Mapping the coupled human and natural disturbance regimes of ’s

Supplementary Materials

Cornelius Senf and Rupert Seidl Table S1: Number of reference plots and their distribution across countries. Country No-Forest Total Validation Calibration

Albania 415 1132 1547 25 1522 1828 2050 3878 73 3805 442 700 1142 181 961 334 1161 1495 27 1468 393 391 784 44 740 418 863 1281 96 1185 433 558 991 50 941 1401 2847 4248 69 4179 401 2452 2853 38 2815 377 365 742 39 703 450 202 652 294 358 413 507 920 477 443 1227 2615 3842 311 3531 412 956 1,368 109 1,259 349 1,199 1,548 81 1,467 Ireland 298 2,474 2,772 61 2,711 Italy 363 660 1,023 262 761 389 483 872 56 816 416 858 1,274 56 1,218 Moldova 302 2,803 3,105 29 3,076 403 485 888 12 876 322 2,782 3,104 31 3,073 394 621 1,015 271 744 Poland 1,312 3,234 4,546 271 4,275 387 691 1,078 77 1,001 436 1,058 1,494 207 1,287 334 750 1,084 77 1,007 1,820 2,640 4,460 43 4,417 412 251 663 18 645 326 557 883 434 449 439 265 704 391 313 Switzerland 1,128 2,581 3,709 36 3,673 439 647 1,086 22 1,064 451 2,115 2,566 519 2,047 United Kingdom 332 2,508 2,840 213 2,627 Sum 19,996 46,461 66,457 5,000 61,457

Figure S2: Mapping workflow used in this study for mapping no-forest areas, undisturbed forests and disturbed forests across continental Europe.

Figure S3: Example of the spatial filter applied to the disturbance maps. Colors indicate different years of disturbance.

Supplementary Note 1

Our disturbance map had an overall accuracy of 92.5 %, with a disturbance commission error of 14.6 % and a disturbance omission error of 32.8 % (Table S1). Omission errors were mainly related to low severity disturbances that could not be separated from noise (Figure S4) and whose detection is beyond the capacity of current satellite time series analysis approaches. The mean absolute error between the estimated disturbance year and the manually interpreted disturbance year was 3 years (Figure S5), with 77 % of the assigned disturbance years being within three years of the manually interpreted disturbance year. Mapped disturbance severity – that is a continuous value between 0 and 100 indicating the loss of canopy cover within a focal cell during a disturbance event – was well able to differentiate between un-disturbed areas, stand-replacing disturbances, and non-stand- replacing disturbances (Figure S6). Overall, our disturbance severity map yielded an accuracy of 71 % when separating stand-replacing and non-stand-replacing disturbances (Figure S7). Table S1: Confusion matrix (expressed as proportions), and errors of omission and commission, both derived from and independent validation sample of n = 5,000.

Interpreted Mapped Total Commission error rate Disturbance Forest Non-forest Disturbance 0.063 0.010 0.001 0.074 0.146 Forest 0.026 0.236 0.007 0.269 0.121 Non-forest 0.005 0.026 0.626 0.657 0.048 Total 0.094 0.272 0.634 1.000

Omission error rate 0.328 0.132 0.012 Overall error rate: 0.075

Figure S4: Spectral change magnitude in Tasseled Cap Wetness (TCW), Normalized Burn Ration (NBR), Landsat shortwave-infrared I (B5), and Landsat shortwave-infrared II (B7) for all reference pixels (n = 5,000) with commission errors, omission errors and no error (i.e., matching label between mapped and interpreted). For omission errors, spectral change magnitudes were substantially lower than for disturbances, highlighting that many omission errors stem from very low spectral changes indistinguishable from noise in current Landsat- based time series methods.

Figure S5: Estimated years versus manually interpreted years of disturbance for 5,000 independent reference pixels. The majority of the pixels is on or close to the 1:1-line, indicating that the correct year of disturbance is assigned.

Figure S6: Disturbance severity (expressed as proportion) among stand-replacing disturbances, non-stand-replacing disturbances and stable forests. The labels were derived from the reference data and are based on manual interpretation of Landsat time series and auxiliary use of aerial photos. Stand-replacing disturbances have the highest disturbance severities and are well separated from non-stand-replacing disturbances (see following Figure S5).

Figure S7: Separating non-stand-replacing from stand-replacing disturbances (based on manual interpretation) using the disturbance severity measure.

Table S2: Statistics on the size, frequency and severity of disturbances across Europe’s forests.

Quantiles Indicator Mean 0 % 1 % 25 % 50 % 75 % 99 % 100 % Size 1.09 0.18 0.18 0.27 0.45 0.90 10.10 16,617.42 Frequency 0.52 < 0.01 0.02 0.20 0.37 0.63 3.01 31.21 Severity 77.19 < 0.01 22.30 65.24 82.61 93.56 100 100.00

Table S3: Statistics on the trends in size, frequency and severity of disturbances across Europe’s forests.

Indicator Mean Mean Proportion Proportion of Proportion (weighted by of hexagons hexagons with of hexagons forest area) with positive positive trends with no trends (weighted by trend forest area) Size Mean 0.41 0.33 0.67 0.65 0.00 50 % quantile 0.21 0.23 0.17 0.19 0.81 75 % quantile 0.53 0.53 0.53 0.54 0.34 100 % quantile 0.35 0.15 0.58 0.52 0.00 Frequency 1.17 1.19 0.74 0.74 0.02 Severity Mean -0.31 -0.33 0.15 0.12 0.00

Figure S8: Changes in disturbance rates (y-axis; percent of forest area disturbed) in relation to changes in disturbance size (color) and disturbance frequency (x-axis). Trends in disturbance rates are mainly explained by changes in disturbance frequencies (71 %), while changes in disturbance size explained a substantial lower proportion (24 %).

Figure S9: Disturbance size, frequency and severity by (x-axis; see Supplementary Table S4 for abbreviations) and country (color and country-code; see Supplementary Table S5 for abbreviations). A higher spread of countries within an ecoregion indicates stronger differences in disturbance size, frequency and severity among varying forest policies despite belonging to the same forest type.

Figure S10: Trends in disturbance size, frequency and severity by ecoregion (x-axis; see Supplementary Table S4 for abbreviations) and country (color and country-code; see Supplementary Table S5 for abbreviations). A higher spread of countries within an ecoregion indicates stronger differences in disturbance size, frequency and severity among varying forest policies despite belonging to the same forest type.

Table S4: Ecoregion names and abbreviations.

Ecoregion Abbreviation Scandinavian and Russian taiga SaRt Atmf Celtic broadleaf forests Clbf Scandinavian coastal forests Sccf Iceland boreal birch forests and alpine tundra Ibbfaat Scandinavian Montane Birch forest and grasslands SMBfag Arctic desert Arcd North Atlantic moist mixed forests NAmmf Srmf Tyrrhenian-Adriatic Sclerophyllous and mixed forests TSamf Cnmf Rodope montane mixed forests Rmmf Kola Peninsula tundra KlPt Aegean and Western sclerophyllous and mixed forests AaWTsamf Mediterranean conifer and mixed forests Mcamf Central European mixed forests CEmf Corsican montane broadleaf and mixed forests Cmbamf Pindus Mountains mixed forests PMmf Euxine-Colchic broadleaf forests E-bf Iberian conifer forests Ibcf Baltic mixed forests Bltmf Northwest Russian-Novaya Zemlya tundra NRZt Carpathian montane forests Crmf East European EEfs Central Anatolian steppe CnAs Ural montane forests and tundra Umfat Dinaric Mountains mixed forests DMmf Po Basin mixed forests PBmf Pontic steppe Pnts Caledon conifer forests Clcf Northwest Iberian montane forests NImf Faroe Islands boreal grasslands FIbg English Lowlands beech forests ELbf Mediterranean woodlands and forests Mwaf Cyprus Mediterranean forests CyMf Saharan halophytics Shrh Crete Mediterranean forests CrMf Western European broadleaf forests WEbf Pnmf Alps conifer and mixed forests Acamf Illyrian deciduous forests Ildf Northeastern Spain and Southern France Mediterranean forests NSaSFMf Crimean Submediterranean forest complex CSfc Blkmf Italian sclerophyllous and semi-deciduous forests Itsasm-f Appenine deciduous montane forests Admf conifer and mixed forests Pcamf Iberian sclerophyllous and semi-deciduous forests Ibsasm-f Northern Anatolian conifer and deciduous forests NAcadf South Appenine mixed montane forests SAmmf Anatolian conifer and deciduous mixed forests Acadmf Southwest Iberian Mediterranean sclerophyllous and mixed forests SIMsamf Central Anatolian steppe and woodlands CAsaw Southern Anatolian montane conifer and deciduous forests SAmcadf Southeastern Iberian shrubs and woodlands SIsaw Eastern Mediterranean conifer-sclerophyllous-broadleaf forests EMcf Mediterranean dry woodlands and steppe Mdwas

Table S5: Country names and abbreviations.

Country Abbreviation AL Austria AT Belarus BY Belgium BE Bosnia and Herzegovina BA Bulgaria BG Croatia HR Czech Republic CZ Denmark DK Estonia EE Finland FI France FR Germany DE Greece GR Hungary HU Ireland IE Italy IT Latvia LV Lithuania LT Moldova MD Montenegro ME Netherlands NL Norway NO Poland PL Portugal PT Romania RO Serbia RS Slovakia SK Slovenia SI Spain ES Sweden SE Switzerland CH North Macedonia MK Ukraine UA United Kingdom GB

Figure S11: Differences in spatial disturbance patterns between countries but within the same ecoregion: (1) Central European Mixed Forests with larger and more frequent disturbances in Poland compared to Germany. (2) Alps Conifer and Mixed Forests with substantially higher disturbance frequencies in Austria compared to Italy. (3) Carpathian Montane Forests, with widely varying disturbances sizes and frequencies among Poland, Slovakia and Ukraine. (4) Scandinavian and Russian Taiga with differences in disturbance size between Norway and Sweden.

Figure S12: Plantation forests in Portugal and Hungary, characterized by a high disturbance frequencies and severities. For color references please see Figure 1 in the main manuscript.