Want to make creations as awesome as this one?

Transcript

Eco-Friendly Estimation of Heavy Metal Contents in Grapevine Foliage Using In-Field Hyperspectral Data and Multivariate Analysis

By M. Mirzaei, J. Verrelst, S. Marofi, M. Abbasi, H. Azadi

From Remote Sensing - MDPI, 2019

Context

Main goals

Issues

Double interest for grapevine:

  • Tolerant to metal-induced stress ➔ phytostabilizing species
  • Already highlighted correlation between quantity of metals in the soil, in the plant and in the fruits.

Idea: quantify metal levels in grapevines to deduce soil pollution

Statement : increase of metal pollution associated with anthropogenic activities.

Existing methods for estimating foliar metal concentration: expensive and destructive

Hyperspectral methods are already known for the analysis of biochemical changes in plants

No hyperspectral method presented for the analysis of metal concentration in grapevine

+

BUT

For Cu, Zn, Pb, Cr, Cd

Determine if spectroscopy is a good estimator of foliar metal content

Compare two algorithms capable of predicting the relationship between foliar spectral response and metal concentration

Find the best indicators for each metal, especially between spectral index and direct wavelength.

1

2

3

Eco-Friendly Estimation of Heavy Metal Contents in Grapevine Foliage Using In-Field Hyperspectral Data and Multivariate Analysis

By M. Mirzaei, J. Verrelst, S. Marofi, M. Abbasi, H. Azadi

From Remote Sensing - MDPI, 2019

Cu, Zn, Pb, Cr, Cd

Materials & Methods

Repeated for the 5 metals studied

84 vines treated in total

  • Vines in a controlled environment: allows conditions (same water, soil, nutiments conditions)
  • 5 types of treatments :
    • Control: no treatment
    • MAL : metal intake according to the maximum legal dose in Iran
    • L2 to L4 : metal content 2, 3 and 4 times greater than the legal maximum

Sampling of 5 leaves per pot to measure their spectra

Each leaf :
  • Dried 24 hours
  • Crushed
  • Digested with acid
  • Filtered
  • Measurement of the metal concentration (according to the original treatment) by a Graphite-Furnace Atomic Absorption Spectrophotometer

Step 1: Scaling (normalization) of the wavelength, spectral index and heavy metal concentration values obtained previously.
Step 2: Modeling of the paths by PLS. New variables are generated taking into account both the starting variables and the response variables. The variables kept will be those with the highest factor loadings. The optimal wavelengths or spectral indices for the following modeling can be deduced.
Step 3: Modeling of the relationship between the best wavelengths or spectral index and the leaf metal concentration, for each of the 5 metals.
2 algorithms will be considered: SVM and RLM methods.

1

4

3

2

Experimental Plantations

Spectra acquisition

Actual metal concentration

Modeling metal/spectra relationship

Correlation between reflectance and metal concentration.
Observation : for Cd, maximum correlation then drop at the RDE zone (680 nm).
This observation is similar for the 4 other metals.
So there is a significant correlation between sheet metal concentration and hyperspectral datathe, and RDE zone would be a good parameter to estimate the metal concentration.

Eco-Friendly Estimation of Heavy Metal Contents in Grapevine Foliage Using In-Field Hyperspectral Data and Multivariate Analysis

By M. Mirzaei, J. Verrelst, S. Marofi, M. Abbasi, H. Azadi

From Remote Sensing - MDPI, 2019

Results & Conclusion

Between healthy and stressed leaves.
Main difference in the VIS region: plant pigments have an effect on the absorption of the plant.
However, plants under metal stress have an absorbance that decrease.

So the metals act on the pigment, and change the biochemical structure of the leaves.

Feature selection.
Using the PLS method, specific wavelengths are isolated for each metal.

VIS (visible) and RDE (red-edge region) are most likely to carry these wavelengths.

1

4

3

2

Modeling the relationship between best wavelengths or spectral index and metal concentration in leaves.

To conclude.

  • Vine is a good model to observe the effect of metals on plants.
  • Response of the vine is different from that of other plants (rice in particular).
  • SVM algorithm is the most adapted, and the most flexible by nature (combination of linear and non-linear functions).
  • Spectral indices are often more accurate than wavelengths.