LAI and fAPAR validation under CEOS/LPV-VALERI Landscape spatial patterns and the VALERI sampling strategy The Shandan (CH) and Gilching (D) cases Thursday, March 15th 2007
Frank Veroustraete Centre for Remote Sensing and Earth Observation (TAP) Flemish Institute for Technological Research (VITO)
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
1
Objectives, developments The objectives of the
project:
¾ To evaluate the absolute accuracy of bio- geophysical products (LAI, fAPAR, fCover) derived with a range of algorithms from large IFOV sensors (e.g. AVHRR, POLDER, VEGETATION, SEAWIFS, MSG, MERIS, AATSR, MODIS, MISR,…). ¾ To inter-compare products derived with different sensors and algorithms.
For this purpose, the
project develops:
¾ A network of sites distributed globally. ¾ A standard methodology designed to directly measure the biogeophysical variables of interest at the proper spatial and temporal scales. scales
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
2
Up-scaling strategy
(Base site)
(ESU)
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
3
Spatial scale Sensor resolution 1km 20m
Space borne SPOT4 / VEGETATION
Space borne SPOT4 / HRVIR
6m Airborne DORNIER / DAIS
dm Ground based Canon EOS, f35mm
cm
System spatial scale
Ground based Canon EOS, f80mm
Tree
Stand ESU
30-3-2007
Forest
Landscape
Region
Base site Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
4
The
30-3-2007
Shandan site (Gansu province, CH)
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
5
The
Shandan site
Base site
Homogeneous grassland
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
6
spatial sampling strategy in a homogeneous landscape A Sampling in 37 ESU’s of 20 m²
B
C 20 m
ESU(x,y) Trimble Geo-Explorer
H
F 20 m E
3 km
G
D
O
3 km
ESU(x,y) = f (Randomness) Landsat ETM+ RGB false-colour for the Shandan site
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
7
spatial sampling strategy in a homogeneous landscape ¾ Each 3 km² base site contains a number of randomly positioned and rotated 20 m² ESU’s. ¾ The angle of ESU line FB with the true North is chosen randomly. ¾ Each 1 km² pixel in the base site has 4 ESU’s. The central pixel has 5 ESU’s. ¾ In each 20 m² ESU, discrete bio- geophysical variables measurements take place at nine points (A, B, C, D, E, F, G, H, O). ¾ In each 20 m² ESU continuous biophysical variables measurements take place along eight lines (AC, CE, EG, GA, AE, CG, BF, HD), or according to the sun’s position. ¾ The ESU centre point (O) is determined with a Trimble Geoexplorer 2 GPS. All other ESU measuring points co-ordinates are thereby determined. ¾ Only one single transfer function for the whole base site is required.
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
8
Example of LAI up-scaling with ETM+ SAVI y = 0.1283x2.5536 R = 0.67; RM SE=0.18
1.2
2
1
LAI
0.8
Transfer function
0.6 0.4 0.2 0 0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
SAVI
Base site
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
9
The
Gilching site (Bavaria, D) 3 km
Base site
3 km
‘European’ patchy landscape
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
10
The
Gilching site (Bavaria) ESU(x,y) = f (landcover) Sampling in 21 ESU’s of 20 m²
A
20 m
H
B
C
O
D
3 km 20 m
G
F
E
3 km Trimble GeoExplorer
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
ESU(x,y)
11
spatial sampling strategy in a heterogeneous landscape ¾
¾ ¾ ¾ ¾ ¾ ¾
Each 3 km² base site contains a number of 20 m² ESU’s positioned according to landcover type. The angle of ESU line FB with the true North is chosen randomly. Each 1 km² pixel in the base site has a number of ESU’s according to the number of different landcover type patches. In each 20 m² ESU, discrete biophysical variables measurements take place at nine points (A, B, C, D, E, F, G, H, O). In each 20 m² ESU continuous biophysical variables measurements take place along eight lines (AC, CE, EG, GA, AE, CG, BF, HD), or according to the sun’s position. The ESU centre point (O) is determined with a Trimble Geoexplorer 2 GPS. All other ESU measuring points co-ordinates are thereby determined. As many transfer functions as land cover types for the base site.
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
12
indirect methods (for LAI) ¾
Indirect methods
9
Tracing radiation and Architecture of Canopies (TRAC – CCRS) 9 9 9 9 9
9
Hemispherical photography (HDP) (Nikon Coolpix 5000 and fishfish-eye) 9 9 9 9 9
9
Effective and ‘true LAI’ LAI’, fAPAR, fAPAR, fCover, fCover, clump size index, canopy architecture Software: INRA Avignon software package CANEYE Discrete measurements on 20 m ESU points A,B,C,D,E,F,G,H,O Time logging (Sun zenith angle) Cloud free conditions needed
Licor LAILAI-2000 9 9 9 9
30-3-2007
Gap size distribution, fAPAR, APAR, clumping index, effective LAI Continuous measurement in 20 m block (steady pace walks). Dedicated software package from CCRS. Time logging (Sun zenith angle) Cloud free conditions needed.
Measurement of effective LAI Discrete measurements on 20 m block points A,B,C,D,E,F,G,H,O Time logging (Sun zenith angle) Interesting comparisons possible with HDP and TRAC (I. Jonckheere, Jonckheere, FEM, 2006)
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
13
indirect method (TRAC)
Beech
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
14
indirect methods (for LAI) ¾
Indirect methods
9
Tracing radiation and Architecture of Canopies (TRAC – CCRS) 9 9 9 9 9
9
Hemispherical photography (HDP) (Nikon Coolpix 5000 and fishfish-eye) 9 9 9 9 9
9
Effective and ‘true LAI’ LAI’, fAPAR, fAPAR, fCover, fCover, clump size index, canopy architecture Software: INRA Avignon software package CANEYE Discrete measurements on 20 m ESU points A,B,C,D,E,F,G,H,O Time logging (Sun zenith angle) Cloud free conditions needed
Licor LAILAI-2000 9 9 9 9
30-3-2007
Gap size distribution, fAPAR, APAR, clumping index, effective LAI Continuous measurement in 20 m block (steady pace walks). Dedicated software package from CCRS. Time logging (Sun zenith angle) Cloud free conditions needed.
Measurement of effective LAI Discrete measurements on 20 m block points A,B,C,D,E,F,G,H,O Time logging (Sun zenith angle) Interesting comparisons possible with HDP and TRAC (I. Jonckheere, Jonckheere, FEM, 2006)
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
15
indirect method (HDP)
Beech
CANEYE
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
16
indirect methods (for LAI) ¾
Indirect methods
9
Tracing radiation and Architecture of Canopies (TRAC – CCRS) 9 9 9 9 9
9
Hemispherical photography (HDP) (Nikon Coolpix 5000 and fishfish-eye) 9 9 9 9 9
9
Effective and ‘true LAI’ LAI’, fAPAR, fAPAR, fCover, fCover, clump size index, canopy architecture Software: INRA Avignon software package CANEYE Discrete measurements on 20 m ESU points A,B,C,D,E,F,G,H,O Time logging (Sun zenith angle) Cloud free conditions needed
Licor LAILAI-2000 9 9 9 9
30-3-2007
Gap size distribution, fAPAR, APAR, clumping index, effective LAI Continuous measurement in 20 m block (steady pace walks). Dedicated software package from CCRS. Time logging (Sun zenith angle) Cloud free conditions needed.
Measurement of effective LAI Discrete measurements on 20 m block points A,B,C,D,E,F,G,H,O Time logging (Sun zenith angle) Interesting comparisons possible with HDP and TRAC (I. Jonckheere, Jonckheere, FEM, 2006)
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
17
indirect method (LAI-2000)
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
18
indirect method (LAI-2000)
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
19
direct method (for LAI) ¾
Direct method
¾
Field harvesting (FH) 9 9 9 9
30-3-2007
Direct measurement of ‘true’ true’ and effective LAI A square m² m² in the centre of an ESU Depends on harvesting permissions Comparison of reference ‘true’ true’ and effective LAI with indirect methods
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
20
D iam eter (cm )
direct method (for LAI)
3.0 2.5 y = -5E-05x 2 + 0.0009x + 2.2305 R2 = 0.9851
2.0 1.5 1.0 0.5 0.0 0
50
100
150
200
250 He ig h t (cm )
Mean
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
Poly . (Mean)
21
Conclusions ¾ The
sampling strategy can be applied for heterogeneous patchy
landscapes (Gilching case). ¾ The complete
procedure has been applied at the Shandan and
Gilching sites. Though both sites are very different at the landscape level, up-scaling to 1 km² pixels in both cases leads to acceptable LAI validation fields. ¾ Nevertheless, not all landscapes are suitable for the validation of biogeophysical vegetation variables. ¾ The base site should be as flat as possible. Hence (very) hilly terrain is excluded. ¾ Water bodies are unwanted in the base site.
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
22
ESU = f(Panorama)
Thanks for your attention
Yellow river
30-3-2007
Vertrouwelijk – © 2007, VITO NV – alle rechten voorbehouden
23
