Foliage temperature extraction from thermal imagery for crop water stress determination

Abstract

Crop water stress determination methods from canopy temperatures, derived from the surface energy balance equations, treat the canopy temperature under the assumption that the canopy behaves as a virtual “big-leaf”, covering the ground surface. Introduction of very high-resolution thermal imagery, 0.01–0.3-m pixel size, acquired from low altitude platforms, enabled finely detailed observation of the whole canopy, raising the question how to select the relevant canopy temperatures. One approach is to select the sunlit leaves confirming to the “big leaf” energy balance paradigm. However, thermal imagery alone is incomplete and needs additional marking or synchronized visible imagery for interpretation, which makes the process complicated and expensive. The other approach, used in reference surface based water stress evaluation, is to use full frame pixel statistics without pattern recognition by selecting the mean temperature of the cold fraction from the pixel histogram. That greatly simplifies processing for large-scale aerial thermography. Here are presented the results of experiments conducted in cotton and vine grapes, where both approaches were evaluated simultaneously. Ground referenced thermal and visible images were overlapped, and sunlit, shaded and whole canopy leaves were selected for crop temperature evaluation. The pixel histograms of the same images were analyzed in a two-step method, after discarding soil pixels where their temperature was 7 °C higher than air temperature at step one, and calculation of the mean temperatures of the lowest 33 and 100 % of the remaining pixels for step two. Several crop water stress indices were compared with leaf and stem water potentials and stomatal conductance. Good agreement was found between both image segmentation and histogram analysis methods, demonstrating the suitability of both methods in canopy temperature evaluation for crop water stress evaluation.