Estimation of leaf water potential by thermographic and spectral measurements in grapevine

Leaf water potential (ΨL) is a useful variable for the water status assessment of crops. The pressure chamber is the current method for its measurement, but has the disadvantage of being too slow and impractical when to asses an important number of plants becomes necessary, as required in Precision Viticulture. The objective of this investigation was to evaluate alternative methods for estimating ΨL, using remote sensors. One of these methods was based on spectral reflectance. This is a non destructive, quick and efficient method. Nevertheless, this technique requires statistical analysis in order to estimate the needed variables. In this study, two analyses were tested. On one side, we calculated indices from spectral values, and on the other side, we tested the Partial Least Squares analysis (PLS). The other tested method was based on Thermography. Canopy temperature (TC) is known to be related to water status. In this case, images of the canopy of the vineyard were taken, and the temperatures of each point of the assessed area were recorded. Two regression models derived from the thermographic data. One of these models was a simple regression with TC vs. ΨL; and the other was a multiple regression, including temperature, the reflectance indices NDVI (R900-R680/R900+R680) and WI (R900/R970). The assessment took place in a Malbec vineyard, in Mendoza, Argentina. Reflectance was measured during the morning and ΨL at noon, just at the same time as the images were taken. By PLS, using reflectances from 325 to 1075 nm, ΨL could be estimated. With simple and multiple regressions the following equations were obtained: ΨL = -1.21659 + 0.445078 * Tº; R2 =0,19 and ΨL = 1,83399 – 0,613766 * NDVI + 0,0447517 * TC -1,45787 * WI, R2 = 0,36, p = 0,0000. When the observed and estimated ΨL obtained by the three procedures were mapped by krigging in order to analyze the likelihood between the spatial distributions, a high level of similarity was found, despite the low regression coefficients. Apparently, the maps include spatialinformation that is absent in the regressions. Higher likelihood was found between the measured ΨL and the estimated by PLS, when compared to the other methods.