning, GeoAI, and CA-ANN model for assessing Spatio-temporal LULC change in Gilgel Gibe I Watershed, Upper Omo-Gibe Basin, Ethiopia

The landscape of the Earth is never truly still. It breathes, shifts, and transforms as people plant crops, build homes, and clear forests to make way for new needs. This constant reshaping of the ground beneath our feet is known as land-use change, and it is one of the most powerful forces altering our planet today. When natural areas like forests and grasslands are converted into farms or cities, the ripple effects touch everything from the water flowing in rivers to the animals that call those places home. To understand these changes, scientists have long relied on looking at the Earth from above, using satellites to take pictures of the same spot over many years. In recent decades, the ability to process these massive collections of images has exploded, thanks to powerful computers in the cloud that can analyze vast amounts of data in seconds. By combining these satellite views with smart computer programs that can learn patterns, researchers can now not only see how a landscape has changed in the past but also simulate how it might look in the future. This capability is vital for communities that need to plan for water security, food production, and the protection of their natural environment.

In the highlands of Ethiopia, a region known as the Gilgel Gibe I Watershed, these changes are happening with striking speed. This area, which feeds into the massive Omo-Gibe river system, is a critical source of water for irrigation, drinking, and generating electricity for the country. However, the land here is under intense pressure. A team of researchers set out to understand exactly what has been happening to the land over the last two decades and to forecast what might happen in the decades to come. They focused on a specific question: as the population grows and the need for food and housing increases, how is the balance between forests, grasslands, and farmland shifting, and what does that mean for the future of the watershed?

To answer this, the scientists turned to a digital toolkit that blends advanced satellite imagery with artificial intelligence. They gathered images from space taken in 2004, 2014, and 2024, covering a period of twenty years. These images came from different satellites, some capturing the Earth in high detail and others providing a broader view. Using a cloud-based system called Google Earth Engine, the researchers fed these images into a machine-learning algorithm known as Random Forest. Think of this algorithm as a very sharp-eyed student that has been shown thousands of examples of what forests, farms, and cities look like from space. By learning from these examples, the computer was able to automatically sort every pixel in the images into one of six categories: forest, water, built-up areas, bare land, grassland, or agriculture. The computer did this with remarkable precision, correctly identifying the land cover more than 93% of the time in the most recent year, giving the team a clear and accurate picture of the landscape at three distinct moments in time.

The results of this historical review revealed a dramatic transformation. In 2004, nearly 20% of the watershed was covered by forest. By 2024, that number had plummeted to just over 8%, a loss of nearly 50,000 hectares of trees. In its place, agriculture has surged. Farmland expanded from covering about 60% of the area in 2004 to nearly 79% in 2024. This shift was not a slow drift but a rapid conversion, where natural forests and grasslands were systematically turned into fields to feed a growing population. At the same time, the area covered by human settlements, or built-up land, exploded. What was once a tiny fraction of the landscape, less than 1%, grew to more than 3% in just twenty years, representing a massive increase in the footprint of towns and villages. The grasslands, which once covered nearly 13% of the watershed, were cut in half, disappearing to make room for crops and settlements.

But the story does not end with the past. The researchers wanted to know where this trajectory was heading. To do this, they used a sophisticated simulation model that combines two powerful concepts: cellular automata and artificial neural networks. In simple terms, this model acts like a digital crystal ball. It looks at the rules that governed the land changes between 2004 and 2014—such as how close a piece of land was to a road, how steep the slope was, or how far it was from a town—and applies those same rules to predict what the land will look like in 2034 and 2044. Before trusting the model for the future, the team tested it by asking it to predict the landscape of 2024 based on earlier data. The model succeeded, matching the actual 2024 map with high accuracy, which gave them confidence to look further ahead.

The projections for the coming decades paint a sobering picture of continued change. If current trends persist, the forest cover in the watershed is expected to shrink even further, dropping to just 6.5% of the total area by 2044. This would mean that over three-quarters of the forest that existed in 2004 would be gone. Conversely, farmland is projected to expand to cover more than 81% of the watershed, approaching a point where there is very little natural land left to convert. The built-up areas are also expected to grow rapidly, doubling in size over the next twenty years as urbanization accelerates. The grasslands, already diminished, are projected to vanish almost entirely, leaving behind a landscape dominated by crops and concrete.

These changes carry profound implications for the health of the watershed and the people who depend on it. Forests and grasslands act as natural sponges, soaking up rain and releasing it slowly into rivers, which helps prevent floods during heavy rains and maintains water flow during dry seasons. As these natural areas are replaced by hard surfaces and intensive farming, the land loses this ability to regulate water. This can lead to faster runoff, increased soil erosion, and a higher risk of both flooding and drought. Furthermore, the sediment washed away from eroded fields can clog the reservoirs that power the region's hydroelectric dams, threatening the long-term supply of electricity. The loss of grasslands also threatens the livelihoods of pastoral communities who rely on open grazing lands for their livestock.

The study does not claim that these future outcomes are inevitable, but rather that they are the likely result if the current patterns of development continue unchecked. The researchers emphasize that these simulations are based on historical trends and do not account for potential changes in policy, new conservation efforts, or shifts in economic behavior. However, the clarity of the data provides a powerful warning. It suggests that without significant intervention to protect the remaining forests and manage agricultural expansion more sustainably, the watershed could face severe environmental challenges. The findings offer a critical tool for planners and policymakers, providing a clear visual and quantitative basis for making decisions about land use. By understanding exactly how the landscape is changing and where it is headed, the region has a chance to steer its development toward a future that balances human needs with the preservation of the vital natural systems that support them.