Global forest fragmentation change from 2000 to 2020

Forest fragmentation is a major driver of global biodiversity loss and ecosystem degradation. Identifying where forest fragmentation is most severe is fundamental, yet global assessments have reported inconsistent patterns, with some studies finding temperate forests more fragmented than tropical forests, while others suggest accelerating fragmentation in the tropics due to recent deforestation. These discrepancies likely stem from differing definitions and the focus on static versus dynamic temporal scales. Fragmentation is a landscape-level process expressed through decreased patch size, increased number of patches, and expanded edge area, with many ecological effects being immediate but prominent within decades after formation. There is a critical need to quantify both current patterns and recent dynamics of global forest fragmentation and to integrate fragmentation with forest cover change to inform management decisions.

Prior work has characterized global forest fragmentation largely using static landscape pattern metrics, often concluding that tropical forests are relatively contiguous and that temperate forests can be more fragmented (e.g., temperate 1.5× more fragmented than tropical in some studies). Conversely, other studies highlight that tropical regions face the most severe and accelerating fragmentation due to deforestation, with, for example, the fraction of tropical forest edge increasing from 27% in 2000 to 31% in 2010 and concomitant net forest loss in the tropics, while many temperate countries achieved net forest gains. The contrasting findings may reflect differences in temporal framing and definitions of fragmentation. Furthermore, fragmentation and forest cover change are related but can be decoupled; gains in cover do not necessarily reverse fragmentation trends, and focusing solely on either coverage or fragmentation undermines credibility of large-scale landscape dynamics estimates. Ecological consequences of fragmentation can persist for decades to a century, reinforcing the need for recent, dynamic quantification.

The authors developed a synthetic Forest Fragmentation Index (FFI) to capture three core dimensions of fragmentation: edge effect, isolation, and patch size, represented by edge density (ED), patch density (PD), and mean patch area (MPA). They computed static FFI for 2000 and 2020, and a dynamic index ΔFFI (AFFI = FFI2020 − FFI2000), to assess both spatial patterns and temporal changes. Data and processing: High-resolution (30 m) global forest cover maps for 2000 and 2020 were sourced from the Global Land Cover and Land Use Change (GLCLUC) dataset (tree height ≥ 5 m classified as forest), converted to binary forest/non-forest maps, and aggregated within a global grid of 5000 m × 5000 m cells. Forest coverage (FC; percent forest area per grid cell) was computed for each year at 5000 m resolution. In total, ~3.41–3.42 million 5-km grid cells covering global forest landscapes were analyzed. Landscape metrics: ED, PD, and MPA were computed at class level (forest/non-forest) using the landscapemetrics R package. ED = Σ eik / A × 10,000; PD = ni / A × 10,000 × 100; MPA = mean area of forest patches (hectares). Metrics were normalized accounting for metric directionality with respect to fragmentation and cross-year comparability; FFI was the average weighted normalized ED, PD, and (1 − normalized MPA). Dynamic index: AFFI was defined as FFI2020 − FFI2000 (range −1 to 1), where negative values indicate decreased fragmentation. Comparative analyses: Static FFI and AFFI were compared across four climatic zones (tropical, subtropical, temperate, boreal; WCR map, 250 m) and along an altitudinal gradient (GLOBE DEM aggregated to 5 km, categorized into 12 altitude classes). Statistical tests included one-way ANOVA with post-hoc Tukey-HSD/LSD for zonal comparisons, generalized additive models for relationships between AFFI and FFI2000, and linear correlations with altitude. Fragmentation process modes: Based on directions of change in ED, PD, and MPA between 2000 and 2020, eight modes were defined (all combinations of up/down in ED, PD, MPA). Spatial distributions and composition proportions were mapped separately for areas with decreased (AFFI < 0) and increased (AFFI > 0) fragmentation. Hotspots: Three decreased-fragmentation hotspots (western Canada, southern Europe, central China) and three increased-fragmentation hotspots (southeastern Amazon, Congo Basin, central Siberia) were analyzed to quantify changes in normalized ED, PD, and MPA and mode compositions. Drivers: General linear models were fitted relating AFFI to seven explanatory factors aggregated to 5 km: agricultural activity (mean cropland coverage, cropland coverage change; GLCLUC cropland 2003 and 2019), socio-economic intensity (mean nighttime light, nighttime light change; NPP-VIIRS-like 2000 and 2020), demographic pressure (mean population density, population density change; WorldPop 2000 and 2020), and natural disturbance (fire frequency 2000–2020; MODIS MCD64A1). Variables were standardized to 0–1; standardized coefficients and confidence intervals were used to infer drivers globally and within hotspots, and a 50 km grid map of dominant drivers was derived. Two-dimensional framework: To jointly assess area and pattern dynamics, forest landscapes were categorized into four types by the signs of ΔFC (AFC) and ΔFFI: FCupFFIdown (deep recovery), FCupFFIup (early recovery), FCdownFFIdown (early degradation), and FCdownFFIup (deep degradation). Spatial distributions were mapped globally at 5 km, area percentages summarized across climatic zones and altitude, and national-scale averages related via Pearson correlation (n = 131 countries).

• Static patterns: Low static fragmentation (FFI < 0.2) concentrated in the tropics, western Canada, western Siberia, and Far East Russia; high static fragmentation (FFI > 0.8) in eastern North America, southern Europe, central and South China, and tropical forest edges. Area proportions by fragmentation class were relatively stable from 2000 to 2020, with low-fragmentation areas increasing slightly (17% to 19%) and high-fragmentation areas decreasing (17% to 13%). • Dynamic patterns (2000–2020): Approximately 75.1% of global forest landscapes showed decreased fragmentation (AFFI < 0), especially in western Canada, western/Far East Russia, and central/South China. Increased fragmentation (AFFI > 0) occurred mainly in tropical regions (southeastern Amazon, Congo Basin, Indochina), and in parts of western North America and central Siberia; fragmentation was relatively stable in the central Amazon, central/eastern Europe, and the southeastern US. • Climatic zones: Subtropics had the highest static FFI in both years (mean ± SD: 0.64 ± 0.34 in 2000; 0.62 ± 0.31 in 2020; P < 0.001). Tropics (0.43 ± 0.38 in 2000) and boreal (0.41 ± 0.20 in 2020) had the lowest static FFI (P < 0.001). Mean AFFI in the tropics was significantly higher (more positive; 0.01 ± 0.104; P < 0.001) than in other zones (−0.06 to −0.02), indicating more severe recent fragmentation in the tropics. AFFI was negatively related to FFI2000 across all zones (GAM, P < 0.001). Static FFI increased with altitude, while AFFI decreased with altitude, implying lowland forests were relatively intact but underwent more severe fragmentation during 2000–2020. • Fragmentation process modes: In decreased-fragmentation areas (AFFI < 0), the dominant mode was EDdown PDdown MPAup (69.8%), widespread globally; EDdown PDdown MPAdown (15.4%) dominated parts of the central Amazon; EDdown PDup MPAup (8.6%) in eastern Europe. In increased-fragmentation areas (AFFI > 0), EDup PDup MPAdown dominated (53.3%), especially in the tropics, western North America, northern Europe, and central Siberia; EDup PDdown MPAdown (23.6%) and EDdown PDdown MPAdown (8.7%) were also notable in the tropics, Russia, and western Africa. • Hotspot metrics: Decreased-fragmentation hotspots saw declines in ED and PD and increases in MPA: MPA rose by 73% (western Canada), 38% (southern Europe), and 50% (central China). Increased-fragmentation hotspots had strong increases in ED and PD with modest MPA declines: ED increased by 41% (southeastern Amazon), 81% (Congo Basin), and 90% (central Siberia); PD increased by 32%, 186%, and 78%, respectively; MPA decreased slightly (−8% to −31%). • Drivers: At the global scale AFFI was not significantly correlated with individual explanatory variables. Regionally, anthropogenic activity (nighttime light levels/changes, cropland coverage/change) dominated AFFI in developed regions (eastern US, Europe, South China). Wildfire frequency strongly influenced AFFI in Canada, Far East Russia, the southeastern Amazon, tropical Africa, and Australia. In hotspots, key drivers included wildfire frequency (western Canada; southeastern Amazon; central Siberia), mean cropland coverage (southern Europe), and cropland change (central China); the Congo Basin’s AFFI was significantly affected by all factors except nighttime light. • Two-dimensional framework (ΔFC vs ΔFFI): FCupFFIdown landscapes were widespread, concentrated in western Canada, northeastern US, northern Eurasia, and central China, and accounted for the largest shares in temperate (50.0%) and boreal (59.2%) zones. FCdownFFIup was most prevalent in the tropics (39.8%), higher than in subtropical (27.9%), temperate (14.3%), or boreal (10.6%) zones, and distributed in the tropics, northern Europe, and central Siberia. FCupFFIup and FCdownFFIdown accounted for 5.7–7.8% and 24.5–34.2% of area, respectively. • National-scale patterns: Across 131 countries, AFFI was negatively correlated with AFC (R² = 0.35, P < 0.001). Among the ten countries with the largest forest areas, dominant patterns included FCupFFIdown (Russia, China, India), FCdownFFIup (Brazil, Australia, Democratic Republic of the Congo, Peru), and FCdownFFIdown (Canada, USA, Indonesia). China exhibited relatively high AFC and low AFFI (AFC = 1.21%, AFFI = −0.07), while Brazil had low AFC and higher AFFI (AFC = −3.22%, AFFI = 0.01).

By integrating edge, isolation, and patch size into a composite FFI and tracking its change (AFFI), the study distinguishes dynamic fragmentation processes from static distribution patterns. While the tropics retain relatively intact forests, they experienced the most severe increases in fragmentation from 2000 to 2020, likely due to intensified deforestation and disturbances. Conversely, highly fragmented regions in Europe and South China showed recovery linked to afforestation and protection, highlighting that dynamic measures better capture ongoing processes than static metrics alone. The two-dimensional framework coupling ΔFFI with ΔFC resolves inconsistencies when either coverage or fragmentation are considered in isolation, revealing areas where fragmentation decreased despite some cover loss and enabling nuanced classification of recovery (FCupFFIdown, FCupFFIup) versus degradation (FCdownFFIdown, FCdownFFIup) stages. Drivers vary spatially: anthropogenic activities dominate in developed regions, while wildfire plays a key role in boreal and tropical hotspots. These insights support targeted strategies to curb deforestation, enhance connectivity, and prioritize interventions in tropical lowlands where recent fragmentation pressures are greatest.

The study provides a comprehensive, dynamic assessment of global forest fragmentation from 2000 to 2020 using a synthetic FFI and its change (AFFI), coupled with forest cover change in a two-dimensional framework. Key contributions include mapping static and dynamic fragmentation, identifying eight process modes, quantifying regional and national dynamics, and diagnosing drivers across global hotspots. Despite an overall decline in fragmentation for 75.1% of forest landscapes, tropical regions underwent increased fragmentation, underscoring urgent needs to curb deforestation, limit edge creation, and improve connectivity. The negative relationship between AFFI and AFC at national scales indicates that increasing forest area can mitigate fragmentation. Future work should incorporate multi-temporal datasets beyond two time points, integrate species- and vegetation-specific responses, and further link fragmentation dynamics to biodiversity outcomes and carbon-cycle feedbacks to guide policy and restoration.

The assessment relies on bi-temporal forest cover maps (2000 and 2020), which cannot capture continuous dynamics or short-term fluctuations, particularly in areas with shifting cultivation or intensive subtropical forestry. Fragmentation processes are complex and context-dependent; climate variability and change (e.g., warming-driven forest expansion at high latitudes; increased fire frequency in many regions) may influence dynamics in ways not fully disentangled here. While multiple drivers were analyzed, global-scale models did not show significant correlations, suggesting localized, targeted analyses are necessary. Results also depend on the accuracy of input datasets (forest cover, cropland, nighttime lights, population, burned area) and the chosen spatial aggregation (5 km grids). Broader ecological impacts (species-specific effects, ecosystem functions) were not directly measured and warrant integration with biodiversity and carbon datasets, ideally using multi-temporal observations.