User Manual Copernicus Land Monitoring Service High Resolution land cover characteristics Tree-cover/forest and change 2015-2018
Consortium Partners:
Consortium Composition of the Copernicus HRL 2018 Lot 2 No. Organisation name Organisation short name Country Consortium 1 GAF AG (Lead) GAF Germany 2 SIRS (Consortium Partner) SIRS France 3 Telespazio Ibérica (Sub-contractor) TPZ-I Spain
Contact: Copernicus Land Monitoring Service (CLMS) European Environment Agency (EEA) Kongens Nytorv 6 – 1050 Copenhagen K. – Denmark https://land.copernicus.eu/
Lead service provider for production: GAF AG Arnulfstr. 199 – 80634 Munich - Germany E-mail: [email protected] – Internet: www.gaf.de
Document version 1.2
Disclaimer: © European Union, Copernicus Land Monitoring Service 2021, European Environment Agency (EEA)
All Rights Reserved. No parts of this document may be photocopied, reproduced, stored in retrieval system, or transmitted, in any form or by any means whether electronic, mechanical, or otherwise without the prior written permission of the European Environment Agency.
HRL Forest 2018 Product User Manual Page 2
Content
I. Executive Summary ...... 7 II. Background of the Document ...... 8 Scope of the Document ...... 8 Content and Structure ...... 8 Applicable Documents ...... 8 III. Review of User Requirements ...... 9 IV. Product Application Areas and Examples of Use Cases...... 10 Use Case: Feasibility study for tree species mapping ...... 10 Use Case: Exploration of forest access areas for forest management activities ...... 11 V. Product Description ...... 12 Overview of the Product Portfolio ...... 12 Thematic Characteristics of the HRL Forest 2018 ...... 13 Technical Details of the HRL Forest 2018 ...... 14 Status Layers and Change Products ...... 14 Aggregated Products ...... 24 Additional Expert Products ...... 31 Reference Products ...... 39 VI. Production Methodology and Workflow ...... 41 Workflow Description ...... 41 Strengths and Limitations ...... 45 VII. Quality Assessment ...... 47 Internal Validation ...... 47 Thematic Accuracy...... 47 QA/QC Procedures ...... 54 VIII. Terms of Use and Product Technical Support ...... 55 Terms of Use ...... 55 Citation ...... 55 Product Technical Support ...... 55 IX. References ...... 56 X. Annexes ...... 57
HRL Forest 2018 Product User Manual Page 3
Annex 1: Naming Convention of HRL 2018 Products ...... 57 Annex 2: File Naming Nomenclature of HRL Forest 2018 Products ...... 59 Annex 3: Download Content ...... 61 Annex 4: Coordinate Reference System Sheet ...... 62 Annex 5: Colour Palettes and Attribute Fields ...... 63 Annex 6: Tree Cover Change Mask 1518 - Initial Aggregation Rules...... 68
LIST OF FIGURES
Figure 1: HRL Forest 2018 product portfolio...... 12 Figure 2: Pan-European status layer Dominant Leaf Type 2018 (Administrative boundaries © EuroGeographics) ...... 15 Figure 3: Pan-European status layer Tree Cover Density 2018 (Administrative boundaries © EuroGeographics) ...... 18 Figure 4: Workflow for the HRL Forest 2018 product portfolio ...... 41 Figure 5: Scatter plot of TCD 2018 map values and VHR reference data ...... 50
LIST OF TABLES
Table 1: Elements to be included/excluded in the tree cover mask 2018 ...... 13 Table 2: Product sheet for the DLT 2018 10m ...... 16 Table 3: Product sheet for the TCD 2018 10m ...... 19 Table 4: Product sheet for the FTY 2018 10m ...... 21 Table 5: Product sheet for the TCCM 1518 20m ...... 22 Table 6: Product sheet for the DLTC 1518 20m ...... 23 Table 7: Product sheet for the TCD 2018 100m ...... 25 Table 8: Product sheet for the FTY 2018 100m ...... 26 Table 9: Product sheet for the BCD 2018 100m ...... 28 Table 10: Product sheet for the CCD 2018 100m ...... 30 Table 11: FADSL: Hierarchical intersection of input layers and resulting thematic classes ...... 32 Table 12: Product sheet for the FADSL 2018 ...... 33 Table 13: Product sheet for the Dominant Leaf Type Data Score Layer 2018 ...... 34 Table 14: Product sheet for the Dominant Leaf Type Confidence Layer ...... 35 Table 15: Product sheet for the Tree Cover Density Confidence Layer ...... 36 Table 16: Product sheet for the Derived Correction Layers ...... 37 Table 17: Area-weighted confusion matrix of the pan-European DLT 2018 status layer ...... 48 Table 18: Area-weighted confusion matrix of the pan-European TCD 2018 status layer with a 30% density threshold ...... 48 Table 19: Area-weighted confusion matrix of the pan-European TCD 2018 status layer with no density threshold applied...... 49
HRL Forest 2018 Product User Manual Page 4
Table 20: Area-weighted confusion matrix of the TCCM 1518 change layer; Class names: 0=unchanged areas with no tree cover, 1=new tree cover, 2=loss of tree cover, 10=unchanged areas with tree cover ...... 52 Table 21: Area-weighted confusion matrix of the DLTC 1518 change layer; Class names: 0=unchanged areas with no tree cover, 1=new broadleaved cover, 2=new coniferous cover, 3=loss of broadleaved cover, 4=loss of coniferous cover, 10=unchanged areas with tree cover ...... 53 Table 22: Example of a Coordinate Reference System Sheet for Hungary ...... 62 Table 23: Colour palette and attributes of new TCD layer ...... 63 Table 24: Colour palette and attributes of DLT layer ...... 63 Table 25: Colour palette and attributes of FTY layer ...... 64 Table 26: Colour palette and attributes of FADSL ...... 64 Table 27: Colour palette and attributes of BCD layer ...... 64 Table 28: Colour palette and attributes of CCD layer...... 65 Table 29: Colour palette and attributes of DLTC layer ...... 65 Table 30: Colour palette and attributes of TCCM layer ...... 66 Table 31: Colour palette and attributes of DLTCL ...... 66 Table 32: Colour palette and attributes of TCDCL ...... 67
HRL Forest 2018 Product User Manual Page 5
ACRONYMS AND ABBREVIATIONS AD Applicable Document BCD Broadleaved Cover Density BfN Bundesamt für Naturschutz CCD Coniferous Cover Density CL Confidence Layer CLC CORINE Land Cover CLMS Copernicus Land Monitoring Service CRS Coordinate Reference Sheet DCL Derived Correction Layer DLT Dominant Leaf Type DLTC Dominant Leaf Type Change DLTCL Dominant Leaf Type Confidence Layer DLTDSL Dominant Leaf Type Data Score Layer EEA European Environment Agency EO Earth Observation ESA European Space Agency ETRS89 European Terrestrial Reference System 1989 FADSL Forest Additional Support Layer FAO Food and Agriculture Organization FTY Forest Type GRI Global Reference Image H2020 Horizon 2020 (EU Research and Innovation programme) HR High resolution JRC Joint Research Centre LAEA Lambert Azimuthal Equal Area LPIS Land Parcel Identification System LULUCF Land Use, Land Use Change and Forestry MMU Minimum Mapping Unit MMW Minimum Mapping Width OSM Open Street Map OTC Open Telecom Cloud QA Quality Assurance QC Quality Control RF Random Forest SOER State of Environment Report TCCM Tree Cover Change Mask TCD Tree Cover Density TCDCL Tree Cover Density Confidence Layer TCM Tree Cover Mask UBA Umweltbundesamt UN-ECE United Nations Economic Commission for Europe VHR Very High Resolution
HRL Forest 2018 Product User Manual Page 6
I. Executive Summary
Copernicus is the European Union's Earth Observation Programme. It offers information services based on satellite Earth observation and in situ (non-space) data. These information services are freely and openly accessible to users through six thematic Copernicus services (atmosphere monitoring, marine environment monitoring, land monitoring, climate change, emergency management and security).
The Copernicus Land Monitoring Service (CLMS) provides geographical information on land cover and its changes, land use, vegetation state, water cycle and earth surface energy variables to a broad range of users in Europe and across the world in the field of environmental terrestrial applications.
CLMS is jointly implemented by the European Environment Agency and the European Commission DG Joint Research Centre (JRC).
This document captures detailed definitions and product specifications for the High Resolution Layer (HRL) Forest for the 2018 reference year. Two previous Forest products exist for the 2012 and 2015 reference years, following a 3-years update cycle. The HRL Forest with reference year 2018 has been fully produced in the European Terrestrial Reference System 1989 (ETRS89) and in Lambert Azimuthal Equal Area (LAEA) projection by a consortium of European service providers.
The two primary status layers Dominant Leaf Type (DLT) and Tree Cover Density (TCD) at 10m spatial resolution were derived from multi-temporal imagery of the Sentinel-2 satellite, operated by the European Space Agency (ESA), providing information about leaf type (broadleaved/coniferous) and the proportional tree cover density at pixel level (TCD in %). This allows users to apply a (national) forest definition, taking any canopy crown cover threshold into account, which best fits their specific needs. The HRL Forest 2018 product portfolio also comprises a Forest Type product, which is largely following the forest definition of the Food and Agriculture Organization (FAO).
Both primary status layers are sharing the same spatial extent and are the basis for all other products of the HRL Forest product portfolio, which includes a series of aggregated 100m products as well as a set of 20m change products 2012-2015-2018 at pan-European level. Additionally, the products of the 2018 reference year comprise a set of additional quality layers (e.g. Confidence Layers) and Derived Correction Layers (DCL) for the reference years 2015 and 2012. These additional layers enable a re-processing of previous HRL Forest status layers and are based on commission and omission errors identified during change analysis 2015-2018.
The HRLs are designed for use by a broad user community as basis for environmental analyses and for supporting political decision-making. Specifically, they can be used for supporting (amongst others) the reporting on Land Use, Land Use Change and Forestry (LULUCF).
HRL Forest 2018 Product User Manual Page 7
II. Background of the Document
Scope of the Document
The Product User Manual is the primary document that users are recommended to consult before using the product. It provides a technical overview of the product characteristics, production methodology and workflows, quality assessment checks and their results, as well as information on underlying user requirements, potential application areas and examples of use cases, together with information on the terms of use and product technical support.
Content and Structure
The document is structured as follows:
▪ Chapter III recalls the user requirements ▪ Chapter IV presents potential application areas and/or example use cases ▪ Chapter V presents a product description (product file naming convention and format(s), product content and characteristics) ▪ Chapter VI provides a description of the production methodology and workflows ▪ Chapter VII summarizes the quality assessment and validation procedure and the results ▪ Chapter VIII provides information about product access and use conditions as well as the technical product support ▪ Chapter IX lists references to the cited literature ▪ Chapter X provides annexes
Applicable Documents
The following applicable documents (AD) serve as further background information to the users. All documents are freely accessible and available for download.
Ref. Document Name AD01 Tender specifications: EEA/IDM/R0/18/009 AD02 H2020 ECoLaSS User Requirement Analysis: Deliverable D3.2 - Service Evolution Requirements Report Vol. 2 AD03 Nextspace User Study: Nextspace database for user requirements
HRL Forest 2018 Product User Manual Page 8
III. Review of User Requirements
In the frame of the Horizon 2020 (H2020) project ECoLaSS a survey (AD02) of key stakeholders has been performed in order to evaluate the user requirements towards the evolution of existing and future Copernicus products. This survey made also use of the results from the Nextspace User Study (AD03) and revealed that HRL users like European institutions, service industry, research and academia, national agencies, regional administrations or private users would in general appreciate:
▪ High accuracy of the products ▪ No data gaps - due to enhanced cloud gap mitigation ▪ Extending the coverage of the products beyond the EEA-39 area ▪ Sufficient spatial and timely resolution concerning both, status layer and change layer ▪ Short update cycles ▪ Change monitoring ▪ Free and open access ▪ High technical quality ▪ High thematic quality/meaningful and application oriented product definitions ▪ Standardized and comparable nomenclature ▪ Transparent and scientific workflows and state-of-the-art methodology ▪ Detailed documentation of the workflow and the respective methodology ▪ Consistency of the Pan-European products enabling synergistic use of all products ▪ Streamlining the pan-European product with global ones ▪ Availability of historic data and compatibility of time series ▪ Open access to the original Copernicus Sentinel data ▪ Sophisticated product presentation and visualisation possibilities in an online viewer on the Copernicus platform ▪ IPCC conformity
It is the strength of the HRL products that many of the mentioned requirements are already satisfied or at least taken into account in current or upcoming implementations. However, key Copernicus users and stakeholders would appreciate even more thematic information within the HRL Forest such as the distinction between tree species. Due to a lack of essential reference and input data, this can’t be fully addressed at present. However, the DLT layer already shows a distinction between broadleaved and coniferous trees.
HRL Forest 2018 Product User Manual Page 9
IV. Product Application Areas and Examples of Use Cases
The HRL Forest is designed for use by a broad user community as basis for environmental and regional analysis and for supporting political decision-making. They do not only show the area of tree covered surface in the European Environment Agency (EEA)-391 countries but give also details on the TCD of those areas and the distribution of coniferous and broadleaved trees. Several additional products provide background information and further details about the mapped area. In addition, the Forest Type (FTY) products use an approximation of the Food and Agriculture Organization (FAO) forest definition on the tree cover data. Apart from mapping the current status, it is possible to monitor change, which can be a result of manifold influences, e.g. related to climate or forestry. This is enabled by the 3 year update cycle of the HRLs which allows to not only map the loss of tree cover but also the gains. All the provided forest products serve as a supporting source of information for the member states in their reporting obligations to the United Nations Economic Commission for Europe (UN-ECE) and FAO. At European level, contributions to the Land Use, Land Use Change and Forestry (LULUCF) reporting and EEA’s State of Environment Report (SOER) are relevant. Apart from that, the freely available thematic layers further support scientific research applications, e.g. on the causes of the identified changes, and the impacts on the related ecosystems. This is of great use for better understanding the connections between the different components of the respective ecosystems and supports the sustainable use of forests.
As an example, please find below some short information on two use cases at national level, for which the Copernicus HRL Forest products represent a fundamental input:
Use Case: Feasibility study for tree species mapping
Umweltbundesamt (UBA), Germany: Explorative use of the HRL Forest as base information and further analysis towards derivation of tree species from Sentinel-2 data in the frame of a feasibility study with German and Austrian partners. Results from 5 case study sites show that a number of 8-16 tree species could be detected using multi-temporal Sentinel-2 data. The study showed that a high number of available cloud- free satellite scenes and the availability of additional adequate local reference data for algorithm training are required for retrieval of the results2.
There is such demand for a tree species map in Germany at:
▪ Umweltbundesamt (UBA -Environmental Protection Agency): Assessment and risk analysis of forest ecosystems, assessment of material discharges (critical loads), monitoring of indicators in the framework of the German Strategy for Adaptation to Climate Change. ▪ Bundesamt für Naturschutz (BfN – Federal Agency for Nature Conservation): renewable energy planning, requiring tree species composition to identify valuable habitats; considering adaptation to climate change.
1 For further details on EEA member countries and cooperating countries (EEA39), please see https://www.eea.europa.eu/countries-and-regions 2 see https://www.umweltbundesamt.de/sites/default/files/medien/1410/publikationen/2019-02-27_texte_16- 2019_copernicus.pdf
HRL Forest 2018 Product User Manual Page 10
▪ Thünen-Institut (TI – Federal Research Institute for Rural Areas, Forestry and Fisheries): Tree species accounting for the German State-of-Forest report („Wald-Zustandsbericht“).
Use Case: Exploration of forest access areas for forest management activities
ThüringenForst, Germany: Utilization of the Copernicus HRL Forest Tree Cover Mask for detection of forest access areas in order to support a series of forest management activities:
▪ Timber harvest ▪ Forest preservation ▪ Forest plantations ▪ Forest development ▪ Hunting
By dedicated GIS operations using the Tree Cover Mask, elevation data, survey data and other sources, an indication layer is being generated, which enables ThüringenForst to fulfil its operational obligations. This cost-efficient method is operational since the very first HRL Forest implementation for the reference year 2012.
HRL Forest 2018 Product User Manual Page 11
V. Product Description
Overview of the Product Portfolio The HRL Forest3 comprises two primary status layers for the reference year 2018: Dominant Leaf Type (DLT) and Tree Cover Density (TCD). These status layers represent the input for all other products of the HRL Forest product portfolio, including a Forest Type status layer following a forest definition, and change products. The latter one are represented by a Tree Cover Change Mask (TCCM) 2015-2018 and a Dominant Leaf Type Change (DLTC) product 2015-2018 in 20m spatial resolution. Further, aggregated products of the status layers at 100m resolution are provided, as well as additional expert products and some reference products. More detailed product descriptions can be found in the following sub-chapters.
Figure 1: HRL Forest 2018 product portfolio.
3 Please be aware that while the overall term for this group of products is “HRL Forest”, there are multiple different definitions of what a “forest” is in technical terms. Strictly speaking: the primary products of this HRL are tree cover products (TCD and DLT), that map trees wherever they occur, also outside of what is (technically) a forest. As part of the HRL “Forest” group of products there is only one product that applies a FAO forest definition, and can therefore be called a (sensu stricto) forest product: the Forest Type (FTY) product (in 10m and 100m resolution). The fact that TCD and DLT do not have a forest definition and filtering applied makes it possible for users to adapt the existing tree cover density / dominant leaf type products to their forest definition (if different from the FAO definition).
HRL Forest 2018 Product User Manual Page 12
Thematic Characteristics of the HRL Forest 2018 The Copernicus HRL Forest defines TCD as the „vertical projection of tree crowns to a horizontal earth’s surface“ and provides information on the proportional crown coverage per pixel. This information is derived from multispectral High Resolution (HR) satellite data using Very High Resolution (VHR) satellite data and/or aerial ortho-imagery as reference data. TCD is assessed on different VHR sources by visual interpretation following a 10x10 point grid approach, resulting in proportional density information on a 100m by 100m grid level. This density information is linked with the mean spectral values from the input satellite data to generate a specific regression model per production unit. Subsequently, the model is transferred to the satellite data assigned to a production unit, and density values are calculated by applying a multiple linear regression.
TCD shows a natural sensitivity towards phenology and radiometric influences (e.g. haze). Consequently, the magnitude of TCD values strongly relies on the availability and quality of adequate satellite input data and reference data and may vary regionally. Furthermore, extreme weather events and climate conditions (e.g. European drought 2018) show a negative influence on the magnitude of density values due to leaf colouring and leaf shedding.
Table 1 provides an overview of the Land Cover (LC) and Land Use (LU) features that shall be included or excluded in the “tree cover” mapping, if detectable from the satellite imagery. The resulting binary Tree Cover Mask (TCM) is being enhanced by means of various processing steps (automatically and semi- automatically) and subsequently filled with the estimated TCD values.
Table 1: Elements to be included/excluded in the tree cover mask 2018
Elements included in the tree covered area Elements excluded from tree covered area
(if tree cover can be detected from the 10m imagery) (if no tree cover can be detected from the 10m imagery)
• Evergreen/deciduous broadleaved, sclerophyllous and • Open areas within forests (roads, permanently open coniferous trees of any use vegetated areas, clear cuts, fully burnt areas, other severe • Forests (grown-up and under development) forest damage areas, etc.) • Orchards, olive groves, fruit and other tree plantations, • Dwarf shrub-covered areas, such as moors and heathland agro-forestry areas • Vineyards • Transitional woodland, forests in regeneration • Dwarf pine / green alder in alpine areas • Groups of trees within urban areas (alleys, wooded • Mediterranean shrublands (macchia, garrigue etc.) parks and gardens) • Shrubland • Forest management/use features inside forests (forest roads, firebreaks, thinnings, forest nurseries, etc.) • Forest damage features inside forests (partially burnt areas, storm damages, insect-infested damages, etc.)
HRL Forest 2018 Product User Manual Page 13
Technical Details of the HRL Forest 2018
Status Layers and Change Products
Status layers provide information on the Dominant Leaf Type and the Tree Cover Density at pixel level for the reference year 2018 in 10m resolution. With the 10m Forest Type status layer a forest product largely following the FAO forest definition is provided to the users. Furthermore, two change products exist, providing information on gain and loss of tree cover and its associated leaf type between the reference years of HRL 2015 (2015/2016) and HRL 2018 (2018) in 20m resolution. The portfolio of status layers and change products consists of the following products:
▪ Dominant Leaf Type 10m ▪ Tree Cover Density 10m ▪ Forest Type 10m ▪ Tree Cover Change Mask 20m ▪ Dominant Leaf Type Change 20m
Dominant Leaf Type 10m The Dominant Leaf Type (DLT) at 10m spatial resolution is one of the primary status layers, derived from Sentinel-2 time series data using a Random Forest (RF) Classifier, and has the following main specifications:
▪ 10m spatial resolution ▪ Providing information on the dominant leaf type: broadleaved or coniferous ▪ No Minimum Mapping Unit (MMU); pixel-based ▪ Minimum Mapping Width (MMW) of 10m ▪ Consistent multi-temporal coverage: 01.03.2018 – 31.10.2018 ▪ Fully identical in its outline extent with the TCD product
HRL Forest 2018 Product User Manual Page 14
Figure 2: Pan-European status layer Dominant Leaf Type 2018 (Administrative boundaries © EuroGeographics)
HRL Forest 2018 Product User Manual Page 15
Table 2: Product sheet for the DLT 2018 10m
Acronym Product category Dominant Leaf Type 10m DLT Status Layer
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Minimum 90% user's / producer's accuracy for both, broadleaved and coniferous class
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-tree covered areas 1: broadleaved trees 2: coniferous trees 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
HRL Forest 2018 Product User Manual Page 16
Tree Cover Density 10m The Tree Cover Density (TCD) at 10m spatial resolution represents one of the primary status layers of the HRL Forest product portfolio and has the following main specifications:
▪ 10m spatial resolution ▪ TCD range of 0-100% ▪ No Minimum Mapping Unit (MMU); pixel-based ▪ Minimum Mapping Width (MMW) of 10m ▪ Consistent multi-temporal coverage: 01.03.2018 – 31.10.2018 ▪ Fully identical in its outline extent with the DLT product In particular, information on crown cover, which is provided with the continuous-scale (0-100%) TCD product for the whole of Europe, can be generally used by different countries, even if different national forest definition regarding the crown/canopy cover exist (e.g. Spain with 5%, Austria with 30%). In this sense, the TCD product largely supports the user-specific derivation of forest-related products according to the European-wide different understandings of forest.
The Tree Cover Density product is also available as aggregated version in 100m spatial resolution, fully aligned to the EEA 100m reference grid.
HRL Forest 2018 Product User Manual Page 17
Figure 3: Pan-European status layer Tree Cover Density 2018 (Administrative boundaries © EuroGeographics)
HRL Forest 2018 Product User Manual Page 18
Table 3: Product sheet for the TCD 2018 10m
Acronym Product category Tree Cover Density 10m TCD Status Layer
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Minimum 90% user's / producer's accuracy for both, broadleaved and coniferous class
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-tree covered areas 1-100: tree cover density in % 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
HRL Forest 2018 Product User Manual Page 19
Forest Type 10m With the Forest Type product the HRL Forest already provides one type of forest products following a forest definition. Contrary to the TCD product non-forest trees are excluded following the forest definition of FAO. This is e.g. specified in the terms and definitions of the Global Forest Resources Assessment (FRA) Working Paper 1, which is accessible online via www.fao.org/docrep/006/ad665e/ad665e06.htm. The forest definition of the FAO includes and excludes the following features/elements:
Includes (FAO): forest nurseries and seed orchards that constitute an integral part of the forest; as well as forest roads, cleared tracts, firebreaks and other small open areas < 0.5 ha and/or < 20m width. Forest in national parks, nature reserves and other protected areas such as those of specific scientific, historical, cultural or spiritual interest; windbreaks and shelterbelts of trees with an area of more than 0.5 ha and width of more or equal than 20m; plantations primarily used for forestry purposes, including cork oak stands.
Excludes (FAO): land predominantly used for agricultural practices. In this sense fruit trees and olive groves are also excluded. Gardens and urban parks are also not considered as forest.
The 10m Forest Type product is produced by applying a „Forest“ definition, largely following the FAO definition, whereas tree cover in traditional agroforestry systems such as Dehesa/Montado is explicitly included for EEA purposes. The product is derived through a spatial intersection of the two primary status layers TCD and DLT and excludes areas under agricultural use and in urban context as provided by the FADSL. It has the following main specifications:
▪ 10m spatial resolution ▪ TCD range of ≥10-100% ▪ Minimum Mapping Unit (MMU; minimum number of pixels to form a patch) of 0.5 ha (50 pixels); applicable both for tree-covered areas and for non-tree-covered areas in a 4-pixel connectivity mode, but not for the distinction of DLT within the tree-covered area for which no such minimum is set. The potentially available leaf type information for areas below 10% density within non-forest patches smaller the MMU is explicitly kept from the pixel-based DLT product to ensure consistency. ▪ MMW of 10m The Forest Type product is also available as aggregated version in 100m spatial resolution, fully aligned to the EEA 100m reference grid.
HRL Forest 2018 Product User Manual Page 20
Table 4: Product sheet for the FTY 2018 10m
Acronym Product category Forest Type 10m FTY Status Layer
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source layers Tree Cover Density, Dominant Leaf Type and FADSL in 10m spatial resolution.
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) 0.5 ha (50 pixels)
Tree cover density threshold 10%
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-forest areas 1: broadleaved forest 2: coniferous forest 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
HRL Forest 2018 Product User Manual Page 21
Tree Cover Change Mask 20m The Tree Cover Change Mask (TCCM) is a pixel-based 20m change layer derived from the binary masks of the DLT status layers 2015 and 2018. It comprises of four thematic classes out of which 2 indicate changes (new tree cover/loss of tree cover) between the two time steps.
The Tree Cover Change Mask product has the following main specifications:
▪ 20m spatial resolution ▪ TCD range of 0-100% ▪ Boundary filter of 1 pixel to address geometric differences ▪ MMU of 1 ha (100 pixels) for change classes (holes within change areas are filtered up to <1 ha MMU)
Table 5: Product sheet for the TCCM 1518 20m
Acronym Product category Tree Cover Change Mask 20m TCCM Change Product
Reference year 2015 (+/- 1 year) to 2018 (March to October)
Geometric resolution Pixel resolution 20m x 20m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source Tree Cover Density 2015/2018 in 20m/10m spatial resolution.
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) 1 ha (100 pixels) for change classes
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: unchanged areas with no tree cover 1: new tree cover 2: loss of tree cover 10: unchanged areas with tree cover
HRL Forest 2018 Product User Manual Page 22
254: unclassifiable in any of parent status layers 255: outside area
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Dominant Leaf Type Change 20m The Dominant Leaf Type Change (DLTC) product is a pixel-based 20m layer covering the time period of 2015 (+/- 1 year) to 2018. It is derived by dedicated GIS operations of the primary status layer, DLT, of both time steps. It includes 10 thematic classes, thereof 6 change classes.
The DLTC product has the following main specifications:
▪ 20m spatial resolution ▪ TCD range of 0-100% ▪ Boundary filter of 1 pixel to address geometric differences ▪ MMU of 1 ha (100 pixels) for change classes Besides the DLTC 1518, a Dominant Leaf Type Change product for the period 2012-2015 is provided by a dedicated recoding of the experimental (unpublished) DLTC 1215 from the HRL 2015 production. However, this product is not fully comparable with the DLTC 1518 as the initial specifications are different.
Table 6: Product sheet for the DLTC 1518 20m
Acronym Product category Dominant Leaf Type Change 20m DLTC Change Product
Reference year 2015 (+/- 1 year) to 2018 (March to October)
Geometric resolution Pixel resolution 20m x 20m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source layers Dominant Leaf Type 2015/2018 at 20m/10m spatial resolution.
Data type 8bit unsigned raster with LZW compression
HRL Forest 2018 Product User Manual Page 23
Minimum Mapping Unit (MMU) 1 ha (100 pixels) for change classes
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: unchanged areas with no tree cover 1: new broadleaved cover 2: new coniferous cover 3: loss of broadleaved cover 4: loss of coniferous cover 10: unchanged areas with tree cover 12: potential change among dominant leaf types 254: unclassifiable in any of parent status layers 255: outside area
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Aggregated Products
The HRL Forest 2018 provides a series of aggregated products, which are primarily derived from the 10m status layers Dominant Leaf Type, Tree Cover Density and Forest Type:
▪ Tree Cover Density 100m ▪ Forest Type 100m ▪ Broadleaved Cover Density 100m ▪ Coniferous Cover Density 100m
Tree Cover Density 100m The Tree Cover Density 2018 at 100m spatial resolution is an aggregated version of the primary status layer, fully aligned to the EEA reference grid. It has the following main specifications:
▪ 100m spatial resolution ▪ TCD range of 0-100% ▪ No Minimum Mapping Unit (MMU); pixel-based ▪ Consistent multi-temporal coverage: 01.03.2018 – 31.10.2018
HRL Forest 2018 Product User Manual Page 24
Table 7: Product sheet for the TCD 2018 100m
Acronym Product category Tree Cover Density 100m TCD Aggregated product
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 100m x 100m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the input source layer Tree Cover Density 2018 at 10m spatial resolution
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-tree covered areas 1-100: tree cover density in % 255: outside area
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Processing steps: The TCD at 100m spatial resolution is derived through spatial aggregation from the 10m status layer according to the following rules: 1) The spatially consistent EEA 100m grid is overlaid to the 10m Tree Cover Density product. 2) For each 100m grid cell the arithmetic mean density of all underlying 10m pixels (with density values from 0-100) is calculated. The thereof resulting mean values from the aggregation (floating
HRL Forest 2018 Product User Manual Page 25
point data type) are rounded and finally converted to integer values (e.g. raw values in the range from 33.5 to 34.4 are converted to a density value of 34).
3) The class 255 = outside area is predefined by the 100m boundary layer in European and national projection and remains unchanged. As a consequence of this aggregation, small patches in the 100m x 100m grid will be generalised (i.e. exaggerated) according to the above aggregation rules.
Forest Type 100m The Forest Type 2018 at 100m spatial resolution is an aggregated version of the secondary status layer, fully aligned to the EEA reference grid. It has the following main specifications: ▪ 100m spatial resolution ▪ TCD range of ≥10-100% ▪ No Minimum Mapping Unit (MMU); pixel-based ▪ Providing information on the leaf type following the CLC forest definition with 3 thematic classes
Table 8: Product sheet for the FTY 2018 100m
Acronym Product category Forest Type 100m FTY Aggregated product
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 100m x 100m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source layer Forest Type 2018 in 10m spatial resolution.
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold 10%
HRL Forest 2018 Product User Manual Page 26
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-forest areas 1: broadleaved forest 2: coniferous forest 3: mixed zones 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Processing steps: The FTY at 100m spatial resolution is derived through spatial aggregation from the 10m status layer according to the following rules: 1) The spatially consistent EEA 100m grid is overlaid to the 10m Forest Type product.
2) Identification of forest covered 100m grid cells by counting the forest covered pixels (broadleaved and coniferous) of the underlying 10m FTY product. If the number of forest covered pixels is ≥ 50, the 100m grid cell is assigned to “forest cover”. Consequently, all grid cells showing less than 50 forest covered pixels will be labelled as 0 = all non-forest areas in the aggregated product. 3) Application of the CORINE Land Cover definition for broadleaved, coniferous and mixed forest for all grid cells being “forest cover”. 100m grid cells with “broadleaved” constituents with ≥75% of the 10m forest pixels are assigned to the thematic class 1 = broadleaved forest. 100m grid cells with “coniferous” constituents with ≥75% of the 10m forest pixels are assigned to the thematic class 2 = coniferous forest. Grid cells in which neither “broadleaved” nor “coniferous” constituents account for ≥75% of the 10m forest pixels are labelled as class 3 = mixed zones. 4) The class 255 = outside area is predefined by the 100m boundary layer in European and national projection and remains unchanged. As a consequence of this aggregation, small patches in the 100m x 100m grid will be generalised (i.e. exaggerated or deleted) according to the above aggregation rules.
Broadleaved Cover Density 100m The Broadleaved Cover Density (BCD) layer is a newly introduced leaf type product at 100m spatial resolution. It is derived through aggregation of the DLT 2015 layer at 20m resolution and the DLT 2018 layer at 10m resolution for the reference years 2015 and 2018, respectively. More specifically the BCD is defined as the
HRL Forest 2018 Product User Manual Page 27 fraction of pixels in the DLT layer being broadleaved, falling in the corresponding 100m grid cell of the EEA reference grid. While the value range for both 2015 and 2018 BCD products is 0-100%, it is worthwhile noting that only the 2018 product will achieve a continuous integer value range from 0-100. This is due to the fact that the aggregation of the BCD from 20m DLT 2015 product typically allows to compute the density only with a step size of 4% (i.e. 1/25 20m pixels).
The Broadleaved Cover Density layer has the following main specifications:
▪ 100m spatial resolution ▪ Providing information on the leaf type fraction (broadleaved) per 100m grid cell in % ▪ No MMU; pixel-based
Table 9: Product sheet for the BCD 2018 100m
Acronym Product category Broadleaved Cover Density 100m BCD Aggregated product
Reference year 2018 (March to October) / 2015 (+/- 1 year)
Geometric resolution Pixel resolution 100m x 100m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source layers Dominant Leaf Type 2018 / 2015.
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-broadleaved covered areas 1-100: broadleaved cover density in % 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area
HRL Forest 2018 Product User Manual Page 28
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Processing steps: The BCD at 100m spatial resolution is derived through spatial aggregation from the 10m Dominant Leaf Type product according to the following rules: 1) The spatially consistent EEA 100m grid is overlaid to the 10m Dominant Leaf Type product. 2) For each 100m grid cell the number of broadleaved covered 10m pixels is counted. The sum of the broadleaved pixels is written in the 100m grid cell as broadleaved cover density. Grid cells showing no tree cover and/or coniferous cover only are labelled as class 0 = all non-broadleaved covered areas. 3) The class 255 = outside area is predefined by the 100m boundary layer in European and national projection and remains unchanged. As a consequence of this aggregation, small patches of broadleaved trees in the 100m x 100m grid will be generalised (i.e. exaggerated) according to the above aggregation rules.
Coniferous Cover Density 100m The Coniferous Cover Density (CCD) layer is a newly introduced leaf type product at 100m spatial resolution. It is derived through aggregation of the DLT 2015 layer at 20m resolution and the DLT 2018 layer at 10m resolution for the reference years 2015 and 2018, respectively. More specifically the CCD is defined as the fraction of pixels in the DLT layer being coniferous, falling in the corresponding 100m grid cell of the EEA reference grid. While the value range for both 2015 and 2018 CCD products is 0-100%, it is worthwhile noting that only the 2018 product will achieve a continuous integer value range from 0-100. This is due to the fact that the aggregation of the CCD from 20m DLT 2015 product typically allows to compute the density only with a step size of 4% (i.e. 1/25 20m pixels).
The Coniferous Cover Density layer has the following main specifications:
▪ 100m spatial resolution ▪ Providing information on the leaf type fraction (coniferous) per 100m grid cell in % ▪ No MMU; pixel-based
HRL Forest 2018 Product User Manual Page 29
Table 10: Product sheet for the CCD 2018 100m
Acronym Product category Coniferous Cover Density 100m CCD Aggregated product
Reference year 2018 (March to October) / 2015 (+/- 1 year)
Geometric resolution Pixel resolution 100m x 100m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source layers Dominant Leaf Type 2018 / 2015.
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-coniferous covered areas 1-100: coniferous cover density in % 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
HRL Forest 2018 Product User Manual Page 30
Processing steps: The CCD at 100m spatial resolution is derived through spatial aggregation from the 10m Dominant Leaf Type product according to the following rules: 1) The spatially consistent EEA 100m grid is overlaid to the 10m Dominant Leaf Type product. 2) For each 100m grid cell the number of coniferous covered 10m pixels is counted. The sum of the coniferous pixels is written in the 100m grid cell as coniferous cover density. Grid cells showing no tree cover and/or broadleaved cover only are labelled as class 0 = all non-coniferous covered areas. 3) The class 255 = outside area is predefined by the 100m boundary layer in European and national projection and remains unchanged. As a consequence of this aggregation, small patches of coniferous trees in the 100m x 100m grid will be generalised (i.e. exaggerated) according to the above aggregation rules.
Additional Expert Products
Besides the status layers, change products and aggregated products, the HRL Forest 2018 portfolio provides a series of additional expert products:
▪ Forest Additional Support Layer 10m ▪ Dominant Leaf Type Data Score Layer 10m ▪ Confidence Layer (CL) for DLT and TCD products ▪ Derived Correction Layer for 2012 and 2015 reference year products ▪ Biophysical Variables
Forest Additional Support Layer 10m The Forest Additional Support Layer (FADSL) provides information on trees under agricultural use or in urban context by utilization of CORINE Land Cover (CLC) 2018 information and HRL Imperviousness 2018. The layer is derived through a spatial intersection of the 10m DLT and TCD layers with CLC 2018 and Imperviousness Degree 2018. It has the following main specifications:
▪ 10m spatial resolution ▪ TCD range of ≥ 10-100% ▪ No MMU; pixel-based ▪ MMW of 10m ▪ Providing information on: o trees predominantly used for agricultural practices – broadleaved (from CLC 2018 class 2.2.2 & 2.2.3) o trees in urban context – broadleaved and coniferous (from CLC 2018 class 1.4.1) o trees in urban context – broadleaved and coniferous (from HRL Imperviousness 2018)
HRL Forest 2018 Product User Manual Page 31
This additional raster product at 10m spatial resolution is used to separate real “forest” areas from non- forest areas (i.e. trees predominantly used for agricultural practices, trees in urban context) in order to derive a Forest Type product which is largely following the FAO definition. The separation between forest and non- forest is being performed by several GIS operations and has certain limitations due to the specifications and thematic accuracies of the different input data (DLT with density values ranging from 10-100%; CLC 2018 with 25 ha MMU; IMD 2018 also representing infrastructure). Processing steps: The FADSL is derived through following processing steps and is provided for consistency reasons and further analysis potential: 1) Reclassification of the Imperviousness Degree layer (pixel range 0-100%) in 10m spatial resolution to a binary mask of impervious with 0 = all non-impervious areas and 1 = all impervious areas. Clouds are explicitly not considered. 2) Filtering of all contiguous all non-impervious patches < 25 ha which are fully surrounded by impervious areas in a 4-pixel connectivity mode and subsequently reclassification to 1 = all impervious areas. 3) Hierarchical intersection of the DLT (with density values ranging from 10-100%), CLC 2018 (full vector resolution) and the gap-filled Imperviousness dataset with subsequent reclassification according to the product specifications, resulting in the thematic classes presented in the table below. 4) The class 255 = outside area is predefined by the 10m boundary layer in European and national projection and remains unchanged.
Table 11: FADSL: Hierarchical intersection of input layers and resulting thematic classes
DLT CLC IMD FADSL Order Class Description Class Class Class Class
trees in urban context – broadleaved and 1.) 1 or 2 any 1 4 coniferous (from HRL Imperviousness context)
trees in urban context – broadleaved and 2.) 1 or 2 141 0 5 coniferous (from CLC class 1.4.1)
222 trees predominantly used for agricultural practices 3.) 1 or any 3 – broadleaved (from CLC classes 2.2.2 and 2.2.3) 223
unclassifiable (no satellite image available, or 254 any any 254 clouds, shadows, or snow)
255 any any 255 outside area
all non-tree areas, and tree cover without urban any other remaining combination 0 context or agricultural use
HRL Forest 2018 Product User Manual Page 32
All pixels within the FADSL showing a class code of 3, 4 and 5 are not considered in the 10m Forest Type product derivation, as they represent "trees predominantly used for agricultural practices - broadleaved" (as derived from CLC classes 2.2.2 Fruit tree and berry plantations and 2.2.3 Olive groves) and "trees in urban context – broadleaved and coniferous" (as derived from the filtered IMD 2018 layer and from CLC class 1.4.1 Green urban areas).
Table 12: Product sheet for the FADSL 2018
Acronym Product category Forest Type Additional Support Layer 10m FADSL Expert Product
Reference year 2018
Geometric resolution Pixel resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Determined by the accuracy of the source layers Tree Cover Density 2018, Imperviousness Degree 2018 and CLC 2018 at 10m spatial resolution.
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold 10%
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: all non-tree areas, and tree cover without urban context or agricultural use 3: trees predominantly used for agricultural practices – broadleaved (from CLC 2018) 4: trees in urban context – broadleaved and coniferous (from IMD 2018) 5: trees in urban context – broadleaved and coniferous (from CLC 2018) 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area
Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
HRL Forest 2018 Product User Manual Page 33
Dominant Leaf Type Data Score Layer 10m The Dominant Leaf Type Data Score Layer (DLTDSL) is a raster dataset produced at 10m spatial resolution and depicts the number of valid (cloud-free) Sentinel-2 observations per pixel, that are effectively used for calculation of the time features as input for the thematic classifications of the DLT and TCD primary status layers. The Data Score Layer (DSL) takes into account that some observations are masked due to cloud cover, cloud shadows, haze and snow cover. In case of frequent cloud/haze cover and/or irradiation effects, the number of valid observations significantly decreases (e.g. over urban areas, permanent snow cover, and as result of frequent cloud/haze cover).
Table 13: Product sheet for the Dominant Leaf Type Data Score Layer 2018
Acronym Product category Dominant Leaf Type Data Score Layer 10m DLTDSL Expert Product
Reference year 2018 (March to October)
Geometric resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European Terrestrial Reference System 1989 (ETRS89)
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy N/A
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 1 - 48: valid cloud-free observations 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
HRL Forest 2018 Product User Manual Page 34
Confidence Layers There are two different expert products: one is a Confidence Layer (CL) for the DLT (DLTCL) which depicts, for each tree-covered pixel, the uncertainty of the classification into broadleaved/coniferous classes respectively. More specifically the DLTCL shows the difference of the probabilities of the two classes, respectively. It tends towards 100 where the probability for one class dominates and towards 0 where the both classes reach similar probabilities.
In addition, a second confidence layer named Tree Cover Density (TCDCL) is provided, which represents, for each tree-covered pixel (i.e. within the TCM), the prediction interval of the respective TCD value at 95% confidence. Assuming that the prediction error 휖0 is normally distributed and that 휖0 is independent of 푌 the prediction interval for a given TCD value 푦0 is given by
2 ′ ′ −1 푦0 ± 푡(훼/2,푛−2)√휎 (1 + 푥0(푋 푋) 푥0)
2 Where 푡(훼/2,푛−2) is the t-statistic at a sample size 푛 and at a given confidence interval 훼 (typically 0.95). 휎 is the variance of the residuals of the model, 푋 is the regression matrix of the model and 푥0 is the vector with a specific set observed values (i.e. the spectral features) from which the value of 푦0 is derived (Seber and Lee, 2003).
Both layers are produced at a spatial resolution of 10m for the reference year 2018. The level of confidence is dependent on the quality of the input training samples (number, distribution, timeliness) and the overall data quality within a production unit.
Table 14: Product sheet for the Dominant Leaf Type Confidence Layer
Acronym Product category Dominant Leaf Type Confidence Layer 10m DLTCL Expert Product
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Minimum 90% user's/producer's accuracy for both, broadleaved and coniferous class
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
HRL Forest 2018 Product User Manual Page 35
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: 0% classification confidence 1-100: classification confidence values 101: automatic enhancements 102: manual enhancements 253: all non-tree covered areas 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Table 15: Product sheet for the Tree Cover Density Confidence Layer
Acronym Product category Tree Cover Density Confidence Layer 10m TCDCL Expert Product
Reference year 2018 (March to October)
Geometric resolution Pixel resolution 10m x 10m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection/national projections
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy Minimum 90% user's/producer's accuracy for both, broadleaved and coniferous class
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
HRL Forest 2018 Product User Manual Page 36
Raster coding (thematic pixel values) 0: 0% prediction interval 1-100: prediction values 253: all non-tree covered areas 254: unclassifiable (no satellite image available, or clouds, shadows, or snow) 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Derived Correction Layer The Derived Correction Layer4 (DCL) is an additional expert product, which is provided for the reference years 2012 and 2015 at 20m spatial resolution. The layers are based on the omission and commission errors identified during the bias reduction process for the change analysis and represent ancillary thematic information enabling expert users to improve the historical 2012 and 2015 status layers to a certain degree.
Table 16: Product sheet for the Derived Correction Layers
Acronym Product category Derived Correction Layer 20m DCL Expert Product
Reference year 2015 (+/- 1 year) / 2012 (+/- 1 year)
Geometric resolution Pixel resolution 20m x 20m, fully conform with the EEA reference grid
Coordinate Reference System European ETRS89 LAEA projection
Geometric accuracy (positioning scale) Less than half a pixel. According to ortho-rectified satellite image base delivered by ESA.
Thematic accuracy N/A
Data type 8bit unsigned raster with LZW compression
Minimum Mapping Unit (MMU) Pixel-based (no MMU)
Tree cover density threshold N/A
4 this expert product can be made available to users by EEA on request
HRL Forest 2018 Product User Manual Page 37
Necessary attributes Raster value, count, class name, area (in km2), percentage (taking outside area not into account)
Raster coding (thematic pixel values) 0: no correction 1: omission in 2012 / 2015 2: commission in 2012 / 2015 255: outside area Metadata XML metadata files according to INSPIRE metadata standards
Delivery format GeoTIFF
Biophysical Variables Due to its short repeat pass cycle, Sentinel-2 provides dense image time series, which have a particularly high potential for the derivation of different biophysical variables of photo-synthetically active vegetation. For the derivation of the Dominant Leaf Type layer, several “simple” time features have been computed as input for the classification process. They yield a good contrast between broadleaved and coniferous tree cover as well as between tree-covered areas and vegetated areas under agricultural use. In particular, the following five features summarise the spatial and temporal dynamics of NDVI time-series within the given observation period from 01-03-2018 to 31-10-2018 (adapted to regional conditions where needed) in a compressed format:
1) Minimum seasonal NDVI (ndvimin) 2) Maximum seasonal NDVI (ndvimax) 3) Mean seasonal NDVI (ndvimean) 4) Seasonal 10th percentile NDVI (ndvip10) 5) Seasonal 90th percentile NDVI (ndvip90)
Biophysical variables (Min, Max, Mean, p10, p90 NDVI) are provided as raster datasets at 10m spatial resolution per production unit including start and end dates of the scene selection in the file naming.
In addition to the Biophysical Variables, the top 10 Time Features used in the DLT classification are provided per production unit. The list of relevant features is documented and stored in the Production Unit Layer. File naming is following
HRL Forest 2018 Product User Manual Page 38
Reference Products
In addition to the above-mentioned additional expert products, the HRL Forest 2018 provides a series of reference products for maximum transparency and reproducibility:
▪ TCD 2018 Reference Database (RDB) for absolute calibration and validation ▪ Reference Database (RDB) for change calibration ▪ Production Unit Layer (PUL) Reference Database for Absolute Calibration and Validation The Reference Database (RDB) for absolute calibration and validation of TCD values has been established in the HRL 2015 production for the first time. For the HRL Forest 2018 production, the existing database has been revised and extended. The RDB provides reliable reference information in a consistent manner to support calibration activities, future product updates and monitoring approaches in the Copernicus Land Monitoring Service in view of tree cover and forest change. The database consists of two reference products:
▪ Primary Sampling Units (PSUs) in form of 100m x 100m reference polygons, also referred to as Tree Cover Density Reference Polygons (TCDRP), ▪ Secondary Sampling Units (SSUs) in form of a regular 10 x 10 point grid within PSUs, also referred to as Tree Cover Density Reference Grid (TCDRG).
The reference polygons are fully aligned with the EEA 100m reference grid and distributed across the EEA39 countries following a stratification approach based on historic HRL Forest 2015 products. In total, 6,250 PSUs have been collected, resulting in 625,000 SSUs. Each point has been interpreted towards the presence/absence of tree cover and its associated leaf type using the best available VHR reference sources.
Reference Database for Change Calibration A second RDB has been set up for supporting the derivation of the Tree Cover Change Mask 2015-2018 by collecting samples within the gain and loss strata of the initial pixel-based 20m Tree Cover Change Mask 2015/2018. This specific reference dataset for statistical calibration of changes is characterizing the changes detected by comparing the historic TCM 2015 with the new TCM for the 2018 reference year to ensure that there are no temporal and spatial inconsistencies between the layers of the different time periods, thus ensuring the overall consistency of the time series.
The aim of this RDB is to identify real changes occurred over the time period 2015-2018 but also omission and commission errors in both, 2015 and 2018 Tree Cover Masks. A random stratified sampling design has been used to draw around 47,000 samples, with a minimum of 50 samples per production unit for each gain and loss strata.
Each sample point location has been interpreted using the best available VHR reference sources according to the surroundings within the sample polygon as real gain, real loss, omission 2015 or 2018 and commission 2015 or 2018. This RDB is used as input for the statistical analysis of change for each production unit in order to produce the Derived Correction Layer (DCL) 2015 or 2018. The database consists of two reference products:
HRL Forest 2018 Product User Manual Page 39
▪ Primary Sampling Units (PSUs) in form of identified raw change polygons, also referred to as Tree Cover Change Mask Reference Areas (TCCMRG), ▪ Secondary Sampling Units (SSUs) in form of sample points within a 20 x 20 m pixel of the associated PSU, also referred to as Tree Cover Change Mask Reference Points (TCCMRP).
The reference polygons are fully aligned to 20m pixel grid and distributed across the EEA39 countries following the loss and gain strata of the initial Tree Cover Change Mask. In total, 47,684 PSUs have been collected, resulting in 47,997 SSUs.
Dominant Leaf Type Production Unit Layer The Dominant Leaf Type Production Unit Layer (DLTPUL) is a vector layer showing the number and distribution of production units for the whole EEA39 area including the French Overseas Departments. In total, there are 229 production units with an average size of 28,580 km². These production units define the spatial basis for production of the primary status layers Dominant Leaf Type and Tree Cover Density, considering the environmental stratification according to Metzger et al. (2012). The layer provides for each production unit information on the relevant observation period, the maximum cloud cover and the top 10 time features used in the thematic classification process.
HRL Forest 2018 Product User Manual Page 40
VI. Production Methodology and Workflow
Workflow Description
The HRL Forest 2018 has been produced to a large part on the Copernicus Data and Information Access Services (DIAS) platform MUNDI Web Services. Mundi is a cloud environment enabling a centralised and concurrent large scale processing of extensive EO data volumes, ancillary and In-situ data. Storage resources and computing power are provided by the Open Telecom Cloud (OTC).
Figure 4: Workflow for the HRL Forest 2018 product portfolio
HRL Forest 2018 Product User Manual Page 41
Input Data and Pre-processing The main input data source for all thematic classifications were optical Sentinel-2A & Sentinel-2B L2A-data from the reference year 2018. The EO data has been pre-processed by ESA with state of-the-art pre- processing steps and the scene selection included several parameters like cloud cover and tile coverage.
Sentinel-2 data pre-processing:
▪ Check of geometric consistency + geometric adaption ▪ Atmospheric correction (BoA) ▪ Topographic normalisation ▪ Cloud masking (based on Sen2cor Scene Classification Layer) ▪ Derivation of indices (NDVI, NDWI, IRECI, BRIGHTNESS, NBR)
Reference Databases: Input for Classification and Product Generation One component of the classification step and main input for most of the products are series of temporal features (time features) derived either from single bands or from indices of the pre-processed EO data. The temporal and spectral evolution of each pixel is thereby summarized through a series of statistical parameters, such as mean, median, minimum, maximum, standard deviation, percentiles or interpolated values. These statistical parameters allow the characterization and differentiation of distinct patterns of seasonal, short-term and long-term changes induced by the high dynamic of vegetation during the growing cycle (canopy density, vitality of the plants, chlorophyll status and expansion).
Since phenological cycles and land cover characteristics differ across the European countries a flexible adaptation of the set of time features, as well as the time window for their calculation, from one regional and biogeographical strata to another is mandatory. For most of the bio-geographical regions, an observation period from March to October has been selected. In the far northern parts of Europe, this observation period has been shortened to April to September.
The second component are ancillary/reference data used for the extraction of training samples, and for calibration and validation tasks, mostly taken from the LUCAS 2018 database, LPIS data, OSM data, and from previous HRLs as well as from VHR satellite data.
Classification & Optimization The classification itself consists of several iterations using varying temporal subsets of time series from S-2 bands and/or indices and a most suitable time window out of the Image and Feature Database plus training samples from the Reference Databases. All the named input data build the basis for classification of the interim products per production unit, using a semi-automatic Random Forest (RF) classifier with 500 trees for creation of the preliminary Tree Cover Mask and Dominant Leaf Type. In case of the TCD, a multiple linear regression model is applied using the median features of the RED, BLUE, GREEN, NIR, SWIR1 and SWIR2 bands of S-2. After the initial creation of the interim products, the preliminary results undergo a dedicated post-
HRL Forest 2018 Product User Manual Page 42 processing in order to correct possible classification errors. This comprises not only rule-based plausibility checks and masking but also visual checks and manual enhancements.
Production of Status Layers After completion of the post-processing, the final TCM is used as basis to mask the raw DLT and TCD results in order to derive the final status layers at 10m, which are the primary products. Confidence layers are calculated for both, DLT and TCD, and presenting the uncertainty of the classification and the prediction interval for each 10m pixel, respectively. The two primary products serve as input for the derivation of all further products.
The DLT and TCD at 10m resolution are the main inputs to the spatial intersection and filtering steps, which result in the Forest Type status layer at 10m resolution. Before the MMU of 0.5 ha is being applied, all pixels covered by the additional expert product FADSL are removed from the initial TCM.
Generation of Change Products The Tree Cover Mask for 2018 is produced at 10m resolution according to the specifications and take full advantage of the Sentinel 2 spatial resolution. However, this improved spatial resolution can lead to substantial “changes” compared to the previous epoch that are purely due to the additional level of details that can be captured thanks to the improved spatial resolution. Therefore, even though this additional level of detail should be captured in the 2018 status layer, it is critical to ensure that these are removed from the change layer to ensure that the areas mapped correspond to actual changes between the 2015 and 2018 layers. A first step to achieve this is to ensure that the resolution of the 2015 and 2018 Tree Cover masks are compatible. Therefore, the 2018 Tree Cover mask is aggregated from 10 to 20m resolution (as indicated in the change products specifications) based on a majority rule: if at least 2 out of 4 pixels are classified as tree- covered in the input layer, the resulting 20m pixel are be labelled as “Tree Cover”. Considering that the increased resolution typically leads to a better representation of small tree covered patches, the majority rule seems to be the best compromise in terms of maintaining the initial geometry, filtering out isolated pixels and maintaining original tree covered areas. Since occurrences of values of 254 / 255 do not fully match among both input status layers, a detailed ruleset has been setup to handle all possible cases consistently (Annex 6).
After aggregation a raw difference map can be computed. Taking into account the classification probabilities for 2015 and 2018 as well as the Reference Database for Change Calibration a re-classification within the change strata is applied to detect false positive changes resulting from commission and omission in both reference years. Commissions and omission 2018 were used directly to update and improve the 2018 status products. Commissions and omissions 2015 are stored in the Derived Correction Layer 2015.
Once the re-classification procedure is finalised, actual change areas can be isolated from omission and commission errors. A one-pixel boundary filter is applied and subsequently a 1 ha MMU filter. In other words only changes that satisfy the MMU after the application of the boundary filter (1-pixel erosion) are retained with their original extent (i.e. without erosion). Holes within change patches that do not satisfy the MMU are also filled with the class code of the largest neighbour polygon to generate the final TCCM 1518.
HRL Forest 2018 Product User Manual Page 43
The Dominant Leaf Type Change 1518 is based on the TCCM 1518 and in addition inherits the leaf type from the DLT 2015 and DLT 2018 to further subdivide tree cover loss and tree cover gain. The class 12 (potential leaf type change) depicts areas with both changes in the leaf type between 2015 and 2018 as well as significant changes in the Tree Cover Density. The underlying assumption is that leaf type changes over a period of 3 years are generally rather unlikely and plausible only if accompanied by significant change in the Tree Cover Density (e.g. selective logging, re-forestation). To this boundary filters and MMU were first applied to filter the raw leaf type differences. Subsequently, the mean TCD change 15-18 (ΔTCD) was computed for each class 12 candidate patch. Only those patches with a ΔTCD >μΔTCD ± 2σΔTCD (μΔTCD: Mean TCD difference in the production unit, σΔTCD: Standard deviation of the TCD difference in the production unit) are retained.
Similarly, the DLTC 1215 was produced through a re-processing of the experimental DLTC 1215 from the HRL2015 production. To address some shortcomings of this previously unpublished product the following processing steps where performed:
▪ Class-aggregation and recoding to match the specifications of the DLTC 1518 ▪ MMU-filtering of 1 ha to also fill holes within change patches that do not satisfy the MMU ▪ Filtering of class 12 according to the TCD difference between 2012 and 2015 in the same manner as for the DLTC 1518 (see previous paragraph). As a result of the filtering all instances of class 12 are completely removed from the DLTC 1215. Finally, the Derived Correction Layer 2012 was produced showing all areas which were identified as likely commission and omission errors in the DCL 2015 and which are hence also likely to correspond to commission/ omission of tree cover in the 2012 status layers.
Derivation of Aggregated Products After production and validation of the status layers, several aggregated products at 100m spatial resolution are derived from the 10m products, such as the TCD, FTY, BCD and CCD. All of them are derived by application of rule-based spatial aggregations as described in chapter V.
Other/National Products The Production Unit Layer (PUL) is a vector-based layer which provides information on the footprint of the production units, the criteria for the Sentinel-2 time series selection as a basis for the production of the DLT and TCD 2018 layers as well as the ten most important time features used for the leaf type classification.
The Data Score Layer (DSL) is an intermediate 10m x 10m raster dataset showing the number of valid, e.g. cloud-free observations per pixel that are effectively used for thematic classification. The Data Score Layer provides a comprehensive overview of the amount and quality of the input data.
For the production of national product versions for the EEA39 member states (32 member countries and 7 cooperating countries) several products are re-projected from the ETRS89 LAEA projection to national projections. Those products are all primary and secondary status layers and the respective confidence layers, the aggregated products at 100m and the two change products (DLTC, TCCM) at 20m resolution.
HRL Forest 2018 Product User Manual Page 44
Strengths and Limitations
The HRL Forest portfolio comes with a sound and state of the art methodology and provides users with highly accurate information on tree cover and its associated DLT at pan-European level. Even if the used Random Forest classifier is able to handle outliers to a certain extent, intensive preparation of the input data and a dedicated post-processing (including automatic and semi-automatic steps) was necessary in order to ensure a valid thematic classification.
However, the methodology and products show some limitations due to:
a) EO data availability and quality, b) In-situ data availability and quality, c) Climate conditions and weather extremes.
EO Data Availability and Quality Despite the rapid growth of available EO data in the recent years, there are still regions where data availability is a limiting factor, such as the far Northern Scandinavian regions or the French Overseas Departments (DOMs). The number of available satellite observations is generally lower due to Sentinel-related observation constraints5 and cloud cover conditions. The latter can have a direct impact on the overall Sentinel time series quality, since the suboptimal cloud masking in the S-2 L2A data typically leads to a series of artefacts due to omitted cloud/haze/snow cover. This effect is emphasised in case of frequent cloud/haze cover (e.g. over French Guiana) or snow cover (e.g. Scandinavia).
Sentinel-2 L2A data are terrain corrected by default. The implemented terrain correction algorithm tends to overcorrect north/northwest oriented slopes, resulting in generally bright to very bright areas (pixels) - depending on the given slope and aspect. As reflectance values are significantly increased within these areas, spectral characteristics of tree covered pixels are altered and the tree cover detection and leaf type differentiation becomes more challenging. This is locally resulting in omissions of tree cover within mountainous regions and influencing the Dominant Leaf Type classification (local underestimation of coniferous forest/tree stands) as well as the Tree Cover Density classification (overestimation/ underestimation).
The Sentinel-2A & -2B imagery available from ESA have not yet reached optimal multi-temporal co- registration accuracy, displaying geometric shifts of up to 12.5m (2σ) over time.6 An improved multi-temporal co-registration of both satellites is planned by ESA using the new Copernicus DEM and the validated Global Reference Image (GRI) from end-2020 onwards. Shifts of more than 1 pixel in each direction can have a negative impact on the derived statistical features since it cannot be fully assured that a particular pixel from the S-2 L2A input data consistently observes the same landscape element within the given observation period (e.g. from March to October 2018). In particular in areas with high spatial heterogeneity, detection of small tree-covered linear features (alleys, tree rows) and sparse tree cover is thus more challenging.
5 see Sentinel High Level Operations Plan of 22/07/2019 6 see Sentinel-2 L1C Data Quality Report of 07/07/2020
HRL Forest 2018 Product User Manual Page 45
In-situ Data Availability and Quality Data calibration, classification, quality control and validation all depend on the timeliness, accuracy and reliability of adequate in-situ data and ancillary data sources. Also for the 2018 production both, the availability of up-to-date reference data as well as its reliability are still an issue. Despite progress to facilitate the access and availability of in-situ data for the Copernicus Service projects, some regions still show a lack of adequate in-situ data (e.g. LUCAS data, which is not covering the EEA39 countries). In addition, access constraints, spatial coverage, timeliness and the overall reliability of in-situ data are an issue for large-scale productions as in case of the HRLs leading to additional effort during production.
Climate Conditions and Weather Extremes Long-term climate conditions as well as seasonal or exceptional weather events have implications for the detection of tree cover. The Mediterranean region and Eastern Turkey for example are challenging due to frequent drought effects in general. Bright (dry and/or bare soil) surfaces with sparse tree cover are typical and widespread land cover types in these regions. Particularly the over-radiation effect of the bright surfaces has an impact on the tree cover detection, as the respective areas typically show hardly any characteristic spectral/temporal signals of tree cover. Therefore, further post-processing steps were necessary to compensate for this lack of spectral information within very dry regions. Northern Scandinavian or Alpine regions are challenging due to the short growing periods, long-lasting snow cover and distinct topography.
The drought of summer 2018 (starting already in 2017) affected vast areas of many European countries (e.g. France, Belgium, The Netherlands, Germany, Poland, and Southern Sweden) and had a negative impact on the tree cover detection and the calculation of TCD values from the satellite data. Heat and drought stress led to an untimely leaf colouring and even leaf shedding, which can be directly observed in the input satellite data. The decreased spectral signal hampers the tree cover detection and leads to lower tree cover density values than under “normal” conditions have to be expected. However, since the Tree Cover Density is radiometrically and phenologically sensitive, density values should be depicted in a realistic manner for the reference year 2018.
HRL Forest 2018 Product User Manual Page 46
VII. Quality Assessment
Internal Validation
Each HRL product is subject to an internal validation. This section provides guidance on how the products are validated by defining suitable indicators or metrics. Although the validation is consortium-internal, objectivity is sought as far as possible by following the best-practice established for the validation of Copernicus land cover products and land cover maps in general. The accuracy requirements to be achieved are 90 % for producer’s accuracy (commission) and 90 % for user’s accuracy (omission).
Classification correctness should be evaluated using misclassification rate and/or misclassification matrix. Contrary to logical consistency or completeness, thematic accuracy cannot be subjected to an exhaustive check and has to rely on subsamples. Thus, thematic accuracy assessment has three components: (i) the sampling design, (ii) the response design and (iii) the estimation and analysis procedures.
The stratification and the sampling design consist primarily in selecting an appropriate sampling frame and sampling unit. It is based on the LUCAS (Land Use/Cover Area frame statistical Survey) sampling approach. LUCAS corresponds to a grid of approximately 1,100,000 points throughout the European Union where land cover or land use type is observed. Using LUCAS points ensures traceability and coherence between the different layers.
The response design is the photointerpretation of each sample unit and is based on the independent assessment at the unit level, using reference data from the VHR_IMAGE_2018 dataset. Guiding data in terms of geometry and time stamp are the Sentinel images of 2018 used in the production. The interpretation is supported by additional data sources including publicly available WMS services, street-level views and median Sentinel-2 time feature stacks from the production. The last step, the analysis procedure, consists in analysing the samples in order to draw conclusions for the thematic accuracy of the product. Thematic accuracy is presented in the form of an error matrix resulting from samples interpretation. The different accuracies (Overall, Producer’s and User’s Accuracy) are provided as well as confidence intervals at 95% confidence level. All accuracy values are documented in the relevant delivery reports and metadata.
Thematic Accuracy
Thematic accuracy of the two primary status layers DLT and TCD has been assessed using a stratified sampling approach with 9,695 100x100m plots based on the LUCAS grid. Each plot is filled with a regular 5x5 point grid and has been visually interpreted in terms of tree cover and its leaf type using VHR_IMAGE_2018 data and Sentinel-2 time series data, complemented by additional data sources like virtual globes (e.g. Google Earth Pro).
Area-weighted internal validation results of the Dominant Leaf Type 2018 at pan-European level exceed the target accuracy of 90% for producer’s and user’s accuracy per class. The slightly lower user accuracy of the broadleaved class can be explained by the natural confusion of other vegetation types (e.g. maize, shrubland, and certain grassland types) in particular with the spectral signature of broadleaved trees. Most of the commission errors can be observed in agricultural areas and in transition zones. Confusion with the
HRL Forest 2018 Product User Manual Page 47 coniferous class can be observed in less dense tree stands (shadow effects) and in steep terrain as result of the S-2 L2A topographic overcorrection (brightening-up of coniferous tree/forest stands).
Table 17: Area-weighted confusion matrix of the pan-European DLT 2018 status layer
REFERENCE DLT 2018 Status Layer (weighted) User No Tree Cover Broadleaved Coniferous Total CI at 95% Accuracy
No Tree Cover 6134.66 61.39 29.90 6225.95 98.53% 0.17%
Broadleaved 128.39 1820.52 69.27 2018.18 90.21% 0.43%
MAP DATA MAP Coniferous 26.63 22.08 1402.15 1450.86 96.64% 0.26%
Total 6289.86 1903.99 1501.31 9694.98
Producer 97.54% 95.62% 93.39% 96.52% Overall Accuracy Accuracy
CI at 95% 0.22% 0.30% 0.36% 0.37% CI at 95%
0.913 Kappa
Area-weighted internal validation results of the Tree Cover Density 2018 at pan-European level exceed the target accuracy of 90% for producer’s and user’s accuracy per class, considering a 30% density threshold. This density threshold has been considered appropriate on pan-European scale by the external Copernicus Validation project and was already applied for the validation of the historic 2012 and 2015 products.
Table 18: Area-weighted confusion matrix of the pan-European TCD 2018 status layer with a 30% density threshold
REFERENCE DATA TCD 2018 Status Layer (weighted) No Tree Cover Tree Cover Total User Accuracy CI at 95% (0-29% TCD) (30-100% TCD)
No Tree Cover (0-29% TCD) 6128,88 260,41 6389,28 95.92% 0.28%
Tree Cover (30-100% TCD) 86,07 3219,63 3305,70 97.40% 0.23% MAP DATA MAP
Total 6214,94 3480,04 9694,98
Producer Accuracy 98.62% 92.52% 96.43% Overall Accuracy
CI at 95% 0.17% 0.28% 0.37% CI at 95%
0.913 Kappa
HRL Forest 2018 Product User Manual Page 48
In addition to the 30% density threshold, an assessment at a cut-off value of 1% is provided in Table 19. This allows to evaluate also errors in the lower density ranges which are typically subject to higher uncertainties resulting in slightly lower user’s accuracies and a lower producer’s accuracy for the No Tree Cover class.
Table 19: Area-weighted confusion matrix of the pan-European TCD 2018 status layer with no density threshold applied
REFERENCE DATA TCD 2018 Status Layer (weighted) Tree Cover No Tree Cover Total User Accuracy Confidence Interval (1-100% TCD)
No Tree Cover 3628,08 335,81 3963,89 91.53% 0.35%
Tree Cover (1-100% TCD) 421,58 5309,51 5731,09 92.64% 0.35% MAP DATA MAP
Total 4049,66 5645,33 9694,98
Producer Accuracy 89.59% 94.05% 92.19% Overall Accuracy
CI at 95% 0.41% 0.30% 0.48% CI at 95%
0.912 Kappa
Calibration accuracy of the derived tree cover density values is determined by a linear regression analysis to estimate the relationship between the 9,695 reference plots and the TCD 2018 map product. The calculated coefficient of determination (R²) shows a high level of fit of the estimated regression model with R²=0.89 (see Figure 5).
Disagreements between reference values and map values at 0% density are referring to omission and commission errors within the Tree Cover Mask and are not an issue of the TCD calibration per se. However, it appears that map density values are slightly underestimated at very low to low tree cover density (1-38%) and at very high density (88-100%). This might be explained by a series of reasons: a) the mixed-pixel effect at lower density values (typically less dense tree stands), b) the reduced quality of multi-temporal co-registration between the Sentinel-2 satellites decreasing the spectral signal of tree covered pixels (in case of single trees, tree lines, forest borders) c) the topographic overcorrection of the S-2 L2A input data influencing the spectral signal in hilly terrain, d) the European drought 2018, having caused drought stress, leaf colouring and leaf shedding, resulting in a significant decrease of the spectral signal and thus lower density values, e) the applied response design with a 5x5 point grid, which tends to overestimate the reference tree cover density. It is very difficult to assess the influence of these points on the calculated TCD map values at pan-European level, but it is quite likely that b), c) and d) had a significant influence on the regression models (depending on the region).
HRL Forest 2018 Product User Manual Page 49
The reduced quality of multi-temporal co-registration (b) leads to a stronger mixture of spectral signatures and thus results in lower density values as it might be realistically the case. This is especially the case for small tree covered features (e.g. tree lines, alleys, small tree covered patches and tree cover in urban areas) and sparse canopies on relatively bright surfaces (e.g. olive groves). The topographic overcorrection (c) is resulting in decreased density values in distinct relief. Especially the canopy cover of coniferous trees is systematically underestimated. Even though this issue was addressed during the production, it could be reduced to a certain degree only. One further aspect is the European drought (d) that affected many parts of the EEA39 production area during the last years and in particular in 2018. The unusual weather conditions not only led to leaf colouring and leaf shedding, but also to reduced spectral signals due to the drought stress. Though this could be partially addressed through regional calibration, the impact is still notable and cannot be fully compensated.
Figure 5: Scatter plot of TCD 2018 map values and VHR reference data
Furthermore, it should be considered that the TCD product is derived from a median composite of the S-2 time series over the vegetation period. While this guarantees a better wall-to-wall consistency of the product it also tends to yield values which are more representative for the average tree cover density throughout the main vegetation period rather than the peak of the vegetation period. For the validation of the two change products TCCM 1518 and DLTC 1518 additional 2,000 sample plots were allocated within the change strata as depicted by the final map products. The oversampling of the change
HRL Forest 2018 Product User Manual Page 50 strata is taken into account in the analysis assigning proportionally lower weights to samples within the change strata and proportionally lower weights to samples allocated outside the change areas.
According to the results of the area-weighted assessment as presented in Table 20, the overall quality of the TCCM1518 can be considered as very good with most classes clearly exceeding or approaching the margin of 90% producer´s and user’s accuracy. Notable exception are the producer’s accuracy for New Tree Cover and Loss of Tree Cover with estimated accuracies of 81.79% and 79.70%, respectively. While this indicates a general underestimation of the changes it also illustrates that the TCCM has no obvious bias towards neither gains nor losses.
HRL Forest 2018 Product User Manual Page 51
Table 20: Area-weighted confusion matrix of the TCCM 1518 change layer; Class names: 0=unchanged areas with no tree cover, 1=new tree cover, 2=loss of tree cover, 10=unchanged areas with tree cover
REFERENCE TCCM 1518 (weighted) User’s 0 1 2 10 Total CI at 95% Accuracy
0 7159.27 0.98 13.85 81.96 7256.06 98.67% 0.16%
1 0.1 11.32 0.12 0.2 11.74 96.45% 0.06%
2 4.0 0.0 54.83 3.63 62.46 87.79% 0.44% MAP DATA MAP 10 106.97 1.54 0.0 4256.24 4364.75 97.51% 0.22%
Total 7270.34 13.84 68.8 4342.03 11695
Producer’s Overall 98.47% 81.79% 79.70% 98.02% 98.18% Accuracy Accuracy
Confidenc CI at 95% 0.17% 0.55% 0.59% 0.19% 0.26% e Interval
0.96 Kappa
Table 21 provides the area-weighted accuracy figures for the DLTC 1518. Similarly to the TCCM 1518 the analysis indicates on overall good quality with most classes reaching or exceeding 90% user’s and producer’s accuracies. Notable issues are the lower producer’s accuracies for 1: New broadleaved cover and 4: Loss of coniferous cover. While the metrics for this classes are dominated by a few samples with high weights in the omission strata where omission errors were detected (1 samples for class 1 and 9 samples for class 4) it indicates a general underestimation of the changes and a bias towards New Coniferous Cover in the gain strata.
While the DLTC 1518 is consistent with the DLT 2015 and DLT 2018 the inheritance of the leaf type also leads to an inheritance of uncertainties in the leaf type from both reference years. This manifests in particular in the user’s accuracy of 70.64% for the Loss of broadleaved cover class. This class is particular affected by the general overestimation of broadleaved cover in the DLT 2015; an issue which could not be easily resolved without introducing inconsistencies among the status and change layers.
HRL Forest 2018 Product User Manual Page 52
Table 21: Area-weighted confusion matrix of the DLTC 1518 change layer; Class names: 0=unchanged areas with no tree cover, 1=new broadleaved cover, 2=new coniferous cover, 3=loss of broadleaved cover, 4=loss of coniferous cover, 10=unchanged areas with tree cover
REFERENCE DLTC 1518 (weighted) User’s 0 1 2 3 4 10 Total CI at 95% Accuracy
0 7159.27 0.98 0.0 1.54 12.31 81.96 7256.06 98.67% 0.16%
1 0.08 6.37 0.16 0.12 0.0 0.18 6.91 92.24% 0.09%
2 0.02 0.02 4.77 0.0 0.0 0.02 4.83 98.77% 0.03%
3 2.86 0.0 0.0 15.36 1.75 1.77 21.74 70.64% 0.65% MAP DATA MAP 4 1.14 0.0 0.0 0.32 37.4 1.85 40.71 91.86% 0.34%
10 106.97 1.54 0.0 0.0 0.0 4256.24 4364.75 97.51% 0.22%
Total 7270.34 8.91 4.93 17.34 51.46 4342.02 11695
Producer’s Overall 98.47% 71.5% 96.78% 88.62% 72.67% 98.02% 98.16% Accuracy Accuracy
CI at 95% 0.17% 0.67% 0.06% 0.42% 0.66% 0.19% 0.27% CI at 95%
0.96 Kappa
An internal analysis of regional accuracy metrics across Europe suggest a generally homogenous quality of both, the status layers and the change layers without major outliers. Nevertheless, a tendency for above average quality in Central Europe and below average in transition areas in Northern Scandinavia and complex landscapes in the Mediterranean seems notable (to be confirmed by the external validation).
HRL Forest 2018 Product User Manual Page 53
QA/QC Procedures
All procedures of technical Quality Assurance (QA) and Quality Control (QC) have been implemented consistently and under the coordination and supervision of the project’s Quality Manager. Within the overall QA/QC scheme, dedicated technical QA/QC has been performed throughout the entire HRL production chain, including:
▪ Continuous monitoring and maintenance of the processing infrastructure, ▪ Quality Assurance within the production process, applying planned and systematic checks at various stages between data collection and the final product, as well as ▪ a final Quality Control after the main production of the HRLs (but still before making final data aggregation and re-projection), where the accuracy and precision of the products is being assessed.
The introduction of a centralised cloud processing environment was the key improvement to previous implementations as a more systematic and objective QA/QC mechanism could be ensured.
Quality assurance followed the ISO9001 standards for Quality Management7 and the INSPIRE data quality elements and comprised dedicated procedures of ongoing quality checks (QA breakpoints) during implementation of the production chain, in order to keep persistent control over the various stages of production, assure fitness-for-purpose of the end-products and that all quality requirements are fulfilled, including
▪ Thematic accuracy & consistency ▪ Geometric accuracy & consistency ▪ Logical / topologic consistency ▪ Thematic coding / attributes ▪ Metadata completeness and compliance to INSPIRE.
Priority was given to the target thematic accuracies to be achieved by each product, as well as to the issues of product consistency (spatial and temporal) and homogeneity.
7 ISO 9001:2015
HRL Forest 2018 Product User Manual Page 54
VIII. Terms of Use and Product Technical Support
Terms of Use
The products described in this document were created in the frame of the Copernicus programme of the European Union by the European Environment Agency (product custodian) and are owned by the European Union. The products can be used following the Copernicus full free and open data policy, which allows the use of the products also for any commercial purpose. Derived products created by end users from the products described in this document are owned by the end users, who have all intellectual rights to the derived products.
Citation
In cases of re-dissemination of the products described in this document or when the products are used to create a derived product it is required to provide a reference to the source, as follows:
“© European Union, Copernicus Land Monitoring Service
Product Technical Support
Product technical support is provided by the product custodian through Copernicus Land Monitoring Service helpdesk at [email protected]. Product technical support doesn’t include software specific user support or general GIS or remote sensing support.
HRL Forest 2018 Product User Manual Page 55
IX. References
Literature: Seber, G.A., Lee, A.J., 2003. Linear regression analysis. John Wiley & Sons.
Metzger, M.J., Bunce, R.G., Jongman, R.H., Sayre, R., Trabucco, A., Zomer, R., 2012. A high-resolution bioclimate map of the world: a unifying framework for global biodiversity research and monitoring. Global Ecology and Biogeography 22, 630–638.
Documents of the EU: EUROPEAN COMMISSION (2017): Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions An Action Plan for nature, people and the economy. COM (2017) 198 final. Brussels.
THE COUNCIL OF THE EUROPEAN COMMUNITIES (1992): COUNCIL DIRECTIVE 92/43/EEC of 21 May 1992 on the conservation of natural habitats and of wild fauna and flora (Habitats Directive). OJ L 206 of 22.7.1992
Links: https://europa.eu/european-union/documents-publications/official-documents_en
HRL Forest 2018 Product User Manual Page 56
X. Annexes
Annex 1: Naming Convention of HRL 2018 Products
The following file naming convention will be applied both to raster and vector products for all High Resolution Layer products. All letters except the THEME descriptor are in small (not capital) letters, and no points (“.”) and/or minus (“-“) within file names. The file naming is based on the following descriptors:
THEME YEAR RESOLUTION EXTENT EPSG VERSION
THEME:
▪ 3 letter abbreviation for main products ▪ 4 letter abbreviation for change products ▪ 5 letter abbreviation for additional and expert products
REFERENCE YEAR
▪ 2018, 2015, etc. in four digits ▪ Change products in four digits (e.g. 1518)
RESOLUTION
▪ Four-digit (e.g. 010m and 100m)
EXTENT
▪ 2-digit country code for country deliveries in national projection ▪ “eu” for all deliveries in European Projection
EPSG
▪ 5-digit EPSG code (geodetic parameter dataset code by the European Petroleum Survey Group), see http://www.epsg-registry.org/ ▪ e.g. “03035” for the European LAEA projection
VERSION
• 4-digit qualifier of the version number, starting with “V1_1” for a first full final version, and allowing to capture re-processing/calculation of small changes as (“V1_2”, “V1_3” etc.). In case of major changes a second version should be used (“V2_1”)
HRL Forest 2018 Product User Manual Page 57
Examples and meaning of full product names for final products:
DLT_2018_010m_eu_03035_V1_0.tif Dominant Leaf Type status layer, 2018 reference year, 10m spatial resolution, pan-European product in European projection (EPSG: 3035), first final version.
DLT_2018_010m_eu_03035_V1_1.tif Dominant Leaf Type status layer, 2018 reference year, 10m spatial resolution, pan-European product in European projection (EPSG: 3035), first final version, second delivery after small changes.
HRL Forest 2018 Product User Manual Page 58
Annex 2: File Naming Nomenclature of HRL Forest 2018 Products
Descriptor To be written as Meaning Comments TCD Tree Cover Density Abbreviation to be used for DLT Dominant Leaf Type main lot 2 products FTY Forest Type FADSL Forest Additional Support Layer DLTC Dominant Leaf Type Change Abbreviation to be used for lot TCCM Tree Cover Change Mask 2 change products BCD Broadleaved Cover Density New additional 100m products CCD Coniferous Cover Density to separate broadleaved and coniferous cover TCDRP Tree Cover Density Reference Reference product: Vector Product (vector data) product with selected 100mx100m cells with density information THEME TCDRG Tree Cover Density Reference Reference product: Tree cover Grid (vector data) density point grid TCCMRP Tree Cover Change Reference Reference Product: Vector Product (vector data) product with selected raw change polygons and relevant tree cover change information TCCMRG Tree Cover Change Reference Reference Product: Tree cover Grid (vector data) change point grid NDVIMIN NDVI minimum Biophysical variables: Results NDVIMAX NDVI maximum are provided per production NDVIMEAN NDVI mean unit NDVIP10 NDVI 10th percentile NDVIP90 NDVI 90th percentile REFERENCE YEAR 2012 Reference year 2012 (+/- 1 year) 2015 Reference year 2015 (+/- 1 year) 2018 Reference year 2018 1215 Change 2012-2015 Only for change products 1518 Change 2015-2018 RESOLUTION 010m 10m spatial (pixel) resolution
020m 20m spatial (pixel) resolution
100m 100m spatial (pixel) resolution EXTENT al Albania 2-letter abbreviation for the at Austria country (in national ba Bosnia and Herzegovina projections), “eu” for deliveries be Belgium in European projection and bg Bulgaria additional EEA cell grid code for ch Switzerland tile based delivery (e.g. cy Cyprus eu_E02N65) cz Czech Republic de Germany dk Denmark ee Estonia es Spain (including Andorra)
HRL Forest 2018 Product User Manual Page 59
Descriptor To be written as Meaning Comments eu European Projection mosaic deliver euExxNxx European Projection tile based delivery fi Finland fr France gb United Kingdom gf French Guiana gp Guadeloupe gr Greece hr Croatia hu Hungary ie Ireland im Isle of Man is Iceland it Italy li Liechtenstein lt Lithuania lu Luxembourg lv Latvia me Montenegro mk Macedonia, FYR of mq Martinique mt Malta nl Netherlands no Norway pl Poland pt Portugal re Réunion ro Romania rs Serbia se Sweden si Slovenia sk Slovakia tr Turkey xk Kosovo yt Mayotte EPSG e.g. 03035 ETRS89 LAEA (European 5-digit EPSG code (see Projection) http://www.epsg-registry.org/) VERSION V1_0 First full final version 4-digit qualifier of the version V1_1 Re-delivery of first full final number, starting with “V1_0” version with small changes for a first full final version, and V2_0 Second full final version allowing to capture re- etc. etc. processing/calculation of small changes as (“V1_1”, “V1_2” etc.). In case of major changes, a second version should be used (“V2_0”).
HRL Forest 2018 Product User Manual Page 60
Annex 3: Download Content
All HRL products in European projection can be downloaded from the CLMS website under https://land.copernicus.eu/pan-european/high-resolution-layers. Please note, that an account need to be created in order to login and to download the products. Products can be downloaded as full pan-European mosaic or as tiles with a side length of 1000 km x 1000 km.
National products and expert/reference products are accessible via the following link: Looking for National projection & Expert products? — Copernicus Land Monitoring Service. Raster products are delivered as GeoTIFF (*.tif) with world file (*.tfw), pyramids (*.ovr), attribute table (*.dbf) and statistics (*.aux.xml). Each product is accompanied with product-specific color tables (*.clr & *.txt) and INSPIRE-compliant metadata in XML format and an INSPIRE Mapping Table. In addition, a Coordinate Reference Sheet (CRS) is provided in PDF format, listing the characteristics of the European Terrestrial Reference System 1989. Vector products are provided in shapefile (*.shp) format.
HRL Forest 2018 Product User Manual Page 61
Annex 4: Coordinate Reference System Sheet
National products are accompanied with a PDF providing Coordinate Reference System (CRS) information, including details of parameters used to transform to ETRS89 LAEA projection. CRS information sheets will be static and named as follows: CRS_Information_Sheet_
Table 22: Example of a Coordinate Reference System Sheet for Hungary
National Datum HD72 / EOV type geodetic valid area Hungary Prime meridian Greenwich longitude 0 Ellipsoid GRS 1967 semi major axis 6378160 inverse flattening 298.247167427 Projection Hotine Oblique Mercator (variant B) latitude of origin 47.14439372222222 longitude of origin 19.04857177777778 latitude of 1st standard parallel latitude of 2nd standard parallel scale factor at origin false easting 650000 false northing 200000 EPSG-Code 23700 Software ArcGIS 10.2 European Datum ETRS89 (European Terrestrial Reference System 1989) type geodetic valid area Europe / EUREF Prime meridian Greenwich longitude 0° Ellipsoid GRS 1980 semi major axis 6 378 137 m inverse flattening 298.257222101 Projection Geographic (Ellipsoidal Coordinate System) Datum shift parameters used Operation method Coordinate Frame geocentric X translation 52.684 geocentric Y translation -71.194 geocentric Z translation -13.975 rotation X-axis 0.312 rotation Y-axis 0.1063 rotation Z-axis 0.3729 correction of scale 1.0191 EPSG-Code 1449 Software ArcGIS 10.2
HRL Forest 2018 Product User Manual Page 62
Annex 5: Colour Palettes and Attribute Fields
All HRL products are delivered with an embedded raster colormap including the following attribute fields in the *.tif.dbf file: ["value", "count", “Class_Name”, "AREA_KM2", "AREA_PERC"]. To all product deliveries, both the GIS file specifying the colour palette, and a table listing the RGB values shall be provided in the following formats: • *.clr for GIS colour palettes • *.txt for other purposes
TREE COVER AND FOREST PRODUCTS
Table 23: Colour palette and attributes of new TCD layer
Class Code Class Name Red Green Blue
0 all non-tree covered areas 240 240 240
1 1% tree cover density 253 255 115
2-49 2% to 49% tree cover density colour shades in between
50 50% tree cover density 76 230 0
51-99 51% to 99% tree cover density colour shades in between
100 100% tree cover density 28 92 36
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
Table 24: Colour palette and attributes of DLT layer
Class Code Class Name Red Green Blue
0 all non-tree covered areas 240 240 240
1 broadleaved trees 70 158 74
2 coniferous trees 28 92 36
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
HRL Forest 2018 Product User Manual Page 63
Table 25: Colour palette and attributes of FTY layer
Class Code Class Name Red Green Blue
0 all non-forest areas 240 240 240
1 broadleaved forest 70 158 74
2 coniferous forest 28 92 36
3 mixed zones (only for aggregated 100m layer) 76 133 67
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
Table 26: Colour palette and attributes of FADSL
Class Code Class Name Red Green Blue all non-tree covered areas, and tree cover without 0 240 240 240 urban context or agricultural use trees predominantly used for agricultural practices 3 204 173 71 – broadleaved (from CLC2018) trees in urban context – broadleaved and 4 255 85 0 coniferous (from IMD 2018) trees in urban context – broadleaved and 5 168 56 0 coniferous (from CLC 2018) unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
Table 27: Colour palette and attributes of BCD layer
Class Code Class Name Red Green Blue
0 all non-broadleaved covered areas 240 240 240
1 1% broadleaved cover density 253 255 115
2-49 2-49% broadleaved cover density colour shades in between
50 50% broadleaved cover density 76 230 0
51-99 51-99% broadleaved cover density colour shades in between
100 100% broadleaved cover density 28 92 36
HRL Forest 2018 Product User Manual Page 64
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
Table 28: Colour palette and attributes of CCD layer
Class Code Class Name Red Green Blue
0 all non-coniferous covered areas 240 240 240
1 1% coniferous cover density 253 255 115
2-49 2-49% coniferous cover density colour shades in between
50 50% coniferous cover density 76 230 0
51-99 51-99% coniferous cover density colour shades in between
100 100% coniferous cover density 28 92 36
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
Table 29: Colour palette and attributes of DLTC layer
Class Code Class Name Red Green Blue
0 unchanged areas with no tree cover 255 255 255
1 new broadleaved cover 20 255 20
2 new coniferous cover 0 150 0
3 loss of broadleaved cover 255 0 0
4 loss of coniferous cover 255 128 0
10 unchanged areas with tree cover 191 191 191
12 potential change among dominant leaf types 220 220 10
254 unclassifiable in any of parent status layers 153 153 153
255 outside area 0 0 0
HRL Forest 2018 Product User Manual Page 65
Table 30: Colour palette and attributes of TCCM layer
Class Code Class Name Red Green Blue
0 unchanged areas with no tree cover 255 255 255
1 new tree cover 20 255 20
2 loss of tree cover 255 0 0
10 unchanged areas with tree cover 191 191 191
254 unclassifiable in any of parent status layers 153 153 153
255 outside area 0 0 0
Table 31: Colour palette and attributes of DLTCL
Class Code Class Name Red Green Blue
0 0% classification confidence 255 0 0
1-49 1-49% classification confidence colour shades in between
50 50% classification confidence 255 255 0
51-99 51-99% classification confidence colour shades in between
100 100% classification confidence 8 99 0
101 automatic enhancements 0 200 150
102 manual enhancements 0 128 128
253 all non-tree covered areas 240 240 240
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
HRL Forest 2018 Product User Manual Page 66
Table 32: Colour palette and attributes of TCDCL
Class Code Class Name Red Green Blue
0 0% prediction interval 8 99 0
1-49 2-49% prediction interval colour shades in between
50 50% prediction interval 255 255 0
51-99 51-99% prediction interval colour shades in between
100 100% prediction interval 255 0 0
253 all non-tree covered areas 240 240 240
unclassifiable (no satellite image available, or 254 153 153 153 clouds, shadows, or snow)
255 outside area 0 0 0
HRL Forest 2018 Product User Manual Page 67
Annex 6: Tree Cover Change Mask 1518 - Initial Aggregation Rules
HRL Forest 2018 Product User Manual Page 68