The planimetric area of grapevine leaf blades (LA) is required as input data in many grapevine growth models and quantitative studies of the soil/plant/atmosphere continuum. A subset of 300 scanned grapevine leaves was used to identify and compare allometric statistical models predicting the leaf area of grapevines (cultivars Cabernet Sauvignon and Shiraz). The mean absolute error (MAE), root mean square error (RMSE), and Δ (RMSE – MAE) were used as discriminatory criteria. Six families of models drawn from the literature were compared with stepwise regression using up to six possible predictor variables. Each family was fitted to each cultivar for three vineyard sites. Generic models were computed by aggregating the data across sites and cultivars. The Queensland (stepwise regressions) family performed best, closely followed by Elsner2 and Montero. The MAE of some generic models was at times less than that of their components because of the influence of sites and/or cultivars. Site- and cultivar-specific stepwise regressions are generally the most accurate methodology for estimating leaf surface area. Simple models were generally less accurate than models integrating several predictor variables.
- Copyright © 2010 by the American Society for Enology and Viticulture