Abstract:

Climate change and water scarcity significantly threaten agriculture in arid regions. The Cele Oasis, located at the southern margin of the Taklimakan Desert in Xinjiang, China, is a typical arid-area oasis with a fragile ecology. An accurate prediction of the water requirements for cultivating crops in this area is crucial for the rational allocation of water resources and the development of sustainable agricultural practices. This study is dedicated to designing a prediction model applicable to the water requirements of the major crops in the Cele Oasis, revealing the intricate relationships among meteorological factors, crop growth characteristics, and water requirements, and circumventing the data-acquisition challenges associated with the Penman formula. This research integrated the Penman formula with the crop coefficient method. The daily water requirement was designated as the target variable. Based on the attribution analysis results, relevant meteorological parameters such as relative humidity, sunshine hours, and maximum temperature were selected to construct “XGBoost”, a water requirement prediction model. Moreover, different base learner types of XGBoost, including gbtree, gblinear, and dart, were explored to identify which among them was most suitable for the model.The results of this study were remarkable. XGBoost-based regression analysis revealed that relative humidity, sunshine hours, and maximum temperature were the dominant meteorological factors influencing crop water requirements, with a cumulative importance ratio reaching 75.81%. Among them, relative humidity demonstrated the highest impact, with an average feature importance of 39.84%, followed by sunshine hours (20.25%) and maximum temperature (15.72%). In terms of performance, the gbtree-XGBoost model demonstrated superior accuracy compared to the gblinear-XGBoost model. The R² value of the former increased by ~84.35% relative to the latter, with the root mean square error decreasing by ~0.625. The gbtree-XGBoost model could capture the complex nonlinear relationships between variables more effectively, and its predictions correlated markedly with the actual crop water requirements. In conclusion, this study successfully established a crop water requirement prediction model for the Cele Oasis. It could effectively capture the complex relationships among meteorological factors, crop growth characteristics, and water requirements. Among them, the gbtree-XGBoost model showed excellent performance and can be a reliable tool for guiding irrigation and allocating water resources in the Cele Oasis. It provides a scientific basis for the rational management of agricultural water resources in arid oases, which is conducive to improving water use efficiency, ensuring better crop yields, and promoting the sustainable development of agriculture in arid regions. This research also provides valuable references for similar studies in other arid areas, contributing to global efforts in designing water-saving agriculture methods and sustainable water resource management.

Key words: crop water requirement, Penman-Monteith equation, XGBoost regression, water requirement prediction model, Cele Oasis