Whole-plant water use and transpiration are key measurements for determining site or ecosystem water balance and understanding the response of vegetation to environmental change. However, there is a methodological tradeoff between spatial coverage and accuracy, with point based methods being very labor intensive and expensive and stand-level estimates unavoidably coarse. The application of high-resolution plot or site scale remote sensing promises to bridge the gap between these two extremes. In this experiment, we used diurnal and weekly time-series of high-resolution thermal infrared imagery to estimate transpiration and water-use as canopy complexity increased from seedling emergence to canopy closure.
Paired thermal and color images of 3 mm pixel resolution were captured from 2 m above a crop of soybean (Glycine max) grown in pots in a greenhouse from seedling emergence to canopy closure. Each week, a sequence of images was captured every 30 minutes for 7 -9 hours. Leaf gas exchange was measured every 2.5 hours on a subset of plants and micrometeorological variables logged every 10 minutes. We combined a machine-learning approach to image segmentation and a suite of leaf energy balance models to estimate whole-plant water use for every image. The model ensemble included empirical and modeled reference surface energy balance methods along with the empirical derivation of scaling relationships between canopy complexity and transpiration. Pixel, leaf, whole-plant and canopy scaling was validated against leaf gas exchange data and gravimetric determination of whole-plant and canopy water use.
Results/Conclusions
The pixel-scale remote-sensing based models that implicitly accounted for canopy complexity were able to estimate whole-plant water use to within 10% of gravimetric determination and were better able to capture canopy transpiration than averaged point-based gas exchange measures. We can use these spatially rich data to examine the heterogeneity in water fluxes from plant canopies at the millimeter to whole-plant scales at a temporal resolution of minutes to weeks. Future work will improve upon our analysis to partition among species in mixed canopies. Potential applications include the disaggregation of bulk-flux measures (e.g. eddy-covariance) into component sources and the quantification of the heterogeneity of ecosystem response to environmental change.