Simulating soil salinity dynamics, cotton yield and evapotranspiration under drip irrigation by ensemble machine...

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Soil salinization threatens cotton production, necessitating efficient water management in arid regions.
• Drip irrigation enhances water and fertilizer use but requires accurate soil salinity and evapotranspiration predictions.
• Traditional hydrological models are complex, limiting practical application.

Purpose of the Study:

• To develop a machine learning (ML) method for simulating soil salinity, evapotranspiration (ET), and cotton yield.
• To evaluate the performance of ensemble ML models compared to single models.
• To identify key input factors influencing prediction accuracy.

Main Methods:

• A global dataset from 134 literature sources was compiled.
• Stacking ensemble ML models were developed and compared with Gradient Boosting Decision Tree (GBDT), Random Forest (RF), and Extreme Gradient Boosting Regression (XGBR).
• Model inputs included soil data, meteorological factors, irrigation data, and other relevant variables.

Main Results:

• ML models achieved high accuracy (R = 0.78-0.99) in predicting soil salinity, ET, and cotton yield.
• Ensemble ML models showed superior performance, increasing R by 0.02%-19.31% over single models.
• Soil depth, distance from dripper, and days after sowing were significant factors influencing model accuracy.

Conclusions:

• Ensemble ML provides a robust and accurate alternative to traditional models for predicting cotton soil salinity and ET.
• The developed models can guide irrigation scheduling in arid and saline environments.
• Further research should focus on optimizing input combinations for enhanced predictive power.