Diapositiva 1

1
LAI MAPPING AT THE TICINO EXPERIMENTAL SITES
(ITALY) R. Colombo(1), M. Meroni(1), L.
Busetto(1), G. Seufert(2) (1) Laboratorio di
Telerilevamento (LTDA) Dip. di Scienze
dell'Ambiente e del Territorio (DISAT) Università
degli Studi di Milano-Bicocca (UNIMIB) Piazza
della Scienza, 1 - 20126 Milano,
Italy roberto.colombo_at_unimib.it (2) Joint
Research Centre (JRC) Institute for Environment
and Sustainability (IES)Climate Change Unit
(CCU) Ispra (Va), Italy
2
Main objective and contents
Mapping carbon exchange in fast growing forests
(poplar plantation and short rotation forestry)

• Experimental sites overview
• LAI ground measurements techniques
• Canopy characteristics from hyperspectral data
• Plans for sharing field data and/or high
  resolution LAI products

3
Experimental sites

1. Poplar plantation equipped for gases exchanges
   measurements with eddy covariance technique 10
   plantations at regional level
2. Short rotation forestry (poplar) equipped for
   gases exchanges measurements with eddy covariance
   technique.

50 km
Ticino Natural Park is an UNESCOs Biosphere
Reserve
4
1. Poplar Plantation site

• Height 23 m
• Area 10.5 m2 ha-1

• Type I-214 clone
• Age 12 years
• Tree density 278 trees/ha

Flux tower

• Eddy covariance flux tower measurements since
  2001
• Poplar distribution at regional scale is
  40.000 ha

5
2. Short Rotation Forestry site

• Type PEGASO clone (poplar)
• Age 1 year
• Tree density 5000-12000 trees/ha

• Eddy covariance flux tower measurements since
  2004
• Height from cm to 4m (1 year)

City of Vigevano
Flux tower
Promising development of an energy industry based
on the utilization of biomass resources.
Environmental benefits of SRF as a renewable
energy source
6
Data collection overview
Field data
Licor LAI2000 PCA, HC Sunscan, SPAD, PEA, Fc (
GPS)
Field radiometric measurements
Leaf samples, biomass (NPPa/b)
7
LAI measurement techniques
Poplar plantation temporal measurements in the
test sites

• Two methods
• LICOR LAI-2000
• Nikon Hemispherical Digital Camera

LAI understory vs LAI overstory
Digital Camera
LAI-2000 the transect length, and the spatial
interval between ground samples (3.5m) were
fixed, and a single LAI value was obtained by
averaging 5 samples, collected along the
transect.
HC the transect length, and the spatial interval
between ground samples (5m) were fixed, and a
single LAI value was obtained by averaging 3
photos, collected along the transect.
8
Licor LAI-2000

1. Gap fraction measurements were acquired under
   diffuse radiation condition at sunset using two
   separate inter-calibrated sensor units with a 45
   view-cap.
2. The effective LAI was then computed excluding the
   fifth ring (67 midpoints) that reduces the
   radius of the circle segment spanned from the
   sensor FOV to approximately 1.7 times the canopy
   height. Depending on the canopy height (8 to
   24m) the area sampled ranges between 14 to 42 m.

9
Nikon Digital Camera
Sunlit components Shaded components Sky
Isodata classification
Single Threshold
10
LAI-2000 vs HC and understory evolution

• For low LAI values, HC seems to overestimate,
  while for high LAI values, HC seems to
  underestimate
• HC appears less sensitive at operators errors
  and less sensitive to sky conditions

Probable meas. error
11
LAI measurement techniques
Short Rotation Forestry measured in different
plots belonging to the same plantation (70 ha)

• Three methods
• LICOR LAI-2000
• Nikon Hemispherical Digital Camera
• Destructive (Leaf Area Meter)

1
2
3
12
Licor LAI-2000
Sampling strategy five transects in two SRF
13
Nikon Digital Camera
t2 below measurements
t1 above measurements
14
Destructive measurements vs LAI-2000
Sampling strategy six plants randomly sampled
for each SRF
Delta-T Image analysis system Dias II
Data measured between May the 27th and July the
19th 2004. LAI measurements are still in progress
15
LAI retrieval from RS

• Sources of RS data
• HYPERSPECTRAL DAIS DATA
• LANDSAT DATA
• MODIS DATA.

• Approaches
• Semi-empirical relationships
• PROSAIL inversion.

16
LAI mapping with hyperspectral data at the
experimental site
Multi view DAIS data
Retrieval approach
PROSAIL inversion with stepwise procedure for
optimal spectral sampling selection
exploitation of prior knowledge about model
variables.
Cost function Retained spectral data set Retained spectral data set Optimal spectral sampling set Optimal spectral sampling set
RMSE (m2m-2) Band number RMSE (m2m-2) Band number
LSN 0.76 36 0.60 21
LSNP 0.57 36 0.47 18
Accuracy in LAI estimation (n20) achieved with
different cost functions and different spectral
sampling sets (LSNP MVA0.40 m2/m2)
17
Comparison with ground estimates
Comparisons between LAI-2000 ground measurements
and LAI map derived from DAIS reflectance were
performed in terms of RMSELAI. The issue of
matching the spatial footprint of LAI-2000
measurements and the one of model inversion
(2.5x2.5m pixel resolution) was accomplished
considering the image spatial resolution with
respect to the ground area sampled with LAI-2000,
represented by a circle segment with radius
approximately ranging from 14 to 42m. For each
digitised transect centre sample points, it was
automatically derived the DAIS pixel located 10m
far in direction opposite to sunset SAA. A 10x10m
(4x4 pixel) window was finally centred on this
DAIS pixel in order to compute the average LAI.
By this way we were able to match the LAI-2000
spatial footprint and the fine-scale LAI
estimates, and hence to evaluate the RMSELAI.
18
LAI mapping at regional scale
Plantation distribution in Lombardia Region (land
use map 110.000)
Plantation test site areas range between 0.2
0.4 km2
19
LAI mapping with satellite data at regional scale
(Landsat)
Retrieval approach
Landsat-5 TM, July 2004
Semi-empirical regression models (OLS and RMA)

1. Training area 3x3 pixel
2. Comparison strategy between LAI maps derived from
   hyperspectral and satellite images with moderate
   resolution images is to be defined.

20
What we learned

1. HC seems to over/underestimate eLAI, regarding to
   LAI-2000, at low/high LAI values
2. HC performances are higher in short vegetation,
   less sensitive to operators errors and to sky
   conditions
3. Retrieving LAI by means of HC may need variable
   threshold for gaps classification
4. eLAI measured with LAI-2000 underestimates the
   LAI (as measured by destructive sampling)
5. PROSAIL model can be successfully inverted with
   multiview hyperspectral DAIS data for LAI
   retrieval in poplar plantation
6. Semi-empirical models perform good in LAI mapping
   with Landsat data

21
Plans for sharing field data and/or high
resolution LAI products

1. Full availability to exchange field data,
   hyperspectral data and high resolution LAI maps
   (years 2001, 2003, 2004)
2. An intensive ground campaign is foreseen in 2005
   at the SRF (LAI, Fapar, water, pigment, NPPa/b,
   reflectance spectra....)
3. An hyperspectral CASI acquisition is foreseen for
   2005
4. Possibility to modify/improve our next campaigns
   following the results of the discussion of this
   meeting
5. Availability to collaborate in other projects for
   intensive campaigns measurements (in progress and
   foreseen).
6. Challenging issues are posed by LAI retrieving in
   our patchy agricultural area with MODIS data
   (unmixing approach is needed).
7. Availability to collect LAI measurements in one
   pair of 1km x 1km next spring for MODIS products
   validation purposes.