Skip to main content
ARS Home » Plains Area » Fort Collins, Colorado » Center for Agricultural Resources Research » Water Management and Systems Research » Research » Publications at this Location » Publication #394390

Research Project: Improving Crop Performance and Precision Irrigation Management in Semi-Arid Regions through Data-Driven Research, AI, and Integrated Models

Location: Water Management and Systems Research

Title: Evaluation of artificial intelligence algorithms with sensor data assimilation in estimating crop evapotranspiration and crop water stress index for precision irrigation water management

item KATIMBO, ABIA - University Of Nebraska
item RUDNICK, DARAN - University Of Nebraska
item ZHANG, JINGWEN - University Of Illinois
item GE, YUFENG - University Of Nebraska
item HEEREN, DEREK - University Of Nebraska
item DeJonge, Kendall
item FRANZ, TRENTON - University Of Nebraska
item SHI, YEYIN - University Of Nebraska
item LIANG, WEIZHEN - University Of Nebraska
item QIAO, XIN - University Of Nebraska
item NAKABUYE, HOPE - University Of Nebraska
item DUAN, JIAMING - University Of Nebraska
item KABENGE, ISA - Makerere University

Submitted to: Smart Agricultural Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/7/2023
Publication Date: 1/20/2023
Citation: Katimbo, A., Rudnick, D.R., Zhang, J., Ge, Y., DeJonge, K.C., Franz, T.E., Shi, Y., Liang, W., Qiao, X., Heeren, D.M., Kabenge, I., Nakabuye, H.N., Duan, J. 2023. Evaluation of artificial intelligence algorithms with sensor data assimilation in estimating crop evapotranspiration and crop water stress index for precision irrigation water management. Smart Agricultural Technology. 4. Article e100176.

Interpretive Summary: Artificial intelligence (AI) models have potential to automate irrigation decision scheduling tools. This study evaluated the performance of several AI models to estimate crop evapotranspiration (ETc) and crop water stress index (CWSI), both of which can provide crop water requirements in real-time. Results found that temperature, relative humidity, and soil water content are the best inputs to predict ETc, and the addition of solar radiation and vapor pressure deficit to those variables were best for predicting CWSI. The results of this study will be helpful to manufacturers of irrigation systems, sensors, and decision support systems to create technologies that will ultimately help farmers schedule and apply irrigation more efficiently under declining water resources.

Technical Abstract: Precise irrigation water management improves water use efficiency and yields when smart irrigation scheduling tools are used under limited water supply. Simple and automated irrigation decision support system can serve this purpose. One which uses artificial intelligence (AI) models and incorporates climate and soil moisture measurements to manage irrigation. Therefore, current study evaluated the performance of AI models – deep learning (DL) and machine learning (ML) models and their ensembles in estimating crop evapotranspiration (ETc) and crop water stress index (CWSI) and both can provide crop water requirements and report stress in real-time. Furthermore, a closed – loop irrigation decision support system (i.e., CLDSS) was proposed for more insights about applying AI models to decide precise timing and irrigation depth. ETc was computed using FAO56 single crop coefficient and neutron soil moisture measurements whereas CWSI was calculated using Jackson’s theoretical approach and canopy temperature measured by infrared thermometers (IRTs). Prediction evaluation based on total ranking scores of used metrics (r2, RMSE, MAE, and MAPE) across all models and input combinations showed model performance on training set while estimating ETc as Stacked-Regr > SVM > Stacked-RF > CatBoost > Weighted-Ensemble > ANN > LSTM > MLR > RF > kNN, in contrast to CWSI as CatBoost > RF > kNN > Weighted-Ensemble > ANN > Stacked-Regr > LSTM > Stacked-RF > SVM > MLR. Considering training and testing periods, metrics including r2, RMSE, MAE, and MAPE, respectively, ranged between 0.638 – 0.999, 0.203 – 1.029 mm d-1, 0.158 – 0.807 mm d-1, and 3.1 – 18.8% for ETc, and 0.401 – 0.999, 0.040 – 0.247 (unitless), 0.020 – 0.191 (unitless), and 2.5 – 15.1% for CWSI. Overall, ensembles specifically Stacked Regression predicted ETc better than other models whereas CatBoost outperformed other models during CWSI estimation. The best input combinations for ETc and CWSI predictions were: four(4) (Tmin, Tmax, RH, ' SWC @0.9m) and ten(10) (Ta, Rs, RH, VPD, u2, ' SWC @0.9m), respectively. Proposed CLDSS is one with a central control station which collects and consolidates data into useful format for all models, best models are voted after training and testing on the input variables then used to automatically decide irrigation depth and timing. However, the designed system must be compared to conventional methods (e.g., soil moisture measurements, etc.) before recommending it to producers, either through simulation with process-based agricultural models (like AQUACROP) or perform actual field experiments to evaluate its ability to reach optimized yields as well as improving water use efficiency and water savings.