Impacts of climate change on spatial wheat yield and nutritional values using hybrid machine learning

Abstract

Wheat’s nutritional value is critical for human nutrition and food security. However, more attention is needed, particularly regarding the content and concentration of iron (Fe) and zinc (Zn), especially in the context of climate change (CC) impacts. To address this, various controlled field experiments were conducted, involving the cultivation of three wheat cultivars over three growing seasons at multiple locations with different soil and climate conditions under varying Fe and Zn treatments. The yield and yield attributes, including nutritional values such as nitrogen (N), Fe and Zn, from these experiments were integrated with national yield statistics from other locations to train and test different machine learning (ML) algorithms. Automated ML leveraging a large number of models, outperformed traditional ML models, enabling the training and testing of numerous models, and achieving robust predictions of grain yield (GY) ( R ^2 > 0.78), N ( R ^2 > 0.75), Fe ( R ^2 > 0.71) and Zn ( R ^2 > 0.71) through a stacked ensemble of all models. The ensemble model predicted GY, N, Fe, and Zn at spatial explicit in the mid-century (2020–2050) using three Global Circulation Models (GCMs): GFDL-ESM4, HadGEM3-GC31-MM, and MRI-ESM2-0 under two shared socioeconomic pathways (SSPs) specifically SSP2-45 and SSP5-85, from the downscaled NEX-GDDP-CMIP6. Averaged across different GCMs and SSPs, CC is projected to increase wheat yield by 4.5%, and protein concentration by 0.8% with high variability. However, it is expected to decrease Fe concentration by 5.5%, and Zn concentration by 4.5% in the mid-century (2020–2050) relative to the historical period (1980–2010). Positive impacts of CC on wheat yield encountered by negative impacts on nutritional concentrations, further exacerbating challenges related to food security and nutrition.

Keywords

• CMIP6
• downscaled NEX scenarios
• automatic machine learning
• stacked ensemble model
• uncertainty
• nutritional concentrations