Upload
others
View
9
Download
0
Embed Size (px)
Citation preview
Hyperspectral leaf reflectance for the non-destructive monitoring of urban tree diseases
Melanka Brackx, Shari Van Wittenberghe, Paul Scheunders and Roeland Samson
Observatree/IPSN Conference 23-24 februari 2016
Urban vegetation biomonitoring
Biomonitoring
Study the burden of environmental toxins to living organisms (plants)
Analyze plant indication of environmental health
2
Problem
• Traffic related air pollution
major risk to health
• Need for air quality monitoring
• Limited number of air quality monitoring stations
3
Solution?
Urban trees for biomonitoring air pollution
• Every tree can be a “monitoring station”
• Trees suffer from air pollution
Leaf physical and biochemical characteristics
Aim of my Phd research:
• Assess the potential of hyperspectral leaf reflectance as a tool for biomonitoring the urban environment
4
Hyperspectral remote sensing
5
Hyperspectral leaf reflectance
6
Measuring leaf reflectance
7
• Agrispec (ASD Inc.)
• VIS-NIR Spectroradiometer
• 350-2500nm
Monitoring of urban tree diseases
Techniques can be adapted from crop protection programs
spectral data was proposed to detect e.g. :
• blight on tomatoes
• rust on sugar cane
• feeding of ladybirds on eggplant
• sugar beet diseases
• apple scab
• citrus greening
8
Challenges
• Differences between species
• Natural variation between trees of the same species
• Different stages of infection
• Multiple infections or stress factors
• Surroundings and background in urban environment
9
Monitoring of urban tree diseases
Three different diseases:
• Powdery mildew
• Tar spot disease
• Horse-chestnut leaf miner
10
Powdery mildew
Two tree species
• pedunculate oak (Quercus robur)
• field maple (Acer campestre)
11
Powdery mildew
Two tree species
• pedunculate oak (Quercus robur)
• field maple (Acer campestre)
Characteristics
• White fungus on upper leaf side
• Affects many species
• Weakens plants and can deform young twigs
• Hyperspectral data of powdery mildew on crops have already been studied
12
Powdery mildew
13
Spectral reflectance
Powdery mildew
14
Spectral reflectance
Tar spot disease (Rhytisma acerinum)
Sycamore maple (Acer pseudoplatanus)
Characteristics
• Clearly delineated black spots appear on leaves in late summer and autumn
• Commonly affects sycamores and maples
• No effect on the tree’s long-term health
15
Tar spot disease
Spectral reflectance
16
Horse-chestnut leaf miner (Cameraria ohridella)
Horse-chestnut (Aesculus hippocastanum)
Characteristics
• Leaf-mining moth of the Gracillariidae family
• Recently introduced invasive pest
17
Horse-chestnut leaf miner
Spectral reflectance
18
Analysis
19
Vegetation indices
• Ratio of reflectance in different bands
• Related to chemical and physical leaf characteristics
• E.g. NDVI
Vegetation indices
20
Statistical analysis
21
Mixed model statistics
• For nested data: Multiple spectra were taken within leaves and multiple leaves were sampled within trees
• MIXED EFFECTS MODELS AND EXTENSIONS IN ECOLOGY WITH R (Zuur et al. 2009)
Within tree correletations
Intra class correlations (ICC)
High ICC lowers effective sample size Neff
e.g. : 7 uninfected trees 15 leaves per tree 105 samples?
22
Conclusions
• Each diseases has clear effects on the reflectance of the leaves
• Strong within-tree correlations
Other factors are also influencing the spectra
Many different trees should be sampled
• Spectral indices are a good tool for the detection of tree diseases, however better indices could be developed
23
Future perspectives
• Early stages of infection
• Discriminate multiple infections or stress factors
• Correct for surroundings and background in urban environment
24
Thank you
Research funded by BOF (Special Research fund) of the University of Antwerp in the frame of the project ’Urban vegetation biomonitoring: exploring the potential of hyperspectral remote sensing’