U.S. flag

An official website of the United States government

Dot gov

Official websites use .gov
A .gov website belongs to an official government organization in the United States.


Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.


Main content area

Wine grape cultivar influence on the performance of models that predict the lower threshold canopy temperature of a water stress index

B.A. King, K.C. Shellie
Computers and electronics in agriculture 2018 v.145 pp. 122-129
air temperature, canopy, cultivars, data collection, model validation, models, prediction, regression analysis, relative humidity, solar radiation, uncertainty, variance, water stress, wind speed, wine grapes
The calculation of a thermal based Crop Water Stress Index (CWSI) requires an estimate of canopy temperature under non-water stressed conditions (Tₙwₛ). The objective of this study was to assess the influence of different wine grape cultivars on the performance of models that predict Tₙwₛ. Stationary infrared sensors were used to measure the canopy temperature of the wine grape cultivars Malbec, Syrah, Chardonnay and Cabernet franc under well-watered conditions over multiple years and modeled as a function of climatic parameters – solar radiation, air temperature, relative humidity and wind speed using multiple linear regression and neural network modeling. Despite differences among cultivars in Tₙwₛ, both models provided good prediction results when all cultivars were collectively modeled. For all cultivars, prediction error variance was lower in neural network models developed from cultivar-specific datasets than regression models developed from multi-cultivar datasets. Overall, the cultivar-specific models had less prediction error variance than multi-cultivar models. Multi-cultivar models generally resulted in prediction bias whereas cultivar-specific models eliminated the prediction bias. All predictive models had an uncertainty of ±0.1 in calculation of the CWSI despite significantly different prediction error variance between models.