Estimation of the rice water footprint based on machine learning algorithms

作者：Ali Mokhtar, Ahmed Elbeltagi, Saman Maroufpoor, Nasrin Azad, Hongming He, Karam Alsafadi, Yeboah Gyasi‐Agyei, Wenming He · 发表于：Computers and Electronics in Agriculture · 年份：2021 · DOI：10.1016/j.compag.2021.106501 · 被引用次数：44 · 研究领域：Environmental Impact and Sustainability、Hydrology and Watershed Management Studies、Water-Energy-Food Nexus Studies

It is essential to investigate the impact of climate change on the water footprint (WF) of rice from both historical simulation and future projections. In this study, four machine learning (ML) models, including random tree (RT), random forest (RF), additive regression (AR) and reduced error pruning tree (REPT) were used to model blue and green water footprint (BWFP and GWFP) for the present and future stages in the Yunnan Province, southwest China. Climate variables of daily precipitation, temperature, solar radiation, sunshine hours, wind speed, relative humidity and vapor pressure deficit data, yield and sown areas of rice were collected form 16 districts from 1990 to 2018. Six different scenarios (Sc1-Sc6) with different combinations of climate variables, crop coefficient and sown areas were used as inputs of the ML models for each of blue and green water footprint. Also, future climate projections of maximum (Tmax) and minimum (Tmin) temperatures, precipitation and sunshine were adopted for two different emission scenarios, RCP 4.5 and 8.5 from 2021to 2050, based on Geophysical Fluid Dynamics Laboratory (GFDL-ESM2M) model. For BWFP, the RT model in Sc1 with inputs of solar radiation, humidity, and vapor pressure deficit, was superior to the other scenarios with root mean square error (RMSE) and mean average percentage error (MAPE) values of 11.82 (m 3 ton -1 ) and 0.5%, respectively. Sc4 (sown area, Tmin, sunshine hours) and Sc5 (sown area, Tmin, Tmax, crop coefficient) ...