Data-driven models for canopy temperature-based irrigation scheduling

Authors: King, Bradley - Brad; Shellie, Krista; Tarkalson, David; LEVIN, ALEXANDER - Oregon State University; SHARMA, VIVEK - University Of Florida; Bjorneberg, David - Dave

Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/7/2020
Publication Date: 10/14/2020
Citation: King, B.A., Shellie, K., Tarkalson, D.D., Levin, A.D., Sharma, V., Bjorneberg, D.L. 2020. Data-driven models for canopy temperature-based irrigation scheduling. Transactions of the ASABE. 63(5):1579-1592. https://doi.org/10.13031/trans.13901.
DOI: https://doi.org/10.13031/trans.13901

Interpretive Summary: Normalized crop canopy temperature, termed crop water stress index (CWSI), was proposed over 40 years ago as an irrigation management tool but has experienced limited adopted in production agriculture. Development of generalized crop-specific upper and lower reference temperature is critical for implementation of CWSI-based irrigation scheduling. The objective of this study was to develop and evaluate data driven models for predicting reference canopy temperatures needed to compute CWSI for sugarbeet and wine grape. Reference canopy temperatures for sugarbeet and wine grape were predicted using artificial neural network and regression models developed using measured canopy temperatures of sugarbeet, grown in Idaho and Wyoming, and wine grape, grown in Idaho and Oregon, over 5 years under full and severe deficit irrigation. The neural network models provided excellent prediction of lower reference canopy temperatures for sugarbeet and wine grape. Daily CWSI of sugarbeet as well correlated with irrigation and soil water status at both locations. A quadratic relationship between daily CWSI and midday leaf water potential of Malbec and Syrah wine grape was significant and demonstrated that the methodology developed in this study to compute CWSI could be used to continuously monitor water stress of wine grape.

Technical Abstract: Normalized crop canopy temperature, termed crop water stress index (CWSI), was proposed over 40 years ago as an irrigation management tool but has experienced limited adopted in production agriculture. Development of generalized crop-specific upper and lower reference temperature is critical for implementation of CWSI-based irrigation scheduling. The objective of this study was to develop and evaluate data driven models for predicting reference canopy temperatures needed to compute CWSI for sugarbeet and wine grape. Reference canopy temperatures for sugarbeet and wine grape were predicted using machine learning and regression models developed using measured canopy temperatures of sugarbeet, grown in Idaho and Wyoming, and wine grape, grown in Idaho and Oregon, over 5 years under full and severe deficit irrigation. Lower reference temperatures were estimated using neural network models with Nash-Sutcliffe model efficiencies exceeding 0.88 and root mean square error less than 1.1 degree Celsius. The relationship between well-watered canopy temperature minus ambient temperature and vapor pressure deficit was represented by a linear model that maximized the regression coefficient rather than minimized the sum of squared error. The linear models were used to estimate upper reference temperatures nearly double values reported in previous studies. Daily CWSI calculated as the average of 15-min values determined between 13:00 and 16:00 MDT for sugarbeet and 13:00 and 15:00 local time for wine grape was well correlated with irrigation events and amounts. A quadratic relationship between daily CWSI and midday leaf water potential of Malbec and Syrah wine grape was significant (p<0.001) with an R2 of 0.67. The data driven models developed in this study to estimate reference temperatures permit automated calculation of CWSI for effective assessment of crop water stress, however, wet canopy conditions or solar radiation < 200 W m-2 can result in irrational values of CWSI. Automated calculation of CWSI using the methodology of this study would need to check for wet canopy or low solar radiation conditions and omit calculation of CWSI if determined to be probable.