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
Introduction ● 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
Corine Land Cover National high resolution CLC: raster with 20 m pixel size and national 4th level classes... 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 on the interpretation of satellite images ○ hierarchical classification with 44 3rd ...which is then generalized to European CLC: vector with level classes 25 ha minimum mapping unit. ● Finland has made CLC200, CLC2006 and CLC2012 ○ Non-standard methodology: • Land use from national spatial databases • Land cover using interpretation of satellite images 3
High Resolution Layers ● 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
HRL Soil Sealing High Resolution Layers ● Production by European service providers ○ Finland: • Soil sealing: Metria / Geoville HRL Forest Type • Forest: Metria / VTT • Grassland: INDRA • Wetland and Water: INDRA / Geomatrix ● Verification and enhancement by member countries or service providers ○ Finland, co-operation between HRL Water • Finnish Forest Research Institute METLA (leads verification) • Finnish Environment Institute SYKE (leads enhancement) • Finnish Geodetic Institute GL 5
LUCAS 2012 ● 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 vs. Finnish CLC: areas of classes ● Important for e.g. green house gases-reporting ○ Which one is correct? LUCAS2009 LUCAS2012 FI HR CLC2012 FI EU CLC2012 Area (km 2 ) - % (20m) (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 229490 - 68.1 243143 - 71.8 225608 - 66.7 209053 - 61.9 CLC324 to Woodland 234823 - 69.4 241623 - 71.5 LCD: Shrubland 13950 - 4.1 3621 - 1.1 17026 - 5.0 39559 - 11.7 CLC324 to Woodland 7812 - 2.3 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 ● 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: accuracy ● 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 CLC Overall accuracies CLC Level-1 Classwise Accuracies HR HR CLC EU CLC CLC (20m) 25 ha (20m) 3x3 maj. MMU Level 1 93.1 92.9 90.1 – 5 classes Level 2 83.3 83.7 76.8 – 15 classes Level 3 60.6 60.9 52.5 – 30 classes 8
LUCAS vs. Finnish CLC: error matrix ● Error matrix of Fin HR CLC 20m ○ Cxx: classification result, CLC level-2 class code ○ Lxx: Lucas (reference data), CLC level-2 class code L11 L12 L13 L14 L21 L22 L23 L24 L31 L32 L33 L41 L42 L51 Sum C11 10 12 0 0 10 0 5 0 9 5 0 0 0 0 51 C12 2 86 1 0 8 0 4 0 29 7 0 0 0 0 137 C13 0 1 3 0 0 0 0 0 5 2 0 0 0 1 12 C14 1 4 0 1 2 0 2 0 20 1 0 0 0 0 31 C21 4 16 0 0 564 0 185 0 21 12 0 0 0 0 802 C22 0 0 0 0 2 0 1 0 1 1 0 0 0 0 5 C23 0 0 0 0 0 0 1 0 1 0 0 0 0 0 2 C24 0 1 0 0 8 0 27 0 4 19 0 0 0 0 59 C31 0 55 0 0 19 1 10 0 4913 93 0 27 0 10 5128 C32 5 28 1 1 20 0 15 0 493 210 0 30 2 1 806 C33 0 0 0 0 0 0 1 0 4 1 0 0 0 1 7 C41 0 2 1 0 1 0 0 0 104 12 0 167 0 31 318 C42 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 C51 0 0 0 0 0 0 1 0 10 0 0 5 0 1130 1147 Sum 22 205 6 2 634 1 252 0 5614 363 0 229 2 1175 ● 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. HRLs: accuracy ● Comparison between LUCAS-points and original and enhanced High Resolution Layers ○ Also error estimates from HRL Verification listed HRL LUCAS vs. LUCAS vs. Verification Original HRL Enhanced HRL HRL Soil Sealing Commission error (%) 187 ± 11.2 72.2 ± 5.0 60.1 ± 5.4 Omiossion error (%) 6.9 ± 1.5 21.6 ± 7.7 23.3 ± 6.5 HRL Forest Commission error (%) 19.5 ± 0.5 19.8 ± 1.0 18.1 ± 1.0 Omiossion error (%) 10.5 ± 0.4 7.4 ± 0.7 7.1 ± 0.7 HRL Wetland Commission error (%) 71.4 ± 2.7 65.2 ± 3.4 64.0 ± 3.9 Omiossion error (%) 31.2 ± 2.9 68.0 ± 3.6 25.5 ± 5.1 HRL Water Commission error (%) 7.9 ± 1.6 8.1 ± 1.5 5.1 ± 1.2 Omiossion error (%) 0.9 ± 0.2 1.3 ± 0.6 1.8 ± 0.7 ○ 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
Conclusions & future directions ● So far, LUCAS2012 has been used for accuracy assessment of 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
Thank You for Your Attention!!! 12
Recommend
More recommend
