Simulating soil hydrologic dynamics using crop growth and machine learning models

Fitsum T. Teshome

doi:10.1016/j.compag.2024.109186

Simulating soil hydrologic dynamics using crop growth and machine learning models

2024-09-01

0

PRE

AI

F

Fitsum T. Teshome

H

Haimanote K. Bayabil *

DOI：10.1016/j.compag.2024.109186

原文链接

原文求助

摘要

En 中文

Accurate measurement of crop evapotranspiration (ETc) and soil moisture content (SMC) is critical for different purposes, including developing irrigation scheduling practices that improve water use efficiency and crop yield. The objectives of this study were to 1) simulate daily ETc and SMC of green beans and sweet corn under full irrigation and three deficit irrigation rates using the Decision Support System for Agrotechnology Transfer (DSSAT) CROPGRO-Green bean and CERES-Sweet corn models and 2) evaluate the performance of three machine learning models for simulating ETc of green beans and sweet corn. The DSSAT models were calibrated using measured phenological, biomass, and yield data collected during two years of experiments at the University of Florida, Tropical Research and Education Center (TREC) during the winter (dry) seasons of 2020-21 and 2021-22. The experiments were conducted under four irrigation treatments: 100% full irrigation (FI), 75% FI, 50% FI, and 25% FI, with four replications. The simulated daily ETc and SMC under each irrigation treatment were compared with estimated ETc based on changes in SMC and measured SMC, respectively. Estimating ETc based on changes in SMC involved quantifying sub-hourly moisture loss during the drying phase of soil and aggregating the results into daily timesteps. The eXtreme Gradient Boosting (XGB), light gradient-boosting machine (LightGBM), and Support Vector Machine (SVM) models were trained using SMC and weather variables to simulate daily ETc of green beans and sweet corn. The results showed that the CROPGRO and CERES models could reasonably simulate SMC dynamics of green beans and sweet corn plots. The models simulated SMC better for the three deficit irrigation treatments than the full irrigation treatment. The CROPGRO model evaluation index of agreement (d-Stat) and mean absolute error (MAE) for 25% FI treatment were 0.72 and 0.03 cm3 cm-3, and 0.75 and 0.03 cm3 cm-3 for the CERES model. However, the DSSAT models did not perform well in simulating ETc for green beans and sweet corn. Average d-Stat and MAE for the CROPGRO during model evaluation were 0.5- and 1.0-mm day-1 and 0.6- and 1.1-mm day-1 for CERES, respectively. The performance of all machine learning (ML) models for simulating daily ETc of green beans and sweet corn was better than the CROPGRO and CERES models. Incorporating machine learning algorithms into the DSSAT model has the potential to enhance its performance in simulating ETc. Overall, the results suggest that DSSAT and ML models could potentially be used as alternative decision support tools for assessing and managing irrigation strategies and optimizing water use efficiency under different management and environmental conditions.

Keyword:

Crop water requirement

DSSAT

Gradient boosting

Simulation

Support vector machine

Variable rate irrigation

Simulating soil hydrologic dynamics using crop growth and machine learning models

摘要

期刊

机构

引用论文