Main content area

Evaluation of neural network modeling to predict non-water-stressed leaf temperature in wine grape for calculation of crop water stress index

King, B.A., Shellie, K.C.
Agricultural water management 2016 v.167 pp. 38-52
Vitis vinifera, air temperature, cultivars, decision support systems, deficit irrigation, environmental factors, evapotranspiration, fruit yield, irrigated farming, irrigation rates, leaves, linear models, monitoring, neural networks, rain, relative humidity, solar radiation, stress response, vigor, vines, water stress, wind speed, wine grapes, Idaho
Precision irrigation management of wine grape requires a reliable method to easily quantify and monitor vine water status to allow effective manipulation of plant water stress in response to water demand, cultivar management and producer objective. Mild to moderate water stress is desirable in wine grape in determined phenological periods for controlling vine vigor and optimizing fruit yield and quality according to producer preferences and objectives. The traditional leaf temperature based crop water stress index (CWSI) for monitoring plant water status has not been widely used for irrigated crops in general partly because of the need to know well-watered and non-transpiring leaf temperatures under identical environmental conditions. In this study, leaf temperature of vines irrigated at rates of 35, 70 or 100% of estimated evapotranspiration demand (ETc) under warm, semiarid field conditions in southwestern Idaho USA was monitored from berry development through fruit harvest in 2013 and 2014. Neural network (NN) models were developed based on meteorological measurements to predict well-watered leaf temperature of wine grape cultivars ‘Syrah’ and ‘Malbec’ (Vitis vinifera L.). Input variables for the cultivar specific NN models with lowest mean squared error were 15-min average values for air temperature, relative humidity, solar radiation and wind speed collected within ±90min of solar noon (13:00 and 15:00 MDT). Correlation coefficients between NN predicted and measured well-watered leaf temperature were 0.93 and 0.89 for ‘Syrah’ and ‘Malbec’, respectively. Mean squared error and mean average error for the NN models were 1.07 and 0.82°C for ‘Syrah’ and 1.30, and 0.98°C for ‘Malbec’, respectively. The NN models predicted well-watered leaf temperature with significantly less variability than traditional multiple linear regression using the same input variables. Non-transpiring leaf temperature was estimated as air temperature plus 15°C based on maximum temperatures measured for vines irrigated at 35% (ETc). Daily mean CWSI calculated using NN estimated well-watered leaf temperatures between 13:00 and 15:00 MDT and air temperature plus 15°C for non-transpiring leaf temperature consistently differentiated between deficit irrigation amounts, irrigation events, and rainfall. The methodology used to calculate a daily CWSI for wine grape in this study provided a daily indicator of vine water status that could be automated for use as a decision-support tool in a precision irrigation system.