A recent study has introduced a novel data-driven model that distinguishes between human-induced and natural water consumption in croplands, providing valuable insights into the sustainability of arid lake ecosystems.
Drylands cover 42% of the Earth's land surface and support 38% of the global population, putting them at the forefront of the struggle for limited water resources. The agricultural expansion has placed pressure on ecosystems, resulting in many terminal lakes shrinking or disappearing entirely due to excessive irrigation.
Scientists developed a model that can differentiate between natural and human-driven water consumption in croplands by utilizing remote sensing and machine learning technologies
To tackle this challenge, a research team has conducted studies in the Ebinur Lake Basin in China, an arid region where agricultural growth has placed increasing pressure on water resources. They developed a model that can differentiate between natural and human-driven water consumption in croplands by utilizing remote sensing and machine learning technologies, providing clearer insights into water usage patterns.
The researchers are from the Aerospace Information Research Institute of the Chinese Academy of Sciences, along with their international collaborators. Their findings were published in the Journal of Remote Sensing.
A key finding of this study is a sophisticated method for distinguishing cropland evapotranspiration (ET) into its natural (ETn) and human-induced (ETh) components. The model demonstrated enhanced accuracy, with R² values ranging from 0.88 to 0.96, indicating that by 2019, human activities accounted for 77% of cropland water consumption. Notably, restoring Ebinur Lake to its optimal surface area of 800 km² would require an additional 0.29 km³ of water annually, indicating the toll agricultural expansion has taken on regional water reserves.
In this study, the researchers monitored cropland and lake dynamics from 2003 to 2019 using Sentinel-2 satellite imagery, deep learning, and machine learning algorithms. A random forest regressor was trained to model the relationship between environmental factors and natural ET, achieving high predictive accuracy.
The study revealed that cropland in the Ebinur Lake Basin increased by 50.65% during the observation period, resulting in a 61% rise in total water consumption. Notably, ETh saw a sharp increase after 2013, which coincided with a rapid expansion of irrigated farmland. These findings were thoroughly validated using water level data from the DAHITI database and surface water measurements from the Global Surface Water Dataset (GSWD).
This study presents a new approach to water resource management by combining high-resolution satellite data with machine learning techniques. Potential applications of this method include real-time water monitoring, optimized irrigation strategies, and proactive conservation efforts to prevent the drying up of lakes in water-stressed regions like Central Asia and beyond.
