Water productivity of soybean production systems: A study integrating machine learning and global meta-analysis

Soybean (Glycine max) is one of the most significant crops globally. With its relatively low water productivity (WP), huge amount of water was consumed annually in soybean production. Understanding soybean WP on a global scale is useful for identifying limitations in WP improvement to reduce water use. A meta-analysis was conducted on global soybean WP, based on 2247 observations from 274 studies published from April 2000 to January 2025, with aims to quantify the current WP and its influencing factors, and estimate the boundary function for WP in the different soybean agroecological zones in the world. The global average WP was 6.4 kg ha−1 mm−1. The mean WP followed the order of Americas (mean: 8.1 kg ha−1 mm−1) > Asia (mean: 5.8 kg ha−1 mm−1) > Europe (mean: 5.6 kg ha−1 mm−1) > Africa (mean: 4.5 kg ha−1 mm−1). The boundary function model indicated that the theoretical potential WP values could reach 15.2 kg ha−1 mm−1, 9.6 kg ha−1 mm−1, 14.6 kg ha−1 mm−1 and 6.2 kg ha−1 mm−1 in Asia, Europe, the Americas and Africa, respectively. Combining machine learning and mixed effects model analysis revealed that key limiting factors of WP included mean annual precipitation (MAP), mean annual temperature (MAT)and soil organic matter (SOM). For different management measures, the meta-analysis showed that plastic film mulching played an important role in improving global soybean WP (30.6 % increase, p < 0.05), followed by nitrogen fertilizer application (24.4 % increase, p < 0.05), using irrigation (21.0 % increase, p < 0.05), and optimized irrigation management (7.4 % increase, p < 0.05). These findings provide a scientific basis for optimizing field management strategies to improve soybean WP.