Land Cover monitoring Current activities and future plans Markus - - PowerPoint PPT Presentation

▶
land cover monitoring
SMART_READER_LITE
LIVE PREVIEW

Land Cover monitoring Current activities and future plans Markus - - PowerPoint PPT Presentation

Oct 30, 2023 193 likes •330 views

Experiences using LUCAS data in Finnish Land Cover monitoring Current activities and future plans Markus Trm (markus.torma@ymparisto.fi) Elise Jrvenp, Pekka Hrm, Lena Hallin- Pihlatie, Suvi Hatunen, Minna Kallio Finnish

slide-1
SLIDE 1

Experiences using LUCAS data in Finnish Land Cover monitoring

Current activities and future plans Markus Törmä (markus.torma@ymparisto.fi) Elise Järvenpää, Pekka Härmä, Lena Hallin- Pihlatie, Suvi Hatunen, Minna Kallio Finnish Environment Institute SYKE NTTS2015 Brussels 11.3.2015

slide-2
SLIDE 2
  • Land monitoring in Finland

○ Several organizations are responsible for operational monitoring programmes ○ Information is integrated to produce spatial datasets ○ Needs of European Environment Agency are fulfilled

  • Corine Land Cover
  • Now also data needs of EUROSTAT taken into account

○ Develop Finnish bottom-up-approach for LM so that also statistical datasets for Lucas survey could be produced

  • Inventory of national datasets and classifications
  • Development of methodology to get Lucas compatible data

○ EUROSTAT grant for 2014: Provision of Harmonized Land Cover Information for LUCAS from the Finnish Datasets

  • Finnish Environment Institute SYKE & Natural Resources Institute

Finland LUKE

2

Introduction

slide-3
SLIDE 3

CORINE (Coordination of information on the environment) programme of the European Commission

  • Collect and coordinate the collection of

consistent information on the state of the environment

  • Corine Land Cover classification based
  • n the interpretation of satellite images

○ hierarchical classification with 44 3rd level classes

  • Finland has made CLC200, CLC2006

and CLC2012 ○ Non-standard methodology:

  • Land use from national spatial databases
  • Land cover using interpretation of satellite

images

3

Corine Land Cover

National high resolution CLC: raster with 20 m pixel size and national 4th level classes... ...which is then generalized to European CLC: vector with 25 ha minimum mapping unit.

slide-4
SLIDE 4
  • Pixelwise interpretations of satellite images from selected themes

○ Soil sealing: Density range 0-100% of impervious surfaces. ○ Forest:

  • Tree Cover Density: Density range 0-100%
  • Forest Type: Categories broadleaved and coniferous forest.

○ Grassland: Mask, ground covered by vegetation dominated by grasses and other herbaceous plants with dominantly agriculture use. ○ Wetland: Mask, areas where water is the primary factor controlling the environment. ○ Water: Mask, the permanent presence of surface water.

  • Purpose: supplement the Corine Land Cover classification by

providing higher resolution information for specific land cover themes

4

High Resolution Layers

slide-5
SLIDE 5
  • Production by European service

providers ○ Finland:

  • Soil sealing: Metria / Geoville
  • Forest: Metria / VTT
  • Grassland: INDRA
  • Wetland and Water: INDRA /

Geomatrix

  • Verification and enhancement by

member countries or service providers ○ Finland, co-operation between

  • Finnish Forest Research Institute

METLA (leads verification)

  • Finnish Environment Institute SYKE

(leads enhancement)

  • Finnish Geodetic Institute GL

5

High Resolution Layers

HRL Soil Sealing HRL Forest Type HRL Water

slide-6
SLIDE 6
  • Land Use / Cover Area frame statistical Survey by EUROSTAT

○ around 271,000 points were visited by the field surveyors in 27 European countries

  • 13482 in Finland
  • Data is used for

○ deriving land cover and land use statistics at European level ○ monitoring changes in agro-environment ○ landscape monitoring ○ ground truth for many space borne information collection activities

  • Data collection for in-situ point include

○ Land cover and use classes ○ Date, location ○ Size of area, width of feature ○ Height of trees ○ Photographs

6

LUCAS 2012

slide-7
SLIDE 7
  • Important for e.g. green house gases-reporting

○ Which one is correct?

  • LUCAS2009 & 2012: Shrubland / Grassland definitions?
  • LUCAS vs. CLC: Class definitions of Woodland / Shrubland,

Cropland / Grassland

  • Differences between FI HR & EU CLC due to generalization

7

LUCAS vs. Finnish CLC: areas of classes

LUCAS2009 Area (km2) - % LUCAS2012 FI HR CLC2012 (20m) FI EU CLC2012 (25ha) LCA: Artificial land 4888 - 1.5 5283 - 1.6 8647 - 2.6 4156 - 1.2 LCB: Cropland 20364 - 6.0 16570 - 4.9 22885 - 6.8 15543 - 4.6 LCC: Woodland CLC324 to Woodland 229490 - 68.1 243143 - 71.8 225608 - 66.7 234823 - 69.4 209053 - 61.9 241623 - 71.5 LCD: Shrubland CLC324 to Woodland 13950 - 4.1 3621 - 1.1 17026 - 5.0 7812 - 2.3 39559 - 11.7 6989 - 2.1 LCE: Grassland 10045 - 3.0 14750 - 4.4 2560 - 0.8 14142 - 4.2 LCF: Bareland 4443 - 1.3 2414 - 0.7 3320 - 1.0 1780 - 0.5 LCG: Water 34101 - 10.1 32711 - 9.7 33098 - 9.8 31906 - 9.4 LCH: Wetland 19572 - 5.8 19940 - 5.9 25273 - 7.5 21715 - 6.4

slide-8
SLIDE 8
  • Finnish Corine Land Cover 2012, versions

○ HR CLC 20m: raster with 20 m pixel size ○ HR CLC 20m with 3x3 majority filtering

  • Effect of spatial inaccuracy?

○ EU CLC 25ha: vector with 25 ha Minimum Mapping Unit

8

LUCAS vs. Finnish CLC: accuracy

CLC Overall accuracies CLC Level-1 Classwise Accuracies

HR CLC (20m) HR CLC (20m) 3x3 maj. EU CLC 25 ha MMU Level 1 – 5 classes 93.1 92.9 90.1 Level 2 – 15 classes 83.3 83.7 76.8 Level 3 – 30 classes 60.6 60.9 52.5

slide-9
SLIDE 9
  • Error matrix of Fin HR CLC 20m

○ Cxx: classification result, CLC level-2 class code ○ Lxx: Lucas (reference data), CLC level-2 class code

  • Sample size of some classes really small
  • Mixing of classes

○ Surprisingly many urban in CLC classified as forest in Lucas and vice versa

9

LUCAS vs. Finnish CLC: error matrix

L11 L12 L13 L14 L21 L22 L23 L24 L31 L32 L33 L41 L42 L51 Sum C11 10 12 10 5 9 5 51 C12 2 86 1 8 4 29 7 137 C13 1 3 5 2 1 12 C14 1 4 1 2 2 20 1 31 C21 4 16 564 185 21 12 802 C22 2 1 1 1 5 C23 1 1 2 C24 1 8 27 4 19 59 C31 55 19 1 10 4913 93 27 10 5128 C32 5 28 1 1 20 15 493 210 30 2 1 806 C33 1 4 1 1 7 C41 2 1 1 104 12 167 31 318 C42 1 1 C51 1 10 5 1130 1147 Sum 22 205 6 2 634 1 252 5614 363 229 2 1175

slide-10
SLIDE 10
  • Comparison between LUCAS-points and original and

enhanced High Resolution Layers ○ Also error estimates from HRL Verification listed

○ Commission error: Proportion of samples belonging to certain class in the classification result that were wrongly classified ○ Omission error: Proportion of samples belonging to certain class in the reference data that were not classified as such

10

LUCAS vs. HRLs: accuracy

HRL Verification LUCAS vs. Original HRL LUCAS vs. Enhanced HRL HRL Soil Sealing Commission error (%) Omiossion error (%) 187 ± 11.2 6.9 ± 1.5 72.2 ± 5.0 21.6 ± 7.7 60.1 ± 5.4 23.3 ± 6.5 HRL Forest Commission error (%) Omiossion error (%) 19.5 ± 0.5 10.5 ± 0.4 19.8 ± 1.0 7.4 ± 0.7 18.1 ± 1.0 7.1 ± 0.7 HRL Wetland Commission error (%) Omiossion error (%) 71.4 ± 2.7 31.2 ± 2.9 65.2 ± 3.4 68.0 ± 3.6 64.0 ± 3.9 25.5 ± 5.1 HRL Water Commission error (%) Omiossion error (%) 7.9 ± 1.6 0.9 ± 0.2 8.1 ± 1.5 1.3 ± 0.6 5.1 ± 1.2 1.8 ± 0.7

slide-11
SLIDE 11
  • So far, LUCAS2012 has been used for accuracy assessment
  • f Corine Land Cover and HRLs

○ Some ”oddities”, their reason?

  • Other uses:

○ Training material for LC/LU classifications

  • Drawback: small sample size for many classes
  • Better integration of LUCAS, Corine and national data sets

○ National data is already used to produce Corine data ○ National data could also be used to produce LUCAS data ○ Harmonization of various classifications ○ Multiple and better use of European in-situ data ○ To avoid duplicate work in national and European level

11

Conclusions & future directions

slide-12
SLIDE 12

Thank You for Your Attention!!!

12