Detecting tree mortality severity using MODIS
satellite data
Jan Verbesselt
14 ARSPC Conference
30 September 2008
Tree mortality
• Severe Ips bark beetle damage
CSIRO. Detecting tree mortality severity
Tree mortality
• Caused by drought stress
CSIRO. Detecting tree mortality severity
Precipitation mm/year
Current monitoring systems
• Aerial sketch mapping and field surveys
CSIRO. Detecting tree mortality severity
May 2007
Current monitoring systems
• Aerial and field surveys
• Few highly skilled individuals in the country
• Expensive and labour demanding
• Infrequent coverage (aircraft hire, cloud coverage)
• -> only annually
• A critical need for a forest health monitoring system
CSIRO. Detecting tree mortality severity
• A critical need for a forest health monitoring system
• MODIS satellite imagery
• Frequent coverage over large areas
• Spatial resolution: 250 - 1000m
Detecting tree mortality
• Aim
• Assess sensitivity of MODIS satellite data
• Which VI’s are most sensitive to tree mortality?
• When are VI’s most sensitive?
• Using aerial ADS 40 digital imagery
• 15 cm spatial resolution
CSIRO. Detecting tree mortality severity
• 15 cm spatial resolution
• Visual dead tree detection
Data
CSIRO. Detecting tree mortality severity
MODIS satellite data
• MOD13Q1 product
• 250 m spatial resolution
• 16 daily composited images
• 2000-2008
• Vegetation indices
• Normalized Difference Vegetation Index (NDVI)
• ~ “greenness”
NIR redNDVI
NIR red
−=
+
CSIRO. Detecting tree mortality severity
• ~ “greenness”
• Leaf chlorophyll, leaf area, canopy cover and architecture
• Enhanced Vegetation index (EVI)
• Improved sensitivity in high biomass regions
• Minimizing soil and atmosphere influence
* red (620-670 nm), near-Infrared (NIR; 841 - 876 nm), blue (459-479 nm)
6 7.5 1
NIR redEVI
NIR red blue
−=
+ × + × +
Time series analysis
CSIRO. Detecting tree mortality severity
Time series analysis
CSIRO. Detecting tree mortality severity
Time series analysis
CSIRO. Detecting tree mortality severity
Time series analysis
• Apply the Relative Index (RI)• To remove seasonality
• Reduce general atmospheric noise
• Increase sensitivity
( , )1
x tVIRI
VI= −
CSIRO. Detecting tree mortality severity
• VI median is median at time t ( t: 2000 -> 2007) for current selected pixels within an area of 10 by 10 km.
( )
1median t
RIVI
= −
Time series analysis
CSIRO. Detecting tree mortality severity
Statistical analysis
• Tree Mortality
• VI2005 ~ amount of alive trees in 2005
• VI2007 ~ amount of alive trees in 2007
• Analysis
• 4 indices: NDVI, EVI, RI.NDVI, RI.EVI
2005 2007TreeMortality VI VI= −
CSIRO. Detecting tree mortality severity
• 4 indices: NDVI, EVI, RI.NDVI, RI.EVI
• 3 periods: Median VI of whole year, Feb – Mar (a), and Jun – Aug (b)
• 2 manifestations: Difference approach, 2007 value
• Modelling
• OLS model, step-wise selection of variables based on AIC value
• correct for over-estimation by resampling of the OLS model
Preliminary results
• 4 VI based models
model R2 (*)
NDVI 0.32
RI.NDVI 0.32
EVI 0.13
RI.EVI -0.07
CSIRO. Detecting tree mortality severity
(*) index corrected for over-estimation
RI.EVI -0.07
Discussion
• Remote sensing challenges
• Tree Mortality only up to maximally 30%
• Drought stress caused overall decline in forest health in 2007
(~ defoliation)
• Understorey influence
(e.g. weeds)
CSIRO. Detecting tree mortality severity
Conclusions & further work
• NDVI is more sensitive then EVI
• Sensitivity changes within a year
• NDVI : Feb-Mar
• EVI: Jun-Aug
⇒Development of specific change detection metrics for each VI
CSIRO. Detecting tree mortality severity
• Further work
• Verify EVI
• Include SWIR based indices
• Derive amount of alive trees in 2005 and 2007 from Quickbird data
• Tree density estimation using local Maxima and image segmentation
Sustainable Ecosystems
Dr. Jan Verbesselt, Dr. Darius Culvenor
Phone: 03/95452265
Email: [email protected]
Contact Us
Phone: 1300 363 400 or +61 3 9545 2176
Email: [email protected] Web: www.csiro.au
Thank you
Acknowledgements This work was undertaken within the CRC for Forestry
Program 1: Monitoring Measuring. Discussions with Andrew Robinson,
Christine Stone, and Angus Carnegie contributed to this study.