Important leaf traits in the identification of stressed plants with hyperspectral thermal infrared spectroscopy

Paper 1258 - Session title: Land Posters.

Plant stress detection is an important issue in the management and conservation of natural and anthropogenic environments. Remote sensing is a tool that can monitor large areas to detect plant stress. Most of research on plant stress detection has been focusing on visible-near infrared (VISNIR). However, the thermal infrared (TIR) seems to contain valuable information on Leaf Water Content (LWC) and structural and microstructural traits of leaves, which could be used as proxies for plant stress detection.

This study explores changes in the TIR in relation to imposed plant stress. Sixty plants of Rhododendron cf. catawbiense were used in an two way factorial experiment with water and temperature stress. In total 75 leaves were tracked and measured after the experiment to identify changes in leaf traits and spectral changes in the TIR (1.4 to 16 µm). All stress treatments (dry-cold, wet-cold, dry-ambient) showed significantly different spectra compared to leaves from normal growing conditions (wet-ambient). Leaves of stressed plants displayed similar changes in LWC with a reduction to 41.1% in the most extreme conditions (dry-cold) compared to 61.8% of LWC in the control (ambient-wet). Also stressed leaves tended to have thicker leaves and upper cuticles, increases in stomata density and reductions in stomata size as strategies to cope with dehydration and extreme temperatures. Multinomial models were fit using leaf trait changes as proxies to determine plant stress.

The stress leaves after the experiment, also had different spectra in the TIR compared to the control plants (ambient-wet conditions). Descriptive statistics, PLSR analysis and multinomial models were conclusive in the identification of LWC, leaf thickness, lignin content, leaf area and stomata density, as the leaf traits more affected by the stress treatments and which can be used as proxies for stress detection. PLSR optimized models (PLSRopt) were fitted to predict changes in leaf traits using changes in distinct bands of the spectra. Most bands selected for these predictive models on the most important leaf traits (e.g. lignin, cellulose and LWC) have been reported as molecular vibrations related to those compounds in other studies.