Evapotranspiration, water use efficiency, and yield for film mulched maize under different nitrogen-fertilization rates and climate conditions

The biodegradable film, as an ideal substitute for plastic film, has broad application prospects. However, it is uncertain in maize actual evapotranspiration (${\mathrm{ET}}_{\mathrm{ac}}$) components, yield, and water use efficiency (WUE) of biodegradable and plastic films during the different rainfall seasons. Therefore, a 4-year field trial with three mulching patterns (FNM: flat planting with non-mulching, RPM: ridge-furrow with plastic film mulching, and RBM: ridge-furrow with biodegradable film mulching) and two N-fertilization levels (0 and 180 kg N ha^–1) was conducted. The results showed that the machine-learning models could accurately estimate maize ${\mathrm{ET}}_{\mathrm{ac}}$ and its partitioning, and the random forest and artificial neural networks models had the highest accuracy and the least input variables after optimization. Compared to FNM, RBM and RPM increased $E_t$ by 10.8 mm, 14.0 mm in the dry season, 9.1 mm, 11.2 mm in the normal season, and 4.0 mm, 7.5 mm in the wet season, respectively, but decreased $E_s$ by 75.8 mm, 82.7 mm in the dry season, 48.6 mm, 56.7 mm in the normal season, 67.1 mm, and 74.9 mm in the wet season, respectively. Therefore, RBM and RPM decreased ${\mathrm{ET}}_{\mathrm{ac}}$ by 65.0 mm, 68.8 mm in the dry season, 39.5 mm, 45.6 mm in the normal season, and 53.1 mm, 67.5 mm in the wet season, respectively, compared to FNM. Nitrogen application had a similar effect on $E_s$ and $E_t$ but only increased ${\mathrm{ET}}_{\mathrm{ac}}$ by 13.3 mm in the dry season, 2 mm in the normal season, and 4.3 mm in the wet season, respectively, compared to N0. Furthermore, RBM and RPM under different nitrogen-fertilizations increased maize yield by 4.0 %, 3.6 % in the dry season, 3.0 %, 3.3 % in the normal season, and 5.3 %, 5.9 % in the wet season, respectively, also increased maize WUE by 23.3 %, 24.1 % in the dry season, 12.9 %, 15.0 % in the normal season, and 21.1 %, 23.4 % in the wet season, respectively, compared to FNM. This study proved that RPM could be replaced by RBM under 180 kg N ha^–1 in the different rainfall seasons in terms of reducing ${\mathrm{ET}}_{\mathrm{ac}}$, increasing maize yield, and improving WUE. The optimized machine learning models in this study also provided a low-cost method for computing regional maize ${\mathrm{ET}}_{\mathrm{ac}}$.