Boom-and-bust of arid vegetation: Insights from the latest satellite study
An analysis of satellite data shows that dry ecosystems, which cover 40% of land and feed over two billion people, are experiencing increasingly unstable greening cycles. The 'boom-and-bust' phenomenon could accelerate desertification and alter climate adaptation strategies.
Images from the Copernicus program captured a 15% increase in the variability of vegetation cover in arid zones of the Sahel, Australia, and southwestern America between 2000 and 2025. This is not just a localized 'greening': areas rapidly shift from productivity peaks to abrupt crashes, a pattern researchers describe as a 'boom-and-bust' dynamic.
What the study reveals: Precise results and key figures
According to the study published in Nature Climate Change, 40% of land area, nearly 150 million km², falls into the category of drylands. These regions support over two billion inhabitants, primarily through subsistence agriculture and livestock. The frequency of 'boom' episodes, where productivity increases by more than 30% compared to the decadal mean, has doubled since the beginning of the century. Conversely, 'bust' phases, marked by a drop of more than 25% in the same metric, have lengthened by 20%.
The authors note that these fluctuations are not uniform: areas most exposed to annual precipitation variations, such as western Sahel, show the most pronounced differences. Furthermore, climate prediction models from ECMWF, enriched with neural networks, confirm that the predictability uncertainty of precipitation in these regions has increased by 0.8 to 1.2 mm day⁻¹, exacerbating the unpredictable nature of the phenomenon.
Research method and data used
The team combined several sources of satellite data: optical observations from Sentinel-2, thermal measurements from MODIS, and radar products from Sentinel-1. These datasets, covering over 25 years, were harmonized using a predictive model based on machine learning, similar to GraphCast, which integrates both atmospheric variables (temperature, humidity, heat flux) and soil characteristics.
To reduce noise from sensors, researchers applied a convolutional neural network trained on time series data from the Copernicus Climate Change Service. This network, calibrated with in-situ data from weather stations and observation towers, allowed distinguishing climate variations from instrumental artifacts. The prediction uncertainty was quantified by comparing model outputs with real observations, providing an average margin of error of ±12% for NDVI estimates.
Additionally, simulations from Pangu-Weather, another AI atmospheric forecasting model, were used to project future precipitation scenarios. This hybrid approach – satellite data + AI – provided a spatial resolution of 1 km, sufficient to identify micro-refuges where vegetation persists despite drought.
