Suitability assessment of film mulching on maize production in Northwest China: Integrating meta-analysis with machine learning

Abstract

Context

Film mulching (FM) is commonly used in Northwest China to address water scarcity in agriculture by promoting soil warming and moisture retention. However, FM is not suitable for all regions, and excessive reliance on this method can lead to overuse, which may harm the farmland ecosystem. Understanding where and how FM is most effective is crucial for ensuring its sustainable use.

Objective

This study assessed the suitability of FM on maize considering yield and water use efficiency (WUE) in Northwest China by integrating meta-analysis with machine learning techniques. In addition, the analysis aimed to assess the regional and environmental factors influencing FM performance.

Methods

A meta-analysis was conducted, synthesizing data from 141 studies, to evaluate the influence of FM on maize yield and WUE. Machine learning models, including Random Forest regression, support vector regression, and gradient boosting regression tree, were applied to predict the regional suitability of FM based on climatic, soil, and management practices. Key factors influencing the effect of FM included climatic factors (mean annual precipitation and mean annual temperature), soil characteristics (bulk density, soil organic matter, and total nitrogen), and fertilization strategies (nitrogen and phosphorus). Pearson correlation analysis was conducted to explore the relationship between the 7 factors and the effectiveness of FM, while Random Forest was utilized to prioritize the importance of each factor.

Results

The meta-analysis revealed that FM increased maize yield by 40.55 % and WUE by 40.79 %. Plastic mulch demonstrated superior effectiveness, improving yield by 43.68 % and WUE by 43.85 %. FM performed best under conditions of scarce resources. Among the 7 factors, mean annual precipitation, mean annual temperature, and total nitrogen were of higher importance in the prediction. Random Forest regression excelled in predicting yield and WUE changes. The spatial analysis revealed notable regional variability of FM, with the best results observed in Xinjiang and Gansu.

Conclusions

This study highlighted the effectiveness of FM in improving maize yield and WUE in Northwest China, with regional variability in its performance. The results indicated that FM was most beneficial in regions with limited water and heat, particularly in Xinjiang and Gansu. Moreover, the study also demonstrated the utility of machine learning models, particularly Random Forest regression, in predicting FM suitability across regions.

Significance

This study offered valuable insights into the regional suitability of FM for maize production in Northwest China, providing guidance for agricultural policy and management decisions to enhance the sustainability of FM. In addition, by integrating meta-analysis with machine learning, it presented an effective method to predict FM's impact on crop yield and water use efficiency.