The European Forest Inventory of Disturbances (EFID)

The European Forest Inventory of Disturbances (EFID) dataset provided a reliable reference dataset of forest disturbances that occurred in Northern Europe in the last three decades (2000-2023). The first product includes the sampled square cells. Specifically, stratified sampling was used to define the sample areas for photointerpretation, ensuring both a focus on disturbed areas and representativeness across Europe. The study area was first divided into 250-meter cells, which were classified into four forest cover classes using the 2018 Copernicus forest-type map. Landsat 5, 7, 8, and 9 data were resampled and processed into annual BAP composites (1998–2023), from which the Normalized Burn Ratio (NBR) and its annual change (1985–2022) were calculated. Changes below a 0.1 threshold were excluded, and the cumulative sum of remaining changes defined a disturbance influence index, used to further classify dense forest pixels into forests with no spectral changes, little spectral changes, several spectral changes, or extreme spectral changes. These seven resulting strata guided sample selection, aiming to maximize variability capture, coverage of disturbance types, and detection of low-magnitude events such as thinnings or drought impacts, while accounting for potential photointerpretation errors. Based on the sampled areas, the forest disturbance database was developed, as follows. Photointerpretation is conducted in a GIS environment using both stored satellite images (Landsat BAP, Sentinel-2 Medoid) and on-demand data (PlanetScope). Within each 250×250 m cell, all forest disturbance polygons occurring between 2000 and 2023 are delineated, using a slightly larger area (minimum 4000×4000 m) for context. The most accurate satellite data available for each year are used: Landsat (2000–2015), Sentinel-2 (from 2015), and PlanetScope (from 2017). Delineation guidelines depend on the image resolution, with corresponding minimum mapping units (MMU) and minimum widths: • Landsat (30 m): MMU ≈ 0.1 ha (1000 m²), width ≥ 30 m • Sentinel-2 (10 m): MMU = 500 m², width ≥ 10 m • PlanetScope (3 m): MMU = 100 m², width ≥ 10 m Disturbances with smaller areas can still be mapped for later assessment. The process involves detecting visible changes across years to distinguish real disturbances from noise. Each polygon is described with a set of forest disturbance attributes (Table 1). Multi-year comparisons are essential for identifying disturbance types and estimating recovery time, which can indicate the likely cause (e.g., fast recovery suggests pest, while delayed recovery may suggest thinning). However, identifying the exact disturbance type—especially for low-magnitude events or low-density forests—remains challenging, with the exception of clear cuts, which are more easily recognized. The disturbance database includes key variables such as disturbance type (clear cut, pest, fire, thinning, land use change, windthrow, other), year of occurrence (2000–2023), magnitude (on a 1–5 scale), and years to recovery, defined as the time required to regain 80% of vegetation activity. Photointerpretation is conducted for each survey year, with changes also monitored in the same cells across subsequent years to ensure accurate temporal attribution. Table 1. Forest disturbance polygon (EFID_Photoint_Dataset) db attributes. Attribute Description Values Distu_mag The magnitude of disturbance was determined through photointerpretation of satellite images, enabling experts to integrate contextual information, spatial patterns, and spectral signatures 1-5 Distu_typ Disturbance type determined through photointerpretation of satellite images Coded as: CC=Clear cut, FF= Forest fire, PE= Pest, WT= Wind throw, TH = Thinning, LC= Land use change, OT= Other. Satellite Satellite used to detect the disturbance, prioritizing the best geometric resolution one Coded as S2= Sentinel-2, LT= Landsat, PS = PlanetScope Note Space for more detail explanations e.g. in the case of “other type” specify more details Sample_ID Sample ID of reference Numeric Init_year Initial year when disturbance occurred Year, e.g. 2014 End_year Year of disturbance recovery Year, e.g. 2018 Y2R Year to recover (calculated): End_year – Init_year, e.g. 4 This research was supported by the project "The European Forest Inventory of Disturbances (EFID)" (G-06-2023-2), funded by the European Union through FORWARDS (Horizon Europe Project No. 101084481) grants to third parties managed by European Forest Institute.