Adobe Systems
Estimation of vegetation parameters from hyperspectral data
̶T. Slanináková, 15.9.2023, SitSem
̶
1

Context
Estimation of vegetation parameters from Hyperspectral Data
2
EnviLab: Platform for providing data, visualizations and analyses of ecosystems in CZ
̶Why?
Features:
̶Aggregating and providing geo data from remote sensing via Web, API
̶Hosting visualizations, analyses, and results of various research groups
̶Providing data analyses
̶Analysis/visualization of bark beetle’s reproduction/spread over Czech forests
̶Vegetation parameters

Context
̶Lots of multi/hyperspectral data from remote sensing
̶
Estimation of vegetation parameters from Hyperspectral Data
3
Source: Single-Cell Analysis Using Hyperspectral Imaging Modalities.

Context
̶Lots of multi/hyperspectral data from remote sensing
̶
̶ESA – Sentinel 1,2,3,5P missions (program COPERNICUS)
̶Many more data sources in the future (FLEX ‘25, CHIME ‘28)
̶CzechGlobe (CAS) – Airborne missions
̶
̶Use of such data:
̶Mining/geology
̶Urban land-use mapping
̶Agriculture: health of the crops
̶Vegetation analysis
̶
̶
Estimation of vegetation parameters from Hyperspectral Data
4
Review of Top-of-Canopy Sun-Induced Fluorescence (SIF) Studies from Ground, UAV, Airborne to
Spaceborne Observations

Context
̶Monitoring the health of forests through assessing vegetation parameters
̶Why:
̶Interesting for CzechGlobe, foresters
̶Attractive for us as a nice use case for EnviLab
̶Interesting research problem with active scientific community
̶
̶
Estimation of vegetation parameters from Hyperspectral Data
5

Outline
1.Context
2.Problem definition
3.Approach
4.Results
5.Next steps
̶
Estimation of vegetation parameters from Hyperspectral Data
6

Problem definition
̶Given data from remote sensing (satellite, airborne) train a model to predict vegetation
parameters
̶
̶
Estimation of vegetation parameters from Hyperspectral Data
7
Data
A diagram of a different color spectrum Description automatically generated with medium confidence

Problem definition
̶Given data from remote sensing (satellite, airborne) and simulated spectra, train a model to
predict vegetation parameters
̶
̶
Estimation of vegetation parameters from Hyperspectral Data
8
Data
Simulated spectra
Vegetation parameters
Spectra
A diagram of a different color spectrum Description automatically generated with medium confidence

Approach
Estimation of vegetation parameters from Hyperspectral Data
9
Our approach

Approach
Estimation of vegetation parameters from Hyperspectral Data
10
Our approach

Approach
Estimation of vegetation parameters from Hyperspectral Data
11
Our approach

Approach
Estimation of vegetation parameters from Hyperspectral Data
12
Our approach

Approach
Estimation of vegetation parameters from Hyperspectral Data
13
CzechGlobe’s approach

Results
Estimation of vegetation parameters from Hyperspectral Data
14
Chlorophyll
Carotenoids
Water content
A graph with blue dots and a line Description automatically generated A graph with blue dots
Description automatically generated
On hand-collected validation data (different locations, different times)
Best model based on (n)rmse
Rmse
nrmse
1.27
8.63
0.00087
0.2
0.184
0.301
A graph with blue dots Description automatically generated

Next steps
̶Improve the quality of labels
̶More detailed simulated spectra
̶Involve more validation data into the process
̶Include more data
̶Extend with data from different time segments (unclouded)
̶Extend with airborne data
̶Try more robust models
̶Prithvi-100M-multi-temporal-crop-classification
̶
Estimation of vegetation parameters from Hyperspectral Data
15