European Forest Map Prototype with VEGETATION Work carried

  • Slides: 26
Download presentation
European Forest Map Prototype with VEGETATION Work carried out by SCOT and KUL presented

European Forest Map Prototype with VEGETATION Work carried out by SCOT and KUL presented at VEGETATION 2000 Conference with the support of CNES and contribution of JRC and VTT GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 1

Objectives of the study m Assessment of the capabilities of VEGETATION for mapping forest

Objectives of the study m Assessment of the capabilities of VEGETATION for mapping forest at regional scale m Preparation of a European forest map prototype m with simple methods applied to standard VGT products (10 days composites S 10) y Partners 4 JRC : methodology as developed for FMERS study 4 KUL : alternative approach for stratification 4 SCOT : processing 4 VTT : clustering method GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 2

Processing steps m Pan-European study area m Data : 21 composites S 10 from

Processing steps m Pan-European study area m Data : 21 composites S 10 from April 1999 to October 1999 m Other data : CORINE LC, JRC/FIRS stratification of European forest ecosystems GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 3

Processing steps m Quality of VEGETATION S 10 data y High visual quality y

Processing steps m Quality of VEGETATION S 10 data y High visual quality y However 4 problems of SWIR sensors 4 cloud mask to be improved 4 remaining directional effects Decade #2 October 1999 Status map • Clouds in dark blue • Ice in light blue • Bad SWIR sensors in red GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 4

S 10 decade #2 October 1999 NIR/SWIR/R GLC 2000 Workshop on methods - JRC,

S 10 decade #2 October 1999 NIR/SWIR/R GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 5

S 10 decade #2 October 1999 NIR/SWIR/R GLC 2000 Workshop on methods - JRC,

S 10 decade #2 October 1999 NIR/SWIR/R GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 6

Processing steps m Mapping procedure y Stratification CORINE Land Cover 4 utilisation of FIRS

Processing steps m Mapping procedure y Stratification CORINE Land Cover 4 utilisation of FIRS strata 4 alternative using landscape criteria y Spectral clustering 4 Selection of the most appropriate S 10 forest types discrimination and generation of a multi-spectral monthly composite 4 Stratified clustering of the monthly composite y Multi-temporal analysis of NDVI forest types GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 7

Stratification Two alternative location of cluster means in three ecosystem regions. A - means

Stratification Two alternative location of cluster means in three ecosystem regions. A - means similarly distributed - no likely need for pre-clustering stratification. B - cluster means on the average different in different regions - a likely need for preclustering stratification. GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 8

Stratification FIRS stratification superimposed over VGT S 10 # 2 October 1999 GLC 2000

Stratification FIRS stratification superimposed over VGT S 10 # 2 October 1999 GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 9

Stratification FIRS main regions after regrouping of main strata GLC 2000 Workshop on methods

Stratification FIRS main regions after regrouping of main strata GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 10

Stratification m Stratification alternative using landscape criteria y Proportion of forest within 5 x

Stratification m Stratification alternative using landscape criteria y Proportion of forest within 5 x 5 Km bloc from CORINE LC fragmentation patterns, e. g. Shannon index. . . y proportion of main land cover categories 0 % < 10 % 11 -40 % 41 -80 % > 81 % GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 11

Spectral clustering m Generation of a monthly composite y remaining y selection clouds in

Spectral clustering m Generation of a monthly composite y remaining y selection clouds in all S 10 products of three S 10 composites 4 D 3 August 1999 4 D 1 September 1999 4 D 2 September 1999 y averaging procedure with elimination of pixels in the SWIR band remaining pixels bad cloudy expected to minimise directional effects GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 12

Spectral clustering m Generation of a monthly composite 10 days composite #3 August 1999

Spectral clustering m Generation of a monthly composite 10 days composite #3 August 1999 monthly composite (#3 Aug / # 1 & 2 September 1999) GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 13

Monthly composite by averaging three decades D 3/Aug, D 1 and 2/Sept 1999 Spectral

Monthly composite by averaging three decades D 3/Aug, D 1 and 2/Sept 1999 Spectral clustering GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 14

Clustering m Unsupervised clustering within each stratum into 50 classes y algorithm developed by

Clustering m Unsupervised clustering within each stratum into 50 classes y algorithm developed by VTT (Finland) and tested in the framework of FMERS for European forest mapping with IRS-Wi. FS data y based pixels on homogeneous signatures within 2 x 2 y automatic sort of the clusters according to their approximate biomass values GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 15

Clustering Likely location of the target classes in the spectral range (FMERS/VTT) GLC 2000

Clustering Likely location of the target classes in the spectral range (FMERS/VTT) GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 16

Clustering results m Unsupervised clustering within each stratum into 50 classes Broadleaved 15 Broadleaved

Clustering results m Unsupervised clustering within each stratum into 50 classes Broadleaved 15 Broadleaved confusion 20 mixed Discrimination with SWIR band mixed coniferous GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 17

Clustering results GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001

Clustering results GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 18

Results m contribution of the SWIR band GLC 2000 Workshop on methods - JRC,

Results m contribution of the SWIR band GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 19

Results m NDVI temporal profiles y definition of temporal April-May June-Aug Sept-Oct indicators y

Results m NDVI temporal profiles y definition of temporal April-May June-Aug Sept-Oct indicators y distinction of three “seasons” for better discrimination of NDVI profiles related to forest types y application to the 21 S 10 composites GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 20

Colour composite NDVI three seasons GLC 2000 Workshop on methods - JRC, Ispra, 28

Colour composite NDVI three seasons GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 21

Results m Temporal profiles of main forest cover types GLC 2000 Workshop on methods

Results m Temporal profiles of main forest cover types GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 22

Results m Temporal profiles of other land cover types GLC 2000 Workshop on methods

Results m Temporal profiles of other land cover types GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 23

Results m Temporal profiles of CORINE classes • Quality of CORINE classes ? •

Results m Temporal profiles of CORINE classes • Quality of CORINE classes ? • Variability of vegetation phenology within broad land cover classes GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 24

Prototype of a European Forest Map derived from VEGETATION data Coniferous forest Mixed forest

Prototype of a European Forest Map derived from VEGETATION data Coniferous forest Mixed forest Broad-leaved forest Evergreen vegetation GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 25

CONCLUSIONS y VEGETATION S 10 products of very good quality : geometry, radiometry 4

CONCLUSIONS y VEGETATION S 10 products of very good quality : geometry, radiometry 4 but some improvements are still needed : correction of directional effects, cloud masking… y the spectral clustering approach is leading to promising results with simple methods and a few S 10 products y dramatic potential of NDVI temporal profiles for discriminating vegetation types : combination with spectral clustering ? 4 need to improve and develop new standard products on NDVI profiles, e. g. phenology, … 4 products on land cover changes GLC 2000 Workshop on methods - JRC, Ispra, 28 -29 March 2001 26