3. Forest ecosystem condition accounts

The Ecosystem Conditions accounts are composed of 3 main outputs:

• Variable (descriptor) = the original ecological measurement (raw variable) used to study a certain aspect of an ecosystem condition;
• Indicator = the rescaled ecological measurement (raw variable) to allow a consistent representation of the dimensions of the conditions captured in the variable. Reference values are used to determine good and bad values for a particular variable in the same type of forest (aka forest with similar characteristics), for the sake of comparing those results against other dimensions of condition in the same type of forest, or against results for the same variable in a different area.
• Index = the composite weighted average of a set of indicators, whose goal is to combine different information from several variables into a unique figure, representative of the overall condition of that type of forest in the area under analysis. 

Break-down (categorization) of the results by European Forest

In this analysis, being forests the main objective of this study, the conditions of forests are disaggregated into the classes of forests obtained by combining the information described in this chapter of the guide.

Forest types

European forests have been initially categorized based on their landcover forest class, as per the Corine dataset, which distinguishes 4 types of forest:

1. Broad-leave forest (311) - https://land.copernicus.eu/content/corine-land-cover-nomenclature-guidelines/html/index-clc-311.html
2. Coniferous forest (312) - https://land.copernicus.eu/content/corine-land-cover-nomenclature-guidelines/html/index-clc-312.html
3. Mixed forest (313) - https://land.copernicus.eu/content/corine-land-cover-nomenclature-guidelines/html/index-clc-313.html
4. Transitional woodland & shrub (324) - https://land.copernicus.eu/content/corine-land-cover-nomenclature-guidelines/html/index-clc-324.html

Bioregion zones

Such categories of forest were also combined with the information on the European bioregion zones, used to define 11 bioregions over:

1. Alpine,
2. Arctic,
3. Atlantic,
4. Black Sea,
5. Boreal,
6. Continental,
7. Macaronesian,
8. Mediterranean,
9. Pannonian,
10. Steppic regions
11. Alpine (Scandinavia)

The Scandinavian Alpine zone was differentiated from the rest of the Alpine bioregion for its characteristics and covers the forest in the Scandinavian mountains on the border between Norway and Sweden.

The combination of landcover forests and bioregions results in 44 forest types, following the definition of Maes, J., Bruzón, A.G., Barredo, J.I. et al. Accounting for forest conditions in Europe based on an international statistical standard. Nat Commun 14, 3723 (2023).- https://doi.org/10.1038/s41467-023-39434-0

The following datasets were used to obtain this information:

• Forest landcovers from Corine year 2000 (vector/ https://land.copernicus.eu/en/products/corine-land-cover/clc-2000), and
• Corine accounting adjusted year 2000 (raster/ https://www.eea.europa.eu/en/datahub/datahubitem-view/a55d9224-a326-4cb1-9b9c-3a324520341a?activeAccordion=1069872%2C1069948);
• European bioregion zones (https://sdi.eea.europa.eu/data/def7ac06-7d3f-4da5-880c-a76a73953cfc).
  
  

Forest ecosystem condition variables

Currently 13 variable accounts can be generated in the application.
These are ordered by their ECT classification:

1. A1.1 Normalized Difference Water Index

2. A2.1 Soil Organic Carbon

3. B1.1 Threatened Forest Bird Species Diversity

4. B2.1 Above-ground Biomass

5. B2.2 Leaf Area Index

6. B2.3 Canopy Cover Percentage (Tree Cover Density)

7. B3.1 Net Primary Production

8. B3.2 Fraction of Green Vegetation Cover (FCOVER)
9. B3.4 Drought Severity

10. B3.5 Normalized Difference Vegetation Index

11. C1.1 Forest Connectivity

12. C1.2 Landscape Naturalness
    

13. C1.3 Forest Fragmentation

The table below summarizes all the ecosystem condition metrics considered in the development of the Forest Conditions Accounts

The full list is still under development

Forest ecosystem condition indicators

Each variable is scaled between 0 and 1, with values getting closer to 1 as the forest conditions improve.
This allows each indicator to represent consistently the information coming from the variables, which in their original values have different ranges but also different interpretations of such values, and not always higher values are associated with better conditions (e.g. if we look at variables like Drought Severity or Forest Fragmentation, the higher the value, the worse the conditions of the forests). 

There is currently an indicator for each variable computed in the application.
These are ordered by their ECT classification:

1. A1.1 Normalized Difference Water Index

2. A2.1 Soil Organic Carbon

3. B1.1 Threatened Forest Bird Species Diversity

4. B2.1 Above-ground Biomass

5. B2.2 Leaf Area Index

6. B2.3 Canopy Cover Percentage (Tree Cover Density)

7. B3.1 Net Primary Production

8. B3.2 Fraction of Green Vegetation Cover (FCOVER)
9. B3.4 Drought Severity

10. B3.5 Normalized Difference Vegetation Index

11. C1.1 Forest Connectivity

12. C1.2 Landscape Naturalness
    

13. C1.3 Forest Fragmentation

There are two methodologies adopted to define the Condition Index Accounts, both follow the theoretical framework developed in the scientific publication Maes, J., Bruzón, A.G., Barredo, J.I. et al. Accounting for forest condition in Europe based on an international statistical standard. Nat Commun 14, 3723 (2023), here in after referred to as the Nature Communication method. The differences in their implementation (and their reasons) are fully documented in this guide.

Reference areas representing healthy forests

These areas are used to identify forests considered in good health, to define the upper reference values of that variable. Forests considered in good condition (the upper reference areas) must meet the two following criteria:

• being primary forest, alias forests where the signs of human impacts, if any, are strongly blurred due to decades without forest management or
• being protected areas (from before the year 2000) and highly tree-covered throughout the time series (areas with a tree-covered density loss equal to or higher than 5% are excluded).

Such criteria are observed based on the information of the following datasets:

• Primary forests from EPFD v2.0 - https://www.nature.com/articles/s41597-021-00988-7
• Protected Areas from IUCN data - https://www.protectedplanet.net/en/thematic-areas/wdpa?tab=WDPA
• Copernicus's High-Resolution Layer Tree Cover Density - https://land.copernicus.eu/en/products/high-resolution-layer-tree-cover-density

Results are obtained by combining the following inputs by using, as consistently as possible, the year 2000 as the reference year, except for the information on the Tree-cover density, whose earliest observation is in the year 2012.

Whenever a forest fails to meet any of the listed criteria, it is not considered an area of reference for good condition

Primary forest in the PeopleEA approach:

Primary forests were identified by taking:

• the polygons of primary forests,
• the points of primary forests and buffer to 200m radius.

Primary forest extents match roughly with the Maes et al. figures (Supplementary Table 8)

Protected Area in the PeopleEA approach:

The Protected Areas that have been protected areas from before the year 2000 and which don't exhibit tree-cover loss higher than 5% were considered for this analysis.
The categories of protected areas considered were:

1. Strict Nature Reserves 1a,
2. Wilderness or wildlands 1b, and
3.  National parks 2.

In the Maes approach, depending on the bioregion analysed, different categories of Protected Areas were selected.
The selection of the Categories 1a, 1b & 2 was the combination that matched more closely the Maes et al. figures (Supplementary Table 8)

Healthy reference areas roughly match, but there are significant differences (>5%) for some European forest categories.
The deviations by forest type and by bioregion are detailed in the tables above (ATBD and Maes et al. 2023 Supplementary tables).

Reference values: upper and lower thresholds

• The upper (healthier conditions) reference threshold was set at the 98th percentile value observed in the reference healthy areas, for each type of forest, for each variable. 
  
• The lower (worse conditions) reference threshold was set at the 2nd percentile value observed in the rest of the forest area
  (Total forest area - healthy reference area).
  
•  The percentile might be inverted in case the healthier conditions correspond to lower values.

Rescaling of the variables

The percentile computed define the thresholds used to rescale the observed variables values into indicators ranging from 0 to 1.  

Xi = (X_observed - X_{LowerReference}) / (X_{UpperReference} - X_{LowerReference})

Forest ecosystem condition indices

The Forest Condition Index quantifies forest health by summarizing the information of several indicator accounts, each representing a different aspect of the forest conditions, thus is computed by taking into account the several dimensions represented by each metric (raw variable).
The indicators composing the index aim to represent comprehensively all the components of ecosystems (abiotic, biotic and landscape). The index is usually composed of a single indicator for each Ecosystem Typology Class (ETC), since the variables belonging to the same ETC are usually highly correlated. The indicators are selected considering their relevance, their direct relationship to forest conditions and the availability of data for their measurement.

Forest Condition Index in the People EA methodology

The index built in the People EA project is the result of the weighted average of these indicators:

• the Net Difference Water Index (NDWI),
• Soil Organic Carbon (SOC),
• Threatened forest bird species diversity (TFBSD),
• Above-Ground Biomass (AGB),
• Net Primary Production (NPP)
• and Forest connectivity (FC).

The weights for indicator accounts are determined according to:

• their spatial resolution (i.e. the size of the smallest feature detected by a satellite sensor or displayed in a satellite image, usually expressed as a single value measuring the length of one side of a square),
• temporal resolution (i.e. the total amount of years in a dataset),
• temporal frequency (i.e. the availability of temporal observations over the same period) and
• dataset quality (referring to the proximity of the latest year to the present).

In particular, indicator accounts are ranked according to each of these categories. Larger values of these criteria are associated with a better representation of the ecological conditions and, rank in a higher position. The relative sum of the indicator accounts positions results in their index weight. When data from more than one indicator is considered equal in one particular aspect, each indicator is attributed to the average of the positions they would represent. For example, if three indicator accounts should be ranked 7, they will be assigned 6 since it is the average of the positions these variables represent (5th, 6th and 7th). Ranking and weights attributed to indicator accounts are detailed in the table below:

The index has been designed to take into account one indicator for each Ecosystem Typology Class.

The table ranks each variable considering the aspects in the column headers.

• Spatial Resolution describes the size of the smallest feature that can be detected by a satellite sensor or displayed in a satellite image. It is usually presented as a single value representing the length of one side of a square,
• Temporal resolution refers to the total amount of years in the dataset,
• Temporal frequency considers the availability of temporal observation over the same period. While all variables are annualized, annual averages obtained from highly frequent observation are more accurate and thus considered better input overall.
• Dataset quality represents the proximity of the first and latest year in the dataset, respectively, to the reference year (2000) and the present.
• When variables are ranked the same in one criterion, they are assigned the average of the positions they would represent. For example, if there are three variables which should be ranked 7, they will be assigned 6 since it is the average of the positions these variables represent (5, 6 and 7).

Indicators weighted using an arithmetic average (PeopleEA index)

Overall condition index = 
(Net Difference Water Index x 0.24) + (Soil Organic Carbon x 0.05) + (Threatened Forest Bird Species Diversity x 0.08)+
(Above-ground Biomass x 0.15) + (Net Primary Production x 0.19) + (Forest Connectivity x 0.12)

For areas where Soil Organic Carbon is missing, because of large areas not covered by the dataset, the index is calculated by redistributing the weights:
Overall condition index =
(Net Difference Water Index x 0.31) + (Threatened Forest Bird Species Diversity x 0.11)+
(Above-ground Biomass x 0.19) + (Net Primary Production x 0.24) + (Forest Connectivity x 0.16)

Indicators weighted using a geometric average (Euclidean distance)

The formulation of the resulting Forest Condition Index is the following:
Forest Condition Index = sqrt(
(NDWI^2 x 0.27) + (SOC^2 x 0.07) + (TBFSD^2 x 0.10) + (AGB^2 x 0.18) + (NPP^2 x 0.22) + (FC^2 x 0.16) )

For areas where Soil Organic Carbon is missing, because of large areas not covered by the dataset, the index is calculated by redistributing the weights:
Forest Condition Index = sqrt(
(NDWI^2 x 0.31) + (TBFSD^2 x 0.11) + (AGB^2 x 0.19) + (NPP^2 x 0.24) + (FC^2 x 0.15) )

Forest Condition Index published in Nature Communications by Maes et al. (2023)   

The Forest Condition Index quantifies forest health by summarizing the information of several indicator accounts, each representing a different aspect of the forest conditions. The indicators composing the index aim to represent comprehensively all the components of ecosystems (abiotic, biotic and landscape). The index is usually composed of a single indicator for each Ecosystem Typology Class (ETC), since the variables belonging to the same ETC are usually highly correlated. The indicators are selected considering their relevance, their direct relationship to forest conditions and the availability of data for their measurement.

Specifically, the Forest Condition Index published in Nature Communications considers:

• the Net Difference Water Index,
• Soil Organic Carbon,
• Threatened forest bird species diversity,
• Tree cover density (Canopy cover),
• Net Difference Vegetation Index (Forest productivity),
• Forest connectivity and
• Landscape naturalness.  

To obtain a unique assessment of the forest condition, the index uses the weighted average of indicator accounts. The weights for indicator accounts are assigned based on their capacity to represent the ecosystem structure (intrinsic relevance), the ecosystem services provided (instrumental relevance), the changes in forest condition derived from external factors (directionality meaning) and the anthropogenic repercussion (sensibility to human influence) as well as on their adequacy to the SEEA EA framework (framework adequacy). Indicators are ranked between 1 (lowest position) and 7 (highest position) for five conceptual criteria proposed to select ecosystem condition variables (SEEAEAWhiteCover2021, CzuczEtAl2021  . Better positioned indicators have a larger weight in the index, and the relative sum of the indicator positions results in their index weight. Positions and weights attributed to indicator accounts are detailed in the table below, which corresponds to Maes et al. figures (Supplementary Table 9).

The formulation of the resulting Forest Condition Index is the following:
Forest Condition Index = 
(NDWI x 0.08) + (SOC x 0.12) + (TBFSD x 0.22) + (TCD x 0.21) + (NDVI x 0.13) + (FC x 0.13) + (LN x 0.11)
For areas where Soil Organic Carbon is missing, because of dataset coverage, the index is calculated by redistributing the weights:
Forest Condition Index = 
(NDWI x 0.10) + (TBFSD x 0.26) + (TCD x 0.23) + (NDVI x 0.15) + (FC x 0.14) + (LN x 0.12)