Abstract
Plant traits are features that characterise and differentiate between species. From these, certain leaf traits, such as broadleaf or coniferous, have been used to characterise the whole
plant, especially when assessing large areas of vegetation. Leaf traits do not only differentiate species but also provide information on plant health. Conventional methods of measuring leaf traits, especially at the molecular level (e.g. water, lignin, and cellulose content), are often expensive and time-consuming. Spectroscopic methods allowing the estimation of leaf traits
through empirical models are becoming a tool for accurate estimations of leaf traits.
This study identifies the most important bands in the infrared related to biochemical and
morphological leaf traits. We generated regression models for eleven leaf traits (including
organic and morphological traits), using an optimisation method of partial least squares
regression models.
This study used spectroscopic data with 6612 bands from the short to the long wave infrared
(1.4-16.0μm) from 20 plant species including herbaceous, woody, temperate and tropical
plants. Fourteen leaf traits were measured in each fresh leaf, including chemical (e.g., leaf
water content, nitrogen, cellulose) and morphological (e.g., leaf area and leaf thickness) traits.
Optimized partial least squares regression models were fitted for each leaf trait with a few
bands (4-10) as explanatory variables. For these remaining bands, we identified a
physiological explanation. From the original pool of leaf traits, cuticle thickness, bundle area
and stomata size resulted in R-squared values lower than 70%. These traits were therefore
not further considered in the analysis. Eleven leaf traits resulted in optimized models with high
R-squared values. The models selected bands from the SWIR where the leaf spectra have features correlated to lignin and cellulose content, which account on average for 28% of the fresh weight of the leaves. The biochemical traits have high R-squares when using the whole spectra (6612 bands) and also in optimized models with five to seven bands. The selected bands match with known molecular bonds of the molecules analysed (water, lignin, cellulose and nitrogen bearing molecules, such as proteins). Morphological leaf traits models have in general good R-squares, especially leaf thickness which has a high correlation with most
bands of the SWIR (1.4-2.5 μm) and a specific feature at 7.38μm.
plant, especially when assessing large areas of vegetation. Leaf traits do not only differentiate species but also provide information on plant health. Conventional methods of measuring leaf traits, especially at the molecular level (e.g. water, lignin, and cellulose content), are often expensive and time-consuming. Spectroscopic methods allowing the estimation of leaf traits
through empirical models are becoming a tool for accurate estimations of leaf traits.
This study identifies the most important bands in the infrared related to biochemical and
morphological leaf traits. We generated regression models for eleven leaf traits (including
organic and morphological traits), using an optimisation method of partial least squares
regression models.
This study used spectroscopic data with 6612 bands from the short to the long wave infrared
(1.4-16.0μm) from 20 plant species including herbaceous, woody, temperate and tropical
plants. Fourteen leaf traits were measured in each fresh leaf, including chemical (e.g., leaf
water content, nitrogen, cellulose) and morphological (e.g., leaf area and leaf thickness) traits.
Optimized partial least squares regression models were fitted for each leaf trait with a few
bands (4-10) as explanatory variables. For these remaining bands, we identified a
physiological explanation. From the original pool of leaf traits, cuticle thickness, bundle area
and stomata size resulted in R-squared values lower than 70%. These traits were therefore
not further considered in the analysis. Eleven leaf traits resulted in optimized models with high
R-squared values. The models selected bands from the SWIR where the leaf spectra have features correlated to lignin and cellulose content, which account on average for 28% of the fresh weight of the leaves. The biochemical traits have high R-squares when using the whole spectra (6612 bands) and also in optimized models with five to seven bands. The selected bands match with known molecular bonds of the molecules analysed (water, lignin, cellulose and nitrogen bearing molecules, such as proteins). Morphological leaf traits models have in general good R-squares, especially leaf thickness which has a high correlation with most
bands of the SWIR (1.4-2.5 μm) and a specific feature at 7.38μm.
| Original language | English |
|---|---|
| Pages | 1 p. + s1-s16 |
| Publication status | Published - 2017 |
| Event | 10th EARSeL SIG Imaging Spectroscopy Workshop - University of Zurich, Zurich, Switzerland Duration: 19 Apr 2017 → 21 Apr 2017 http://www.earsel.org/SIG/IS/workshops/10-IS-Workshop/ |
Conference
| Conference | 10th EARSeL SIG Imaging Spectroscopy Workshop |
|---|---|
| Country/Territory | Switzerland |
| City | Zurich |
| Period | 19/04/17 → 21/04/17 |
| Internet address |