generating fine resolution leaf area index maps for boreal forests of finland.pptx
TRANSCRIPT
www.helsinki.fi/yliopisto
Generating fine resolution leaf area index maps for boreal forests of
Finland
Janne Heiskanen, Miina Rautiainen, Lauri Korhonen,
Matti Mõttus, Pauline Stenberg
IGARSS 2011, 24–29 July 2011, Vancouver, Canada
www.helsinki.fi/yliopisto 2
Leaf area index (LAI)• Key variable in modeling vegetation-atmosphere
interactions, particularly carbon and water cycle
• One half of the total leaf surface area per unit ground surface area
Several global-scale LAI products, but finer spatial resolution (e.g. Landsat and SPOT) is needed to describe the spatial heterogeneity of LAI
Empirical, vegetation index (VI) based methods are typically used in fine resolution mapping, but more physically-based approach could generalize better in space and time, and between sensors
Introduction
www.helsinki.fi/yliopisto 3
Generate fine-resolution forest LAI maps for Finland using satellite image mosaics at 25 m resolution
LAI estimation methods• Empirical model based on reduced simple ratio (RSR)
• Inversion of forest reflectance model (PARAS)
Compare upscaled LAI maps with MODIS LAI (V005)
Objectives
www.helsinki.fi/yliopisto
> 1000 field plots measured with LAI-2000 PCA or hemispherical photography (2000–2008)
SPOT HRVIR and Landsat ETM+ images from the same summer (atmospherically corrected)
LAI field measurements
www.helsinki.fi/yliopisto 5
Requires min and max SWIR reflectance factors
Best model fit if values are determined separately for each scene (scene-specific RSR) instead of general values (global RSR)
RSR-Le regression models
min_max_
max_
SWIRSWIR
SWIRSWIR
red
NIRRSR
Le
RSR
www.helsinki.fi/yliopisto
PARAS forest reflectance model
L
LLgroundcanopy
p
pifcgfcgfBRF
1202121 )(),()()(
θ1 and θ2: view and Sun zenith anglescgf = canopy gap fractionρground = BRF of the forest background
f= canopy upward scattering phase function i0(θ2 ) = canopy interceptance ωL = leaf albedo
Photon recollision probability (p): the probability by which a photon scattered from a leaf (or needle) in the canopy will interact within the canopy again
Rautiainen & Stenberg 2005, RSE
ground component canopy component
p
p
pp
www.helsinki.fi/yliopisto
Can use field measurements of canopy structure and optical properties of foliage and understory
Calculation of p from LAI-2000 PCA data (Stenberg 2007, RSE)
30,000 simulations for training neural networks
• LAI-2000 PCA (cgf, p)
• Leaf (needle) albedo from images
• Mixtures of forest understory spectra (Lang et al. 2001)
• Red, NIR and SWIR
PARAS simulations
Empirical data
BRFred
BR
FN
IR
LAI
DIFNp
1
1
DIFN = ‘diffuse non-interceptance’
www.helsinki.fi/yliopisto
Accuracy at an independent validation site
PARAS
RMSE = 0.59 (25.1%)Bias = -0.27 (-11.4%)
r = 0.88
Measured Le
Est
ima
ted
Le
RSR (scene-specific)
Measured Le
Est
ima
ted
Le
RMSE = 0.57 (24.2%)Bias = -0.30 (-12.7%)
r = 0.90
Heiskanen et al. 2011, JAG
www.helsinki.fi/yliopisto 9
Country-wide mosaics (IMAGE2000/2006) produced by Finnish Environmental Institute (SYKE)
• 37 Landsat ETM+ scenes, 1999–2002
• 83 IRS P6 LISS and SPOT-4 HRVIR scenes, 2005 or 2006
Input data for Finnish Corine Land Cover databases (CLC2000/2006)
Images have been atmospherically corrected, but red and SWIR reflectance factors were calibrated using satellite data from the field sites
Satellite image mosaics
www.helsinki.fi/yliopisto 10
RSR
Effective LAI (Le)LAI
estimation methods
Correction for shoot-level clumping
LAI
Validation
Field plots(6 sites)
MODIS LAI Intercomparison
Satellite image mosaics
(2000/2006)
Land cover maps
(2000/2006)
Heiskanen et al. 2011, JAG
www.helsinki.fi/yliopisto
Scene-specific RSR
SWIR BRF Forest maskScene-boundaries (2006)
+ +
www.helsinki.fi/yliopisto
Data set ρSWIR Range Mean SD
IMAGE2000(n = 37)
Min 0.057–0.118 0.082 0.016
Max 0.208–0.276 0.235 0.012
IMAGE2006(n = 83)
Min 0.063–0.133 0.089 0.019
Max 0.193–0.285 0.221 0.015
Global values based on sample plotsρSWIR_min = 0.063ρSWIR_max = 0.244
Scene-specific RSR:ρSWIR_min, ρSWIR_max
www.helsinki.fi/yliopisto
Accuracy at modelling sites
RSR (scene-specific) RSR (global) PARAS
Measured Le
Est
ima
ted
Le
Measurement site Date RSR (scene-specific) RSR (global) PARAS
LAI Image R2 RMSE R2 RMSE R2 RMSE
Puumala (n = 395) 6/2000 2.8.1999 0.64 0.52 0.64 0.54 0.69 0.56
Saarinen (n = 370) 7/2001 27.6.2001 0.61 0.65 0.61 0.72 0.66 0.84
Hirsikangas (n = 24) 5–6/2005 6.8.2006 0.42 0.55 0.42 0.55 0.48 0.57
Rovaniemi (n = 20) 6/2005 2.7.2006 0.70 0.54 0.75 0.29 0.79 0.46
Tähtelä (n = 261) 6/2006 7.6.2006 0.45 0.55 0.45 0.35 0.38 0.38
Hyytiälä (n = 73) 6–7/2008 17.7.2006 0.58 0.73 0.57 0.79 0.63 0.88
www.helsinki.fi/yliopisto
LAI maps (global RSR)
2000 2006
www.helsinki.fi/yliopisto
LAI≤ 1.0
1.1–2.0
2.1–3.0
3.1–4.0
4.1–5.0
5.1–6.0
> 6.0
LAI 2006 MODIS LAI(IMAGE2006 dates)
MODIS LAI(July average 2002–2010)
White = non-forest (< 50% forest), Black = clouds Good quality (main algorithm with or without saturation)
LAI 2006 and MODIS LAI (V005)
www.helsinki.fi/yliopisto
Comparison with MODIS LAI
MODIS LAI includes also understory LAI
Scene-wise averages
www.helsinki.fi/yliopisto 17
Empirical and forest reflectance model based methods for estimating LAI
• Empirical model based on RSR (global) was selected for generating LAI maps for Finnish forests
• Realistic LAI patterns but the highest values are underestimated
Reflectance data and land cover maps• Systematic difference in red and SWIR bands
• Phenological differences between the images
• Clumping correction
Further validation of MODIS LAI (V005)
Conclusions
18www.helsinki.fi/yliopisto
Thank you!
http://www.mm.helsinki.fi/~mxrautia/lai/index.htm
Heiskanen, J, M Rautiainen, L Korhonen, M Mõttus & P Stenberg (2011). Retrieval of boreal forest LAI using a forest reflectance model and empirical regressions. International Journal of Applied Earth Observation and Geoinformation 13: 595–606. doi:10.1016/j.jag.2011.03.005