Linking Structural Forest Heterogeneity and Ecological Processes Using Sentinel-2 and FAD-Based Zoning

Linking Structural Forest Heterogeneity and Ecological Processes Using Sentinel-2 and FAD-Based Zoning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Linking Structural Forest Heterogeneity and Ecological Processes Using Sentinel-2 and FAD-Based Zoning Sanjana Dutt, Jakub Wojtasik², Dimitri Justeau-Allaire, Tarmo K. Remmel, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7923328/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 15 You are reading this latest preprint version Abstract Structural heterogeneity within forests strongly influences ecological function, yet zone-specific diagnostics remain limited. This study presents a reproducible, zone-aware framework that links forest configuration to ecological processes using open remote-sensing and inventory data. In the Tuchola Forest Biosphere Reserve (Poland), we assessed degradation, moisture stress, habitat quality, and structural maturity across Foreground Area Density (FAD)–based zones—Core (≥ 90%), Transitional (40–60%), and Rare (≤ 10%)—by integrating multi-temporal Sentinel-2 imagery (2016, 2020, 2024) with field-observed ecological attributes. Interpretable ensemble models (Extra Trees, LightGBM) and partial-dependence analyses revealed consistent contrasts: Rare zones exhibited early-stage canopy stress linked to structural openness and edge exposure, while Core interiors maintained stable moisture regimes and mature canopy structure. Site-type and stand-age patterns showed that spectral similarity at open edges can mimic maturity, underscoring the importance of combining spectral and structural information in future monitoring. Validated against field observations, the workflow offers a spatially explicit, transferable benchmark for diagnosing ecological variability from open data. The results support zone-specific management strategies—preserving Core interiors, stabilizing Transitional areas through adaptive corridors, and restoring connectivity in Rare zones. Biological sciences/Ecology Earth and environmental sciences/Ecology Earth and environmental sciences/Environmental sciences structural heterogeneity ecological processes Sentinel-2 Foreground Area Density (FAD) interpretable machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction Understanding how forest structural continuity relates to ecological functioning remains a central challenge in remote sensing and landscape ecology. Forest fragmentation—the division of continuous forest cover into smaller, more isolated patches—disrupts processes that regulate biodiversity, hydrology, and biomass productivity [ 20 , 38 ]. Fragmentation per se (independent of habitat loss) alters patch configuration and increases edge exposure, intensifying microclimatic stress through greater solar radiation, wind, and desiccation, thereby elevating fire susceptibility [ 2 , 16 ]. These pressures are particularly pronounced in temperate conifer systems such as the Tuchola Forest Biosphere Reserve (TFBR), where even-aged Scots pine (Pinus sylvestris) stands (> 90% of the area) exhibit uniform canopy structure and shallow rooting. Such homogeneity heightens vulnerability to edge-driven moisture stress, bark-beetle outbreaks, and fire relative to mixed or deciduous stands [ 40 , 7 ]. The resulting changes—canopy thinning, moisture loss, and reduced connectivity—limit the persistence of interior-dependent species that rely on large, contiguous forest tracts [ 4 , 15 ]. Although small patches may function as stepping stones for certain taxa, extensive and well-connected interiors remain essential for maintaining stable populations [ 4 , 15 , 38 ]. Remote sensing provides a powerful means to assess these structural and ecological patterns across scales. Sentinel-2’s 10 m multispectral data enable repeated monitoring of vegetation condition through indices sensitive to canopy greenness and biomass (NDVI, EVI), moisture dynamics (NDMI), and pigment-related structural traits (CI-red-edge, NDRE) [ 24 , 37 , 41 ]. These vegetation indices (VIs) respond to disturbance regimes that often co-vary with fragmentation, yet relatively few studies have integrated multi-temporal Sentinel-2 data, fragmentation-derived structural zoning, and field-based ecological attributes to evaluate how forest structural context influences spectral responses to disturbance [ 25 , 37 ]. Here, we use the Foreground Area Density (FAD) metric to stratify the landscape into Core (≥ 90%), Transitional (40–60%), and Rare (≤ 10%) zones, representing a gradient in canopy continuity and edge exposure. FAD-based zones are thus treated as proxies for fragmentation intensity, providing a structural framework for comparing ecological and spectral variation across forest density classes. This approach enables disentangling of ecological signals from background spectral variation by explicitly accounting for neighborhood configuration [ 36 , 38 ]. To demonstrate this, we integrate multi-temporal Sentinel-2 composites (2016, 2020, 2024) with field-based ecological attributes—degradation, moisture content, site type, and stand age (hereafter FEAs)—from the Polish Forest Data Bank (BDL). These FEAs serve as ground-observed indicators of forest condition, allowing us to link satellite-derived vegetation traits with ecological realities on the ground. Our goal is not to measure fragmentation directly but to develop a reproducible, transferable workflow that connects forest structure and function through open data and interpretable machine learning. By comparing ecological processes across fragmentation-derived structural zones, this study establishes a benchmark for operational, zone-based forest monitoring under intensifying climatic and disturbance pressures. The specific objectives are to: Identify sensitive indicators – determine which Sentinel-2 vegetation indices best capture biomass productivity, moisture stress, pigment dynamics, and structural maturity across fragmentation-derived zones; Map zone-specific patterns – quantify how relationships between vegetation indices (VIs) and FEAs vary among Core, Transitional, and Rare zones; and Evaluate predictive performance – assess how accurately open Sentinel-2 indices predict field-observed ecological attributes using interpretable ensemble models (Extra Trees, LightGBM). By coupling Sentinel-2 data with FAD-based stratification and explainable learning, the framework provides a scalable tool for diagnosing ecological differentiation across structural gradients of fragmentation, supporting targeted management and reproducible monitoring in forested landscapes. 2. Materials and Methods 2.1. Study Area The Tuchola Forest Biosphere Reserve (TFBR; 53°30′N, 17°50′E; ~3,195 km², Fig. 1 ) occupies nutrient-poor fluvioglacial sands on the Pomeranian outwash plain of northern Poland and contains > 900 kettle lakes and Sphagnum spp. peatlands that generate sharp hydrological and edaphic gradients [ 27 ]. TFBR is dominated by even-aged Pinus sylvestris plantations (> 90%), with minor Betula spp., Quercus robur , and Alnus glutinosa . Fragmentation arises primarily from silvicultural clear-cuts and salvage logging, compounded by biotic outbreaks (e.g., Panolis flammea ) and extreme events—the 2012 F3 tornado and the 2017 windstorm are notable examples [ 9 , 12 ]. As shown using Bayesian mapping, edge expansion in TFBR is strongly associated with cropland proximity, younger stands, and high wind exposure [ 12 ]. This configuration-driven change has progressed even where total forest area remains relatively stable, a pattern consistent with fragmentation- per se effects emphasized by Fahrig and observed in other Polish landscapes [ 14 , 23 ]. These characteristics make TFBR an apt natural laboratory for examining how local forest density and neighborhood context modulate ecological processes under fragmentation. 2.2. Data Collection 2.2.1 Sentinel-2 Imagery We used Sentinel-2 Level-2A surface reflectance for the growing season (1 May–31 August) in 2016, 2020, and 2024. When Level-2A data were unavailable in 2016, the corresponding Level-1C scenes were converted to L2A using ESA’s Sen2Cor (atmospheric correction; also provides the Scene Classification Layer, SCL). Cloud-affected pixels were excluded using SCL classes 3 (cloud shadow), 8 (cloud – medium probability), 9 (cloud – high probability), 10 (thin cirrus), and 11 (snow/ice). Bands B2, B3, B4, B5, B8, and B11 were retained for their sensitivity to vegetation biochemistry and canopy structure [ 24 , 41 ]. The complete workflow—pre-processing, vegetation-index (VI) calculation, FAD-based zoning, model fitting, and interpretation—is summarized in Fig. 2 . 2.2.2. Field Data Forest inventory data from the Polish Forest Data Bank (BDL) provided polygon-level attributes of degradation, moisture content, site type, and stand age (FEAs). Polygons correspond to operational management units delineated by the State Forests National Forest Holding. These standardized observations serve as the reference for evaluating how Sentinel-2 VIs represent on-the-ground ecological conditions (cf. [ 32 , 29 ]). Encodings for categorical FEAs are listed in Supplementary Tables S2–S4. All tabular inputs and replication code are archived in Zenodo [ 43 ]. 2.3 Image Processing and Data Integration All Sentinel-2 inputs were Level-2A surface reflectance (0–1). We produced seasonal median composites (May–August) in Google Earth Engine to reduce cloud and phenology noise and reprojected them to PUWG 1992 (EPSG: 2180) at 10 m. Twenty-metre bands (B5, B11) were resampled to 10 m using bilinear interpolation to harmonize spatial resolution, while categorical layers (e.g., masks, classes) were resampled with nearest neighbour. Forested areas were restricted using a binary mask from the Dynamic World “trees” probability layer with a threshold of 0.60 [ 8 , 12 ]. Seventeen VIs (Table 1 ) were computed in Python (rasterio 1.3; NumPy 1.26) spanning greenness/biomass (NDVI, EVI, EVI2, GNDVI, GRNDVI, GSAVI, LAI, DVI), moisture (NDMI, GVMI), pigment/chlorophyll (GARI, MCARI, MTVI2, NDRE, GBNDVI), and soil/shadow correction (CVI, MSAVI) [ 24 , 37 , 41 ]. FEAs were rasterized to 10 m using nearest neighbour to preserve categorical integrity and aligned to the composite grid. Where polygon boundaries did not coincide with pixel edges, attributes were assigned to the pixel containing the polygon centroid, and overlay checks flagged potential misalignments [ 9 , 8 ]. Table 1 Sentinel-2 vegetation indices, ecological domains, and references. Functional Domain Indices Ecological Focus References Greenness / Biomass NDVI, EVI, EVI2, GNDVI, GRNDVI, GSAVI, LAI, DVI Leaf area, productivity, biomass 41, 37 Moisture Stress NDMI, GVMI Canopy water content, drought 37 Pigment / Chlorophyll GARI, MCARI, MTVI2, NDRE, GBNDVI Chlorophyll, nutrient status 24 Soil / Shadow Correction CVI, MSAVI Soil background, shadow 41, 24 2.4. Landscape Stratification FAD was computed in GuidosToolbox [ 36 ] using a moving 243 × 243 pixel window (2.43 km per side; 5.90 km²) on the 10 m grid to quantify neighbourhood forest density—used here as a structural proxy for fragmentation intensity—and highlight sparsely forested areas (Fig. 3 ). The Dynamic World “trees” mask (threshold 0.60; Section 2.3 ) was applied before FAD calculation. From the six native FAD classes, we retained Core (FAD ≥ 90%) , Transitional (40–60%) , and Rare (≤ 10%) as structurally distinct zones representing varying canopy continuity and edge exposure, hence serving as proxies for fragmentation intensity [ 12 , 8 ]. FAD provides a compact, spatially explicit measure of local forest amount and configuration. Although complementary landscape metrics such as edge density or patch cohesion can refine fragmentation characterization, they were not computed here; instead, FAD-based zones were used as the primary structural framework for ecological comparison. FEAs ( degradation , moisture content , site type , stand age ) were linked to FAD classes (Core, Transitional, Rare) for 2016, 2020, and 2024, then rasterized to the 10 m grid for pixel-wise modelling using nearest-neighbour interpolation to preserve categorical integrity: Degradation : eight ordered classes from Degraded to Transformed (Table S2). Moisture content : thirteen ordered classes from very wet bogs to fresh soils (Table S3). Site type : fifteen nominal categories spanning soil–moisture gradients (Table S4). Stand age : continuous years of growth summarized into developmental stages. The combined L2A dataset was > 99% complete; pixels with missing values (< 1%) were removed. No outliers were discarded to retain ecologically meaningful extremes. Ordinal variables ( degradation , moisture ) were integer-encoded to preserve rank; site type was one-hot encoded; stand age remained numeric. Tree-based models required no additional feature scaling [ 31 ]. 2.5 Modelling and Interpretation 2.5.1 Modelling Framework We compared two ensemble regressors—Extremely Randomized Trees (ET) and LightGBM (LGBM)—to predict FEAs within Core, Transitional, and Rare zones across 2016, 2020, and 2024 [ 17 , 22 ]. Regression was preferred over classification because several ordered levels (e.g., degradation, moisture) were uneven or absent in certain zones; treating them as quasi-continuous preserved rank information and avoided unseen categories [31]. Hyperparameters were tuned using spatial k-fold cross-validation with non-overlapping geographic partitions and evaluated on held-out folds. Key ET parameters (number of trees, features per split, minimum samples to split/leaf) were optimized per FEA × zone × year (Supplementary Table S5). Model performance (MSE / MAE) is summarized in Supplementary Table S6; ET was selected for interpretation due to its stability and transparency [26]. 2.5.2 Variable Importance and Partial Dependence Following Breiman [ 5 ], impurity-based importance (from ET) was combined with permutation importance (PI) to identify key drivers across FAD zones and years [ 19 , 30 ]. Impurity scores summarize split-level variance reductions but may favour high-cardinality features; PI provides a model-agnostic estimate by shuffling a feature and recording the increase in mean squared error (MSE), which better reflects out-of-sample impact. PI was repeated with multiple random permutations per feature and the resulting spread summarized. Partial Dependence Plots (PDPs) visualize how VIs influence model predictions while averaging over the distribution of other features [ 28 ]. Let \(\:f\) (⋅) denote the trained model and split the features into a set of interest \(\:{x}_{\text{S}\text{}}\) ​ and its complement \(\:{x}_{c}\) The partial dependence of \(\:f\) with respect to \(\:{x}_{\text{S}\text{}}\) is is defined as $$\:{F}_{S\left(z\right)}=\:{\mathbb{E}}_{\left\{{x}_{C}\right\}\left[f\left(z,\:{x}_{C}\right)\right]}=\:\int\:f\left(z,\:{x}_{C}\right)p\left({x}_{C}\right)d{x}_{C}$$ 1 \(\:\:\) where \(\:p\left({x}_{C}\right)\:\) is the marginal distribution of the complementary features. In practice, we approximate this expectation using the brute-force empirical average over the observeddataset \(\:\:\:\:\:\:\:\:\) $$\:{\widehat{F}}_{S\left(z\right)}=\:\left(\frac{1}{n}\right){\sum\:}_{i=1}^{n}f\left(z,\:{x}_{C},i\right)$$ 2 where \(\:{x}_{\left\{C,i\right\}\:}\) ​are the observed values of the complementary features for sample \(\:i\) , and \(\:n\) is the number of samples used in the average. Because fragmentation effects are context-dependent, we emphasized 2D PDPs (i.e., \(\:\mid\:S\mid\:=2)\) to capture nonlinear responses and interactions between key VI pairs for each FEA, e.g., NDRE–GARI (degradation), NDRE–NDMI (moisture content), NDVI–NDRE (site type), and CVI–NDRE (stand age). To ensure comparability across conditions, PDPs were organized in 3×3 grids—rows corresponding to FAD classes (Rare, Transitional, Core) and columns to years (2016, 2020, 2024). PDPs were implemented using scikit-learn ’s brute-force method, evaluating grids within observed feature ranges to avoid extrapolation. 2.5.3 Model Accuracy Assessment The predictive performance of ET and LGBM models was evaluated using the Mean Squared Error (MSE) and Mean Absolute Error (MAE), computed separately for training and test datasets to assess model fit and generalization. For each FEA, predicted values \(\:{ŷ}_{i}\) were compared with ground-truth values \(\:{y}_{i}\) derived from the Polish Forest Data Bank, rasterized at 10 m resolution to align with Sentinel-2 L2A imagery (Section 2.3.1). MSE was calculated as the average squared difference between predicted and actual values, emphasizing larger deviations: $$\:MSE\:=\:\left(\frac{1}{n}\right){\sum\:}_{i=1}^{n}{\left({y}_{i}-\:{ŷ}_{i}\right)}^{2}$$ 3 MAE was computed as the average absolute difference, providing an interpretable metric in the original units of the ground variable: $$\:MAE\:=\:\left(\frac{1}{n}\right){\sum\:}_{i=1}^{n}\left|{y}_{i}-\:{ŷ}_{i}\right|$$ 4 Here, \(\:n\) represents the number of sampled pixels per fragmentation zone (10% stratified random sample). Metrics were computed per FEA, per zone (Rare, Transitional, Core), and per year (2016, 2020, 2024) to preserve ecological context. Error distributions were further summarized with boxplots comparing predicted and observed values to evaluate model stability across zones. All metrics are reported in Supplementary Table S6. 3. Results The ET model, tuned via hyperparameters such as the number of trees and split thresholds (Supplementary Table S5; Section 2.5.3 ), predicted all FEAs across structural zones: Core, Transitional, and Rare zones for 2016, 2020, and 2024. Its performance was compared with LGBM on held-out test data (Supplementary Table S6; Section 2.5.5). As shown in Fig. 4, ET yielded lower or more stable errors for most FEAs—particularly Degradation , Site Type , and Stand Age —while LGBM performed slightly better for Moisture Content in some zones but with higher variability overall. No statistically significant inter-model or inter-annual differences were observed (Supplementary Table S6), indicating comparable predictive capacity between models. ET was therefore retained for subsequent interpretability analyses owing to its higher stability and transparency. 3.2. FEA Prediction Performance For each FEA, we selected a representative VI pair based on PI rankings (Section 2.5.2 ; Supplementary Figures S3, S5, S7, S9). This procedure ensured interpretability and comparability across structural zones and years while retaining ecological relevance. The selected pairs were NDRE + GARI (Degradation), NDRE + NDMI (Moisture Content), NDVI + NDRE (Site Type), and CVI + NDRE (Stand Age). Representative PI results for each FEA are summarized in the main text, whereas the complete PI grids and alternative VI combinations are presented in the Supplementary Figures S4, S6, S8, and S10. This organization allows readers to follow the principal findings directly in the manuscript while providing full transparency and reproducibility through detailed supplementary materials. 3.2.1. Degradation PI analysis (Supplementary Figure S3) identifies NDRE as the most influential vegetation index for predicting degradation, particularly in Rare zones, while GARI and GRNDVI also rank highly in fragmented forest contexts. PDPs for the NDRE + GARI pair (Fig. 5 ) reveal consistent spectral patterns associated with increasing canopy degradation. In these surfaces, higher response values correspond to intensified canopy stress, characterized by declining red-edge reflectance (NDRE ≈ 0.35–0.45) and reduced pigment-resistant greenness (GARI ≈ 0.60–0.70). Core zones maintain relatively stable NDRE and GARI ranges through all years, indicating limited physiological stress and structural disturbance. Transitional zones exhibit a gradual shift toward lower NDRE–GARI values by 2024, reflecting chlorophyll depletion and partial canopy thinning. Rare zones display the widest spectral spread, with pronounced low-NDRE regions (< 0.40) and depressed GARI values, signifying pigment loss, crown opening, and exposure effects typical of fragmented or repeatedly disturbed stands. Comparable spectral responses are observed for alternative indices such as GRNDVI and NDWI (Supplementary Figure S4), which likewise capture stress-related reductions in chlorophyll content and canopy water retention under conditions of increased structural openness and edge exposure. 3.2.2. Moisture Content PI rankings (Supplementary Figure S5) highlight NDMI and NDRE as dominant predictors, especially under the drier 2020 conditions when moisture stress intensified in Rare zones. PDPs for NDRE + NDMI (Fig. 6) show clear spectral gradients corresponding to hydrological variability. Higher response values denote greater moisture stress, linked to lower near-infrared reflectance (NDMI ≈ 0.20–0.30) and weaker red-edge activity (NDRE ≈ 0.40–0.50). Core zones remain relatively stable, maintaining high NDRE–NDMI values consistent with fresh or moist soils. Transitional zones display moderate declines in NDMI by 2024, indicating partial drying and disrupted water balance. Rare zones exhibit the sharpest decrease (< 0.25 NDMI; < 0.45 NDRE), reflecting progressive desiccation and canopy-level moisture loss caused by fragmentation and exposure. Comparable responses are observed for moisture-sensitive indices such as NDWI and CI red-edge (Supplementary Figure S6), confirming these spectral traits as reliable indicators of hydrological stress in fragmented forests. 3.2.3. Site Type PI results (Supplementary Figure S7) identify NDVI and NDRE as the main predictors of site-type variability, with CI red-edge gaining importance in fragmented zones. PDPs for NDVI + NDRE (Fig. 7) reveal distinct spectral separations among habitat types. Higher response values correspond to fertile or moist forest sites with high NDVI (> 0.70) and stable red-edge reflectance (NDRE ≈ 0.45–0.55), while lower responses represent bog or swamp habitats characterized by reduced chlorophyll activity. Core zones consistently retain high NDVI–NDRE combinations, indicating stable fresh or moist broadleaf and coniferous stands. Transitional zones show intermediate values (~ 0.55–0.65 NDVI) by 2024, suggesting mixtures of moist broadleaf and swamp forests. Rare zones display pronounced spectral contrasts (NDVI < 0.55; NDRE < 0.45), marking shifts toward bog coniferous and riparian floodplain habitats associated with disturbance and altered water tables. Additional VI pairings for site-type prediction are presented in Supplementary Figure S8. 3.2.4. Stand Age PI analysis (Supplementary Figure S9) indicates CVI as the leading predictor of stand-age variation in Core and Transitional zones, with NDRE capturing younger regrowth dynamics in Rare areas. PDPs for CVI + NDRE (Fig. 8 ) show clear spectral relationships between canopy structure and forest maturity. Higher response values denote older stands, characterized by greater chlorophyll density (CVI ≈ 3.0–4.5) and stronger red-edge reflectance (NDRE ≈ 0.55–0.65), whereas lower responses correspond to younger, regenerating canopies with diminished pigment and structural complexity. Core zones are dominated by high-CVI, high-NDRE combinations across all years, reflecting mature, undisturbed stands. Transitional zones exhibit mixed-age structures by 2024, with intermediate spectral ranges (CVI ≈ 2.0–3.0; NDRE ≈ 0.45–0.55) consistent with selective disturbance and regrowth. Rare zones predominantly show low-CVI (< 2.0) and low-NDRE (< 0.45) responses, indicative of young secondary forests and patchy regeneration occurring in structurally open, disturbance-prone zones. Additional VI pairings for stand-age prediction are provided in Supplementary Figure S10. 4. Discussion 4.1 Ecological Drivers and Degradation Trends Across FAD Zones Across FAD-based structural zones, NDRE, GARI, and GRNDVI consistently captured canopy stress, reflecting their sensitivity to pigment decline and early warning signals [ 24 , 34 ]. Treating degradation as a continuous gradient revealed subtle transitions preceding visible structural change, aligning with trait-based remote-sensing frameworks emphasizing anticipatory indicators [ 24 , 37 ]. In Core areas, PDPs were notably stable—consistent with buffered microclimates and structural continuity that reduce edge exposure and maintain physiological equilibrium[ 4 , 15 ]. In contrast, Transitional and Rare zones showed steeper, more variable gradients, symptomatic of greater microclimatic variability, nutrient depletion, and wind exposure near structurally open edges[ 6 , 2 ]. These patterns echo global evidence that canopy pigment decline accelerates once forest continuity drops below critical thresholds of ~ 20–30%[ 20 , 14 ]. The strong and recurring influence of NDRE and GARI in Rare zones underscores their value as early indicators of canopy stress under open or discontinuous cover. Interpreted through FAD-based stratification, these responses reveal how pigment-sensitive indices vary with local canopy density and connectivity, providing a transferable approach for detecting early degradation in temperate forest mosaics. 4.2 Moisture Dynamics and Spectral Predictability Moisture responses diverged clearly by structural context. NDMI and NDRE consistently ranked as the most informative indices for soil–canopy interactions under variable canopy closure [ 32 , 21 ]. Core zones retained stable, fresh-moist conditions—consistent with dense canopies and established root systems that buffer against drought [ 35 ]. Rare zones, in contrast, exhibited stronger moisture contrasts and higher variability, reflecting localized drying under reduced canopy shade and altered evapotranspiration—hallmarks of structurally open forest patches [ 6 , 39 ]. By 2024, slight brightening in Rare-zone PDPs may signal partial hydrological recovery following thinning or disturbance, a pattern also observed elsewhere[ 21 , 35 ]. These contrasts highlight the diagnostic value of FAD-based stratification for isolating hydrological responses otherwise obscured in aggregate forest metrics. Together, they demonstrate how open-data, reproducible workflows can support climate-resilience assessment across structurally heterogeneous forest landscapes. 4.3 Site Type as a Landscape Filter Site-type variation reflected underlying fertility and hydrological gradients captured by NDVI and NDRE, with CIred-edge contributing additional pigment-based contrast. Core zones exhibited homogeneous PDP responses consistent with fertile, moist broadleaf–conifer conditions and the relative stability of intact interiors [ 25 ]. Transitional and Rare zones displayed stronger heterogeneity, reflecting spatial variability in soils, drainage, and successional stage influenced by historical disturbance [ 29 , 9 ]. Modeling site type on a continuous scale enabled direct comparison among zones but may smooth over fine-scale heterogeneity, especially in Rare fragments combining pioneer regrowth with mature remnants. Integrating spectral indices with LiDAR-derived structural variables or soil data could sharpen habitat delineation in such mosaics [1]. Even so, the clear spectral separation across zones shows that commonly used indices such as NDVI and NDRE remain efficient proxies for fertility and site quality in large-area inventories. This integration of spectral and structural information underscores how FAD-based zoning can serve as a practical habitat-filter framework, linking trait-based ecology with reproducible monitoring at the landscape scale. 4.4 Stand Age and Structural Maturity Stand-age patterns linked CVI and NDRE with canopy development, capturing pigment and structural changes during forest succession. Core and Transitional zones displayed stable PDPs characteristic of mature, uniform stands, confirming that optical indices can reliably reflect stand-age variation in low-disturbance contexts [ 40 ]. Rare zones, however, presented greater spectral ambiguity: regenerating edges often exhibited reflectance signatures resembling older stands, risking overestimation of maturity—a known limitation in structurally heterogeneous forests [ 11 ]. Incorporating LiDAR or GEDI-derived metrics alongside vegetation indices would enhance age-structure interpretation and ecological realism, ensuring that optical proxies reflect both pigment and physical canopy development[ 3 , 10 ]. Such integration would improve the precision of maturity mapping in complex forest mosaics, bridging spectral traits and structural dynamics within open-data, reproducible frameworks. 4.5 From Process Detection to Management Application Zone-specific analysis offers a framework for tailoring management interventions to structural context. Core zones, characterized by stable spectral and ecological conditions, remain priorities for strict protection to preserve carbon stocks and interior-dependent biodiversity [ 4 , 15 ]. Transitional zones, which exhibited intermediate and variable responses, could benefit from adaptive strategies such as selective thinning, mixed-species enrichment, and the establishment of ecological corridors to stabilize pigment and moisture dynamics [ 16 ]. Rare zones, displaying the strongest stress signals, align with the need for restoration or passive rewilding aimed at rebuilding canopy continuity and mitigating desiccation risk. These interpretations are consistent with broader landscape evidence showing that small, well-managed forest fragments can significantly enhance connectivity and ecosystem function [ 14 , 18 ]. Spectral indicators such as NDRE, NDMI, and NDVI therefore serve as efficient tools for diagnosing stress and monitoring ecological recovery across structurally heterogeneous forests. Because the workflow relies entirely on open data, similar zone-based diagnostics can be readily applied to other temperate and boreal systems, supporting regionally adapted conservation strategies grounded in transparent and reproducible evidence. 4.6 Limitations and Methodological Considerations Several methodological considerations qualify the interpretation of these findings. Treating categorical field attributes ( degradation , moisture , site type ) as continuous variables improved detection of ecological gradients but may obscure abrupt thresholds—particularly in edge-dominated Rare zones, where variability in composition and microclimate is high. Spatial resolution presents an additional constraint: Sentinel-2’s 10 m pixel size can underrepresent microhabitat heterogeneity, especially in species-rich or structurally diverse stands. Integrating UAV or LiDAR data would offer finer structural context and improve ecological interpretation [ 1 , 10 ]. Transferability beyond temperate, pine-dominated forests may also require local calibration, as spectral overlap among canopy species can reduce generality [ 13 , 25 ]. Despite these caveats, combining ET with PI and PDP visualization effectively linked structural forest heterogeneity to ecological gradients. The open-data and code-based design enhances transparency and reproducibility, while the framework’s interpretability supports future extensions that integrate climatic, edaphic, and LiDAR-based covariates to refine resilience forecasting under different disturbance regimes. 5. Conclusion This study examined how structural forest heterogeneity influences key ecological processes—degradation, moisture dynamics, habitat quality, and structural maturity—across Core, Transitional, and Rare zones in the Tuchola Forest Biosphere Reserve using open Sentinel-2 vegetation indices. Pigment-sensitive indices such as NDRE and GARI consistently acted as early indicators of canopy stress in Rare zones, where structural openness and edge exposure precede physiological decline and reduced habitat quality [ 4 ]. Moisture-sensitive indices like NDMI distinguished the hydrological stability of Core zones from the variability and drying trends observed in Rare areas, reflecting the cumulative effects of disturbance and shallow rooting typical of monodominant Scots pine stands [ 15 , 40 ]. Site-type predictions captured fertility contrasts between closed interiors and structurally mixed mosaics, while stand-age modelling emphasized the risk of misclassifying regenerating edges as mature stands—highlighting the need for integrated spectral and structural interpretation. By combining Foreground Area Density (FAD)–based stratification with interpretable ensemble learning (Extra Trees, LightGBM, Partial Dependence), the analysis establishes a reproducible framework linking spectral traits to ecological processes across structural gradients [ 17 , 22 , 28 ]. The findings demonstrate that open, zone-aware approaches can reveal functional variation otherwise masked in aggregate forest metrics, offering a transferable template for ecological diagnosis. Although developed for a temperate, pine-dominated landscape in northern Poland, the workflow’s modular design—built entirely on open data, transparent preprocessing, and explainable modelling—makes it readily adaptable to other forest systems. Incorporating LiDAR, GEDI, or climatic variables would further refine the framework’s capacity to trace recovery trajectories and anticipate resilience loss under disturbance and drought [ 3 , 10 ]. More broadly, integrating vegetation indices with FAD-based structural zoning advances remote sensing from descriptive mapping toward diagnostic, reproducible assessment of ecological function. Such frameworks can support scalable forest monitoring aligned with global biodiversity and climate objectives by identifying where action is most needed: safeguarding Core interiors, adapting management in Transitional areas, and restoring connectivity in Rare zones. In this way, the study contributes to the broader transition toward interpretable, open-data monitoring of forest resilience and provides a practical foundation for climate-smart, structure-aware conservation planning [ 18 , 38 ]. Declarations Acknowledgements The authors thank Dr. Melaine Aubry Kientz and Prof. Grégoire Vincent for valuable advice and guidance during the internship at the AMAP Laboratory, Montpellier. We also thank the Statistical Analysis Center at Nicolaus Copernicus University for access to their computational cluster and expert statistical feedback. We acknowledge the Bank Danych o Lasach – Lasy Państwowe (Forest Data Bank – State Forests) for providing high-quality forest inventory data collected by professional foresters. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Author contributions S.D. conceived the study, curated data, performed analyses, developed methods and software, validated results, prepared visualizations, and wrote the manuscript. J.W. contributed software, statistical analysis, visualization, validation, and manuscript review and editing. D.J-A. provided supervision and manuscript review and editing. T.K.R. contributed to manuscript review, interpretation of results, and critical revisions. M.K. provided resources, supervision, map visualizations, and manuscript review and editing. Ethics statement This research did not involve any experiments on humans, human data, or live vertebrates/higher invertebrates. All analyses were based on open-access remote sensing datasets and publicly available forest inventory data from the Polish State Forests. Therefore, no ethical approval or informed consent was required. Data availability Derived datasets and code to reproduce all results are archived at Zenodo (DOI: 10.5281/zenodo.17241364 ). The archive includes: (i) tabularized vegetation and ground indices for 2016, 2020, and 2024 across FAD-classified zones (Core, Transitional, Rare); (ii) Python code for preprocessing, FAD zoning, and model replication; and (iii) documentation (README, metadata, data dictionary) and example panels. Sentinel-2 Level-2A imagery and Dynamic World “trees” probabilities are publicly available from the Copernicus Open Access Hub and Google Earth Engine. Ground attributes were obtained from the Polish Forest Data Bank (BDL) under open-data provisions. A formal data citation is included in the reference list. Additional information Competing interests The authors declare no competing interests. Use of AI tools Generative AI tools (ChatGPT by OpenAI and Grok) were used only for language editing and readability improvements. No text, figures, or analyses were generated automatically. All content was reviewed and approved by the authors, who take full responsibility for the work. References Alonzo, M., McFadden, J. P., Nowak, D. J. & Roberts, D. A. Mapping urban forest structure and function using hyperspectral imagery and lidar data. Urban Urban Green. 17 , 135–147. https://doi.org/10.1016/j.ufug.2016.04.003 (2016). Arroyo-Rodríguez, V., Saldana-Vazquez, R. A., Fahrig, L. & Santos, B. A. Does forest fragmentation cause an increase in forest temperature? Ecol. Res. 32 , 81–88. https://doi.org/10.1007/s11284-016-1411-6 (2017). Bauer, L. P. L., Huth, A., Bogdanowski, A., Müller, M. & Fischer, R. Edge effects in Amazon forests: integrating remote sensing and modelling to assess changes in biomass and productivity. Remote Sens. 16 , 501. https://doi.org/10.3390/rs16030501 (2024). Blake, J. G. & Karr, J. R. Species composition of bird communities and the conservation benefit of large versus small forests. Biol. Conserv. 30 , 173–187. https://doi.org/10.1016/0006-3207(84)90065-X (1984). Breiman, L. Random forests. Mach. Learn. 45 , 5–32. https://doi.org/10.1023/A:1010933404324 (2001). Briant, G., Gond, V. & Laurance, S. G. Habitat fragmentation and the desiccation of forest canopies: eastern Amazonia. Biol. Conserv. 143 , 2763–2769. https://doi.org/10.1016/j.biocon.2010.07.024 (2010). Britton, J., Graham, G. & Woloszyn, M. Exploring relationships between drought indices and ecological drought impacts using machine learning and explainable AI. J. Appl. Serv. Climatol . 18 , e009001. https://doi.org/10.46275/JOASC.2024.09.001 (2024). Brown, C. F. et al. Dynamic World, near real-time global 10 m land use land cover mapping. Sci. Data . 9 , 251. https://doi.org/10.1038/s41597-022-01307-4 (2022). Budniak, P. & Zięba, S. Effects of forest fragmentation on the volume of wood resources in managed pine-dominated forests in Poland. Forests 13 , 590. https://doi.org/10.3390/f13040590 (2022). Burns, P., Hakkenberg, C. R. & Goetz, S. J. Multi-resolution gridded maps of vegetation structure from GEDI. Sci. Data . 11 , 881. https://doi.org/10.1038/s41597-024-03668-4 (2024). Dobor, L. et al. Post-disturbance recovery of forest carbon in a temperate forest landscape under climate change. Agric. Meteorol. 263 , 308–322. https://doi.org/10.1016/j.agrformet.2018.08.028 (2018). Dutt, S., Batar, A. K., Sulik, S. & Kunz, M. Forest ecosystem on the edge: fragmentation susceptibility in Tuchola Forest, Poland. Ecol. Indic. 161 , 111980. https://doi.org/10.1016/j.ecolind.2024.111980 (2024). Fahrig, L. Effects of habitat fragmentation on biodiversity. Annu. Rev. Ecol. Evol. Syst. 34 , 487–515. https://doi.org/10.1146/annurev.ecolsys.34.011802.1324199 (2003). Fahrig, L. Ecological responses to habitat fragmentation per se . Annu. Rev. Ecol. Evol. Syst. 48 , 1–23. https://doi.org/10.1146/annurev-ecolsys-110316-022612 (2017). Fahrig, L. et al. Is habitat fragmentation bad for biodiversity? Biol. Conserv. 230 , 179–186. https://doi.org/10.1016/j.biocon.2018.12.026 (2019). Fletcher, R. J. et al. Is habitat fragmentation good for biodiversity? Biol. Conserv. 226 , 9–15. https://doi.org/10.1016/j.biocon.2018.07.022 (2018). Geurts, P., Ernst, D. & Wehenkel, L. Extremely randomized trees. Mach. Learn. 63 , 3–42. https://doi.org/10.1007/s10994-006-6226-1 (2006). González-Ávila, S., Ortega, E. & Martín, B. Potential priority areas for forest-dwelling species in Spain based on the degree of forest fragmentation. J. Maps . 19 , 2223597. https://doi.org/10.1080/17445647.2023.2223597 (2023). Grömping, U. Variable importance assessment in regression: linear regression versus random forest. Am. Stat. 63 , 308–319. https://doi.org/10.1198/tast.2009.08199 (2009). Haddad, N. M. et al. Habitat fragmentation and its lasting impact on Earth’s ecosystems. Sci. Adv. 1, e1500052 (2015). https://doi.org/10.1126/sciadv.1500052 He, L. et al. Temporal dynamics of soil moisture in a northern temperate mixed successional forest after a prescribed intermediate disturbance. Agric. Meteorol. 180 , 22–33. https://doi.org/10.1016/j.agrformet.2013.04.014 (2013). Ke, G. et al. LightGBM: a highly efficient gradient boosting decision tree. Adv. Neural Inf. Process. Syst. 30 , 3149–3157. 10.5555/3294996.3295074 (2017). https://dl.acm.org/doi/ Kozak, J. et al. Forest-cover increase does not trigger forest-fragmentation decrease: Polish Carpathians. Sustainability 10 , 1472. https://doi.org/10.3390/su10051472 (2018). Lausch, A., Erasmi, S., King, D. J., Magdon, P. & Heurich, M. Understanding forest health with remote sensing—Part I: spectral traits, processes and characteristics. Remote Sens. 8 , 1029. https://doi.org/10.3390/rs8121029 (2016). Lausch, A., Erasmi, S., King, D. J., Magdon, P. & Heurich, M. Understanding forest health with remote sensing—Part II: approaches and data models. Remote Sens. 9 , 129. https://doi.org/10.3390/rs9020129 (2017). Łoś, H. et al. Evaluation of XGBoost and LGBM performance in tree species classification with Sentinel-2 data. In IGARSS 2021 – IEEE International Geoscience and Remote Sensing Symposium 5803–5806 (IEEE, 2021). https://doi.org/10.1109/IGARSS47720.2021.9553031 Łuców, D. et al. Pine forest management and disturbance in northern Poland: combining high-resolution paleoecology and remote sensing. Front. Ecol. Evol. 9 , 747976. https://doi.org/10.3389/fevo.2021.747976 (2021). Molnar, C. Interpretable Machine Learning: A Guide for Making Black Box Models Explainable. Leanpub, (2019). https://christophm.github.io/interpretable-ml-book Nguyen, H. T., Nguyen, T. D. & Kappas, M. Land cover and forest type classification by values of vegetation indices and forest structure of tropical lowland forests in central Vietnam. Int. J. For. Res. 8896310 (2020). (2020). https://doi.org/10.1155/2020/8896310 Nicodemus, K. K., Malley, J. D., Strobl, C. & Ziegler, A. The behaviour of random forest permutation-based variable importance measures under predictor correlation. BMC Bioinform . 11 , 110. https://doi.org/10.1186/1471-2105-11-110 (2010). Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12 , 2825–2830 (2011). https://jmlr.org/beta/papers/v12/pedregosa11a.html Peters, J. et al. Random forests as a tool for ecohydrological distribution modelling. Ecol. Model. 207 , 304–318. https://doi.org/10.1016/j.ecolmodel.2007.05.011 (2007). Riitters, K. & Wickham, J. Decline of forest interior conditions in the conterminous United States. Sci. Rep. 2 , 653. https://doi.org/10.1038/srep00653 (2012). Rossini, M., Panigada, C., Meroni, M. & Colombo, R. Assessment of oak forest condition based on leaf biochemical variables and chlorophyll fluorescence. Tree Physiol. 26 , 1487–1496. https://doi.org/10.1093/treephys/26.11.1487 (2006). Schwartz, N. B., Budsock, A. M. & Uriarte, M. Fragmentation, forest structure, and topography modulate impacts of drought in a tropical forest landscape. Ecology 100 , e02677. https://doi.org/10.1002/ecy.2677 (2019). Vogt, P. & Riitters, K. GuidosToolbox: universal digital image object analysis. Eur. J. Remote Sens. 50 , 352–361. https://doi.org/10.1080/22797254.2017.1330650 (2017). Wang, K., Franklin, S. E., Guo, X. & Cattet, M. Remote sensing of ecology, biodiversity and conservation: a review from the perspective of remote-sensing specialists. Sensors 10 , 9647–9667. https://doi.org/10.3390/s101109647 (2010). Wang, Z., Han, L., Wang, L., Shi, H. & Luo, Y. Neighboring patch density or patch size? Which determines the importance of forest patches in maintaining overall landscape connectivity in Kanas, Xinjiang, China. Biology 14 , 881. https://doi.org/10.3390/biology14070881 (2025). Wei, X. et al. Forest harvesting and hydrology in boreal forests: under an increased and cumulative disturbance context. Ecol. Manage. 522 , 120468. https://doi.org/10.1016/j.foreco.2022.120468 (2022). Wulder, M. A., White, J. C., Andrew, M. E., Seitz, N. E. & Coops, N. C. Forest fragmentation, structure, and age characteristics as a legacy of forest management. Ecol. Manage. 258 , 1938–1949. https://doi.org/10.1016/j.foreco.2009.07.041 (2009). Xue, J. & Su, B. Significant remote sensing vegetation indices: a review of developments and applications. J. Sensors 1353691 (2017). (2017). https://doi.org/10.1155/2017/1353691 Yu, X. et al. Spatiotemporal variation and predictability of vegetation coverage in the Beijing–Tianjin–Hebei metropolitan region, China. Theor. Appl. Climatol . 145 , 47–62. https://doi.org/10.1007/s00704-021-03616-x (2021). Dutt, S. & Wojtasik, J. Tabularized datasets of Sentinel-2 vegetation indices (VIs) and ground indices (GIs) for the Tuchola Forest (2016, 2020, 2024). Zenodo . (2025). https://doi.org/10.5281/zenodo.17241364 Additional Declarations No competing interests reported. Supplementary Files SupplementarymaterialsLinkingStructuralForestHeterogeneity.docx Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 07 May, 2026 Reviews received at journal 05 May, 2026 Reviewers agreed at journal 03 May, 2026 Reviews received at journal 02 May, 2026 Reviewers agreed at journal 01 May, 2026 Reviewers agreed at journal 01 May, 2026 Reviewers agreed at journal 29 Apr, 2026 Reviewers agreed at journal 24 Apr, 2026 Reviews received at journal 31 Jan, 2026 Reviewers agreed at journal 21 Jan, 2026 Reviewers invited by journal 07 Nov, 2025 Editor invited by journal 03 Nov, 2025 Editor assigned by journal 25 Oct, 2025 Submission checks completed at journal 25 Oct, 2025 First submitted to journal 22 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7923328","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":546236742,"identity":"aff9ccfa-bd4f-4794-b28f-fccdf005960c","order_by":0,"name":"Sanjana Dutt","email":"data:image/png;base64,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","orcid":"","institution":"Nicolaus Copernicus University","correspondingAuthor":true,"prefix":"","firstName":"Sanjana","middleName":"","lastName":"Dutt","suffix":""},{"id":546236743,"identity":"47dc49db-4cab-41ca-bd4c-7e10fdc90e20","order_by":1,"name":"Jakub Wojtasik²","email":"","orcid":"","institution":"Nicolaus Copernicus University","correspondingAuthor":false,"prefix":"","firstName":"Jakub","middleName":"","lastName":"Wojtasik²","suffix":""},{"id":546236744,"identity":"a1591533-90bc-4166-a75a-f720d994693a","order_by":2,"name":"Dimitri Justeau-Allaire","email":"","orcid":"","institution":"AMAP Laboratory (Botany and Modeling of Plant Architecture and Vegetation)","correspondingAuthor":false,"prefix":"","firstName":"Dimitri","middleName":"","lastName":"Justeau-Allaire","suffix":""},{"id":546236745,"identity":"18713911-27b6-4613-8ee9-8034a64f33a8","order_by":3,"name":"Tarmo K. Remmel","email":"","orcid":"","institution":"York University","correspondingAuthor":false,"prefix":"","firstName":"Tarmo","middleName":"K.","lastName":"Remmel","suffix":""},{"id":546236746,"identity":"09eaf680-cfbd-43da-a2e4-181bf935addc","order_by":4,"name":"Mieczysław Kunz¹","email":"","orcid":"","institution":"Nicolaus Copernicus University","correspondingAuthor":false,"prefix":"","firstName":"Mieczysław","middleName":"","lastName":"Kunz¹","suffix":""}],"badges":[],"createdAt":"2025-10-22 12:08:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7923328/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7923328/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96250406,"identity":"78ccd59b-8b28-43d0-9db0-c14a62cb086d","added_by":"auto","created_at":"2025-11-19 07:38:17","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3327197,"visible":true,"origin":"","legend":"","description":"","filename":"manuscriptLinkingStructuralForestHeterogeneity.docx","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/08c1081fdc68b9ea5ce2fb6f.docx"},{"id":96252290,"identity":"9bb9b735-68e2-466c-b97d-1a13d5f62419","added_by":"auto","created_at":"2025-11-19 07:40:45","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7904,"visible":true,"origin":"","legend":"","description":"","filename":"bee4cf3a38da49868a1a85b68001a7c1.json","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/ff85a4cdf17fda9006e62a24.json"},{"id":96164259,"identity":"1366c783-8cac-4398-8e80-aee1451613fc","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1986395,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementarymaterialsLinkingStructuralForestHeterogeneity.docx","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/3bb48a90e6d4c503f253d68d.docx"},{"id":96252183,"identity":"70d92f3a-9d24-46f2-b0e9-962be3fc0897","added_by":"auto","created_at":"2025-11-19 07:40:36","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":120758,"visible":true,"origin":"","legend":"","description":"","filename":"bee4cf3a38da49868a1a85b68001a7c11enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/73e61a84f7e70b7874430da9.xml"},{"id":96250482,"identity":"af2a0b3c-5f51-4e2f-a67d-3cc8c1547642","added_by":"auto","created_at":"2025-11-19 07:38:27","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":922122,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/fb8f3c333fde751a26de2e2b.jpeg"},{"id":96164261,"identity":"51070da1-4208-44ce-bea0-86a6f83aec0d","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":90494,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/50c7ac21d5a29f2205c0ddc0.png"},{"id":96164278,"identity":"71bce3b9-aefb-4cf3-9871-6a869d96e3eb","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":268505,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/bf26321dc8cbd031dbd91220.jpeg"},{"id":96164266,"identity":"003c1781-7815-46db-969d-ab62ceb2b59a","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":128282,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/1f6bcf9f8c2ee316a1bc471a.png"},{"id":96251242,"identity":"9dd0b0a5-ccce-4c3a-8eaf-33d0237904e4","added_by":"auto","created_at":"2025-11-19 07:39:33","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":480299,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/9b280c042b62d374f07f21b4.png"},{"id":96164280,"identity":"d4448f83-6975-409b-866c-6097e72a4853","added_by":"auto","created_at":"2025-11-18 09:16:56","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":477835,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/5a16fdf9a2452549bbe69dd1.png"},{"id":96164262,"identity":"6fe15d2c-c194-489b-a3bc-d0521b1677c2","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":448525,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/3f5e491d62d32fab5692345e.png"},{"id":96164264,"identity":"0fd51d9f-5c59-4c98-b398-9542b60cb6e3","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":446878,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/b070cb186d7ec2ab88a248df.png"},{"id":96164276,"identity":"81ae7fb9-2c66-4df8-9fbf-1f6183a3b5d7","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":420858,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/ea10bcb68cc5888c9518fa9c.png"},{"id":96164275,"identity":"fd65ee46-5802-42d5-a55b-ee5c47a2fb5c","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17576,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/acf27ba8efa4b6a52140e57a.png"},{"id":96164270,"identity":"4e9d19b6-0387-4f51-93de-45d62b1eff75","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57942,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/4d9a41de1d3053748d8e0237.png"},{"id":96250968,"identity":"461495a7-d39d-443a-b426-e74c993dc4a4","added_by":"auto","created_at":"2025-11-19 07:39:09","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":38109,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/0180c941a1fadb0391fed322.png"},{"id":96164274,"identity":"d3238e02-06ee-4e0e-b464-cc4205a43215","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96029,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/87343acdf27d4838f503a050.png"},{"id":96252244,"identity":"ee78fdaa-0507-4c11-b9ee-e235af02dfcd","added_by":"auto","created_at":"2025-11-19 07:40:42","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80662,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/2a2a64c135e44a82c935e63d.png"},{"id":96164272,"identity":"22e9f1d6-13ca-42d3-a14d-6d34d6b73898","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75943,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/10a2be823386193fc771f690.png"},{"id":96250486,"identity":"984f2fea-4dfa-4d2e-b63b-ee9cf2852e66","added_by":"auto","created_at":"2025-11-19 07:38:27","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":81303,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/5b2c9b94a6f33b2b052e849b.png"},{"id":96164273,"identity":"8321b8fb-600a-4f7a-bb65-716bb02990c8","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"xml","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":120846,"visible":true,"origin":"","legend":"","description":"","filename":"bee4cf3a38da49868a1a85b68001a7c11structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/9dce01db18488f6bfcde16f6.xml"},{"id":96164279,"identity":"6c00b22b-bbff-4b70-8784-b7e940863bb2","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"html","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136009,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/4986afd08ef6452344214ed4.html"},{"id":96164252,"identity":"fe8d87b9-4895-4fd1-b641-d1fbfadbdb69","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":922122,"visible":true,"origin":"","legend":"\u003cp\u003eLocation of the Tuchola Forest Biosphere Reserve (TFBR) in northern Poland, overlaid with CORINE land cover types (2018), major roads, and commune boundaries.\u003c/p\u003e","description":"","filename":"image1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/167455c3f17d2700051dc832.jpeg"},{"id":96164251,"identity":"3de2b10b-dcde-4db7-9947-0de797d42347","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":97803,"visible":true,"origin":"","legend":"\u003cp\u003eForest Fragmentation Analysis Workflow. Overview of the methodology combining Sentinel-2 imagery and forest inventory data, preprocessing steps, vegetation index calculation, FAD-based fragmentation zoning, ensemble modeling (Extra Trees, LightGBM), and visualization through PDPs and boxplots.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/c8e8e8f33786d067c0f192ce.png"},{"id":96164254,"identity":"d258df5d-bc22-4e94-a629-3b6a35df91e6","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":268505,"visible":true,"origin":"","legend":"\u003cp\u003eForeground Area Density (FAD)-derived fragmentation zones—Core (green, ≥90% forest cover), Transitional (orange, 40–60%), and Rare (red, ≤10%)—mapped across TFBR in 2016, 2020, and 2024, showing spatial patterns of forest density.\u003c/p\u003e","description":"","filename":"image3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/910f0e22f6c899298e35fab0.jpeg"},{"id":96164253,"identity":"900a5eaa-fd82-40eb-8302-ca0b74d7b9fe","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":128282,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eTest-set error distributions for ET (orange) and LGBM (blue) across FAD classes (Rare, Transitional, Core) and years (2016, 2020, 2024). Similar distributions across years indicate model consistency; ET was retained for interpretability in subsequent analyses.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/dc152e86e4ef93a7680f595d.png"},{"id":96164256,"identity":"3c2992c7-205d-4b5b-a252-5263b84f57df","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":480299,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ePDPs illustrating degradation using NDRE + GARI across Core, Transitional, and Rare zones (2016, 2020, 2024). Brighter areas indicate more severe degradation (see Supplementary Table S2 for degradation classes).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/77fff4dbae2f90d0daa20eb1.png"},{"id":96164258,"identity":"aa9db8f2-2750-409c-b88e-4398daae3556","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":477835,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ePDPs depicting moisture content using NDRE + NDMI across Core, Transitional, and Rare zones (2016, 2020, 2024). Brighter regions represent drier conditions (refer to Supplementary Table S3 for detailed moisture classes).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/c4f8bf00d39f8ef71e71f6f5.png"},{"id":96164268,"identity":"6f5224cb-cec1-4d3f-829c-2a4bb1e5d45d","added_by":"auto","created_at":"2025-11-18 09:16:55","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":448525,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ePDPs illustrating site type conditions using NDVI + NDRE across Core, Transitional, and Rare zones (2016, 2020, 2024). Brighter regions indicate habitats with more fertile and fresh conditions, whereas darker regions represent bog or swamp habitats (see Supplementary Table S4 for detailed site type categories).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/61ddd69541ae0a3afd22f607.png"},{"id":96251295,"identity":"3b0c0e5e-9e17-43fc-9781-2805f6e89070","added_by":"auto","created_at":"2025-11-19 07:39:36","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":446878,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ePDPs illustrating stand age patterns using CVI + NDRE across Core, Transitional, and Rare zones (2016, 2020, 2024). Brighter regions indicate older forest stands.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/1ba67cd24eb3ccd52d6bb685.png"},{"id":96452777,"identity":"98299161-d2c7-49c5-87c5-67759a720527","added_by":"auto","created_at":"2025-11-21 09:44:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3960812,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/5519f475-36c7-49bd-a057-c4102201de92.pdf"},{"id":96252161,"identity":"a44b9a25-923f-4597-8e96-02562d7b44b6","added_by":"auto","created_at":"2025-11-19 07:40:33","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1986395,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementarymaterialsLinkingStructuralForestHeterogeneity.docx","url":"https://assets-eu.researchsquare.com/files/rs-7923328/v1/73f36498bef63582b8ebf0a1.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Linking Structural Forest Heterogeneity and Ecological Processes Using Sentinel-2 and FAD-Based Zoning","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eUnderstanding how forest structural continuity relates to ecological functioning remains a central challenge in remote sensing and landscape ecology. Forest fragmentation\u0026mdash;the division of continuous forest cover into smaller, more isolated patches\u0026mdash;disrupts processes that regulate biodiversity, hydrology, and biomass productivity [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Fragmentation per se (independent of habitat loss) alters patch configuration and increases edge exposure, intensifying microclimatic stress through greater solar radiation, wind, and desiccation, thereby elevating fire susceptibility [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. These pressures are particularly pronounced in temperate conifer systems such as the Tuchola Forest Biosphere Reserve (TFBR), where even-aged Scots pine (Pinus sylvestris) stands (\u0026gt;\u0026thinsp;90% of the area) exhibit uniform canopy structure and shallow rooting. Such homogeneity heightens vulnerability to edge-driven moisture stress, bark-beetle outbreaks, and fire relative to mixed or deciduous stands [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. The resulting changes\u0026mdash;canopy thinning, moisture loss, and reduced connectivity\u0026mdash;limit the persistence of interior-dependent species that rely on large, contiguous forest tracts [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Although small patches may function as stepping stones for certain taxa, extensive and well-connected interiors remain essential for maintaining stable populations [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eRemote sensing provides a powerful means to assess these structural and ecological patterns across scales. Sentinel-2\u0026rsquo;s 10 m multispectral data enable repeated monitoring of vegetation condition through indices sensitive to canopy greenness and biomass (NDVI, EVI), moisture dynamics (NDMI), and pigment-related structural traits (CI-red-edge, NDRE) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. These vegetation indices (VIs) respond to disturbance regimes that often co-vary with fragmentation, yet relatively few studies have integrated multi-temporal Sentinel-2 data, fragmentation-derived structural zoning, and field-based ecological attributes to evaluate how forest structural context influences spectral responses to disturbance [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eHere, we use the Foreground Area Density (FAD) metric to stratify the landscape into Core (\u0026ge;\u0026thinsp;90%), Transitional (40\u0026ndash;60%), and Rare (\u0026le;\u0026thinsp;10%) zones, representing a gradient in canopy continuity and edge exposure. FAD-based zones are thus treated as proxies for fragmentation intensity, providing a structural framework for comparing ecological and spectral variation across forest density classes. This approach enables disentangling of ecological signals from background spectral variation by explicitly accounting for neighborhood configuration [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTo demonstrate this, we integrate multi-temporal Sentinel-2 composites (2016, 2020, 2024) with field-based ecological attributes\u0026mdash;degradation, moisture content, site type, and stand age (hereafter FEAs)\u0026mdash;from the Polish Forest Data Bank (BDL). These FEAs serve as ground-observed indicators of forest condition, allowing us to link satellite-derived vegetation traits with ecological realities on the ground.\u003c/p\u003e\u003cp\u003eOur goal is not to measure fragmentation directly but to develop a reproducible, transferable workflow that connects forest structure and function through open data and interpretable machine learning. By comparing ecological processes across fragmentation-derived structural zones, this study establishes a benchmark for operational, zone-based forest monitoring under intensifying climatic and disturbance pressures.\u003c/p\u003e\u003cp\u003eThe specific objectives are to:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eIdentify sensitive indicators\u003c/b\u003e \u0026ndash; determine which Sentinel-2 vegetation indices best capture biomass productivity, moisture stress, pigment dynamics, and structural maturity across fragmentation-derived zones;\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eMap zone-specific patterns\u003c/b\u003e \u0026ndash; quantify how relationships between vegetation indices (VIs) and FEAs vary among Core, Transitional, and Rare zones; and\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eEvaluate predictive performance\u003c/b\u003e \u0026ndash; assess how accurately open Sentinel-2 indices predict field-observed ecological attributes using interpretable ensemble models (Extra Trees, LightGBM).\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\u003cp\u003eBy coupling Sentinel-2 data with FAD-based stratification and explainable learning, the framework provides a scalable tool for diagnosing ecological differentiation across structural gradients of fragmentation, supporting targeted management and reproducible monitoring in forested landscapes.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1. Study Area\u003c/h2\u003e\u003cp\u003eThe Tuchola Forest Biosphere Reserve (TFBR; 53\u0026deg;30\u0026prime;N, 17\u0026deg;50\u0026prime;E; ~3,195 km\u0026sup2;, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) occupies nutrient-poor fluvioglacial sands on the Pomeranian outwash plain of northern Poland and contains\u0026thinsp;\u0026gt;\u0026thinsp;900 kettle lakes and \u003cem\u003eSphagnum spp.\u003c/em\u003e peatlands that generate sharp hydrological and edaphic gradients [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. TFBR is dominated by even-aged \u003cem\u003ePinus sylvestris\u003c/em\u003e plantations (\u0026gt;\u0026thinsp;90%), with minor \u003cem\u003eBetula\u003c/em\u003e spp., \u003cem\u003eQuercus robur\u003c/em\u003e, and \u003cem\u003eAlnus glutinosa\u003c/em\u003e. Fragmentation arises primarily from silvicultural clear-cuts and salvage logging, compounded by biotic outbreaks (e.g., \u003cem\u003ePanolis flammea\u003c/em\u003e) and extreme events\u0026mdash;the 2012 F3 tornado and the 2017 windstorm are notable examples [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. As shown using Bayesian mapping, edge expansion in TFBR is strongly associated with cropland proximity, younger stands, and high wind exposure [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. This configuration-driven change has progressed even where total forest area remains relatively stable, a pattern consistent with fragmentation-\u003cem\u003eper se\u003c/em\u003e effects emphasized by Fahrig and observed in other Polish landscapes [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. These characteristics make TFBR an apt natural laboratory for examining how local forest density and neighborhood context modulate ecological processes under fragmentation.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2. Data Collection\u003c/h2\u003e\u003cdiv id=\"Sec5\" class=\"Section3\"\u003e\u003ch2\u003e2.2.1 Sentinel-2 Imagery\u003c/h2\u003e\u003cp\u003eWe used Sentinel-2 Level-2A surface reflectance for the growing season (1 May\u0026ndash;31 August) in 2016, 2020, and 2024. When Level-2A data were unavailable in 2016, the corresponding Level-1C scenes were converted to L2A using ESA\u0026rsquo;s Sen2Cor (atmospheric correction; also provides the Scene Classification Layer, SCL).\u003c/p\u003e\u003cp\u003eCloud-affected pixels were excluded using SCL classes 3 (cloud shadow), 8 (cloud \u0026ndash; medium probability), 9 (cloud \u0026ndash; high probability), 10 (thin cirrus), and 11 (snow/ice). Bands B2, B3, B4, B5, B8, and B11 were retained for their sensitivity to vegetation biochemistry and canopy structure [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. The complete workflow\u0026mdash;pre-processing, vegetation-index (VI) calculation, FAD-based zoning, model fitting, and interpretation\u0026mdash;is summarized in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e2.2.2. Field Data\u003c/h2\u003e\u003cp\u003eForest inventory data from the Polish Forest Data Bank (BDL) provided polygon-level attributes of degradation, moisture content, site type, and stand age (FEAs). Polygons correspond to operational management units delineated by the State Forests National Forest Holding. These standardized observations serve as the reference for evaluating how Sentinel-2 VIs represent on-the-ground ecological conditions (cf. [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]).\u003c/p\u003e\u003cp\u003eEncodings for categorical FEAs are listed in Supplementary Tables S2\u0026ndash;S4. All tabular inputs and replication code are archived in Zenodo [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Image Processing and Data Integration\u003c/h2\u003e\u003cp\u003eAll Sentinel-2 inputs were Level-2A surface reflectance (0\u0026ndash;1). We produced seasonal median composites (May\u0026ndash;August) in Google Earth Engine to reduce cloud and phenology noise and reprojected them to PUWG 1992 (EPSG: 2180) at 10 m. Twenty-metre bands (B5, B11) were resampled to 10 m using bilinear interpolation to harmonize spatial resolution, while categorical layers (e.g., masks, classes) were resampled with nearest neighbour.\u003c/p\u003e\u003cp\u003eForested areas were restricted using a binary mask from the Dynamic World \u0026ldquo;trees\u0026rdquo; probability layer with a threshold of 0.60 [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eSeventeen VIs (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) were computed in Python (rasterio 1.3; NumPy 1.26) spanning greenness/biomass (NDVI, EVI, EVI2, GNDVI, GRNDVI, GSAVI, LAI, DVI), moisture (NDMI, GVMI), pigment/chlorophyll (GARI, MCARI, MTVI2, NDRE, GBNDVI), and soil/shadow correction (CVI, MSAVI) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFEAs were rasterized to 10 m using nearest neighbour to preserve categorical integrity and aligned to the composite grid. Where polygon boundaries did not coincide with pixel edges, attributes were assigned to the pixel containing the polygon centroid, and overlay checks flagged potential misalignments [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSentinel-2 vegetation indices, ecological domains, and references.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFunctional Domain\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIndices\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEcological Focus\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eReferences\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGreenness / Biomass\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNDVI, EVI, EVI2, GNDVI, GRNDVI, GSAVI, LAI, DVI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLeaf area, productivity, biomass\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e41, 37\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMoisture Stress\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNDMI, GVMI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCanopy water content, drought\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e37\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePigment / Chlorophyll\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGARI, MCARI, MTVI2, NDRE, GBNDVI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eChlorophyll, nutrient status\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e24\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSoil / Shadow Correction\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCVI, MSAVI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSoil background, shadow\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e41, 24\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.4. Landscape Stratification\u003c/h2\u003e\u003cp\u003eFAD was computed in GuidosToolbox [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e] using a moving 243 \u0026times; 243 pixel window (2.43 km per side; 5.90 km\u0026sup2;) on the 10 m grid to quantify neighbourhood forest density\u0026mdash;used here as a structural proxy for fragmentation intensity\u0026mdash;and highlight sparsely forested areas (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The Dynamic World \u0026ldquo;trees\u0026rdquo; mask (threshold 0.60; Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e2.3\u003c/span\u003e) was applied before FAD calculation. From the six native FAD classes, we retained \u003cem\u003eCore (FAD\u0026thinsp;\u0026ge;\u0026thinsp;90%)\u003c/em\u003e, \u003cem\u003eTransitional (40\u0026ndash;60%)\u003c/em\u003e, and \u003cem\u003eRare (\u0026le;\u0026thinsp;10%)\u003c/em\u003e as structurally distinct zones representing varying canopy continuity and edge exposure, hence serving as proxies for fragmentation intensity [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. FAD provides a compact, spatially explicit measure of local forest amount and configuration. Although complementary landscape metrics such as edge density or patch cohesion can refine fragmentation characterization, they were not computed here; instead, FAD-based zones were used as the primary structural framework for ecological comparison.\u003c/p\u003e\u003cp\u003e\u003cem\u003eFEAs\u003c/em\u003e (\u003cem\u003edegradation\u003c/em\u003e, \u003cem\u003emoisture content\u003c/em\u003e, \u003cem\u003esite type\u003c/em\u003e, \u003cem\u003estand age\u003c/em\u003e) were linked to FAD classes (Core, Transitional, Rare) for 2016, 2020, and 2024, then rasterized to the 10 m grid for pixel-wise modelling using nearest-neighbour interpolation to preserve categorical integrity:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eDegradation\u003c/b\u003e: eight ordered classes from \u003cem\u003eDegraded\u003c/em\u003e to \u003cem\u003eTransformed\u003c/em\u003e (Table S2).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eMoisture content\u003c/b\u003e: thirteen ordered classes from \u003cem\u003every wet bogs\u003c/em\u003e to \u003cem\u003efresh soils\u003c/em\u003e (Table S3).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eSite type\u003c/b\u003e: fifteen nominal categories spanning soil\u0026ndash;moisture gradients (Table S4).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eStand age\u003c/b\u003e: continuous years of growth summarized into developmental stages.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eThe combined L2A dataset was \u0026gt;\u0026thinsp;99% complete; pixels with missing values (\u0026lt;\u0026thinsp;1%) were removed. No outliers were discarded to retain ecologically meaningful extremes. Ordinal variables (\u003cem\u003edegradation\u003c/em\u003e, \u003cem\u003emoisture\u003c/em\u003e) were integer-encoded to preserve rank; \u003cem\u003esite type\u003c/em\u003e was one-hot encoded; \u003cem\u003estand age\u003c/em\u003e remained numeric. Tree-based models required no additional feature scaling [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Modelling and Interpretation\u003c/h2\u003e\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\u003ch2\u003e2.5.1 Modelling Framework\u003c/h2\u003e\u003cp\u003eWe compared two ensemble regressors\u0026mdash;Extremely Randomized Trees (ET) and LightGBM (LGBM)\u0026mdash;to predict FEAs within Core, Transitional, and Rare zones across 2016, 2020, and 2024 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Regression was preferred over classification because several ordered levels (e.g., degradation, moisture) were uneven or absent in certain zones; treating them as quasi-continuous preserved rank information and avoided unseen categories [31]. Hyperparameters were tuned using spatial k-fold cross-validation with non-overlapping geographic partitions and evaluated on held-out folds. Key ET parameters (number of trees, features per split, minimum samples to split/leaf) were optimized per FEA \u0026times; zone \u0026times; year (Supplementary Table S5). Model performance (MSE / MAE) is summarized in Supplementary Table S6; ET was selected for interpretation due to its stability and transparency [26].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\u003ch2\u003e2.5.2 Variable Importance and Partial Dependence\u003c/h2\u003e\u003cp\u003eFollowing Breiman [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], impurity-based importance (from ET) was combined with permutation importance (PI) to identify key drivers across FAD zones and years [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Impurity scores summarize split-level variance reductions but may favour high-cardinality features; PI provides a model-agnostic estimate by shuffling a feature and recording the increase in mean squared error (MSE), which better reflects out-of-sample impact. PI was repeated with multiple random permutations per feature and the resulting spread summarized.\u003c/p\u003e\u003cp\u003e\u003cb\u003ePartial Dependence Plots (PDPs)\u003c/b\u003e visualize how VIs influence model predictions while averaging over the distribution of other features [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Let \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:f\\)\u003c/span\u003e\u003c/span\u003e(\u0026sdot;) denote the trained model and split the features into a set of interest \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{\\text{S}\\text{}}\\)\u003c/span\u003e\u003c/span\u003e​ and its complement \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{c}\\)\u003c/span\u003e\u003c/span\u003eThe partial dependence of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:f\\)\u003c/span\u003e\u003c/span\u003e with respect to \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{\\text{S}\\text{}}\\)\u003c/span\u003e\u003c/span\u003e is is defined as\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:{F}_{S\\left(z\\right)}=\\:{\\mathbb{E}}_{\\left\\{{x}_{C}\\right\\}\\left[f\\left(z,\\:{x}_{C}\\right)\\right]}=\\:\\int\\:f\\left(z,\\:{x}_{C}\\right)p\\left({x}_{C}\\right)d{x}_{C}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\)\u003c/span\u003e\u003c/span\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\left({x}_{C}\\right)\\:\\)\u003c/span\u003e\u003c/span\u003eis the marginal distribution of the complementary features. In practice, we approximate this expectation using the brute-force empirical average over the observeddataset\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\:\\:\\:\\:\\:\\:\\)\u003c/span\u003e\u003c/span\u003e\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:{\\widehat{F}}_{S\\left(z\\right)}=\\:\\left(\\frac{1}{n}\\right){\\sum\\:}_{i=1}^{n}f\\left(z,\\:{x}_{C},i\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{\\left\\{C,i\\right\\}\\:}\\)\u003c/span\u003e\u003c/span\u003e ​are the observed values of the complementary features for sample \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:i\\)\u003c/span\u003e\u003c/span\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e is the number of samples used in the average.\u003c/p\u003e\u003cp\u003eBecause fragmentation effects are context-dependent, we emphasized 2D PDPs (i.e., \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\mid\\:S\\mid\\:=2)\\)\u003c/span\u003e\u003c/span\u003e to capture nonlinear responses and interactions between key VI pairs for each FEA, e.g., NDRE\u0026ndash;GARI (degradation), NDRE\u0026ndash;NDMI (moisture content), NDVI\u0026ndash;NDRE (site type), and CVI\u0026ndash;NDRE (stand age). To ensure comparability across conditions, PDPs were organized in 3\u0026times;3 grids\u0026mdash;rows corresponding to FAD classes (Rare, Transitional, Core) and columns to years (2016, 2020, 2024). PDPs were implemented using \u003cem\u003escikit-learn\u003c/em\u003e\u0026rsquo;s brute-force method, evaluating grids within observed feature ranges to avoid extrapolation.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\u003ch2\u003e2.5.3 Model Accuracy Assessment\u003c/h2\u003e\u003cp\u003eThe predictive performance of ET and LGBM models was evaluated using the Mean Squared Error (MSE) and Mean Absolute Error (MAE), computed separately for training and test datasets to assess model fit and generalization. For each FEA, predicted values \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{ŷ}_{i}\\)\u003c/span\u003e\u003c/span\u003e were compared with ground-truth values \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{y}_{i}\\)\u003c/span\u003e\u003c/span\u003e derived from the Polish Forest Data Bank, rasterized at 10 m resolution to align with Sentinel-2 L2A imagery (Section 2.3.1).\u003c/p\u003e\u003cp\u003eMSE was calculated as the average squared difference between predicted and actual values, emphasizing larger deviations:\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:MSE\\:=\\:\\left(\\frac{1}{n}\\right){\\sum\\:}_{i=1}^{n}{\\left({y}_{i}-\\:{ŷ}_{i}\\right)}^{2}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eMAE was computed as the average absolute difference, providing an interpretable metric in the original units of the ground variable:\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$\\:MAE\\:=\\:\\left(\\frac{1}{n}\\right){\\sum\\:}_{i=1}^{n}\\left|{y}_{i}-\\:{ŷ}_{i}\\right|$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e represents the number of sampled pixels per fragmentation zone (10% stratified random sample). Metrics were computed per FEA, per zone (Rare, Transitional, Core), and per year (2016, 2020, 2024) to preserve ecological context. Error distributions were further summarized with boxplots comparing predicted and observed values to evaluate model stability across zones. All metrics are reported in Supplementary Table S6.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eThe ET model, tuned via hyperparameters such as the number of trees and split thresholds (Supplementary Table S5; Section \u003cspan class=\"InternalRef\"\u003e2.5.3\u003c/span\u003e), predicted all FEAs across structural zones: Core, Transitional, and Rare zones for 2016, 2020, and 2024. Its performance was compared with LGBM on held-out test data (Supplementary Table S6; Section 2.5.5). As shown in Fig.\u0026nbsp;4, ET yielded lower or more stable errors for most FEAs\u0026mdash;particularly \u003cem\u003eDegradation\u003c/em\u003e, \u003cem\u003eSite Type\u003c/em\u003e, and \u003cem\u003eStand Age\u003c/em\u003e\u0026mdash;while LGBM performed slightly better for \u003cem\u003eMoisture Content\u003c/em\u003e in some zones but with higher variability overall. No statistically significant inter-model or inter-annual differences were observed (Supplementary Table S6), indicating comparable predictive capacity between models. ET was therefore retained for subsequent interpretability analyses owing to its higher stability and transparency.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n\u003ch2\u003e3.2. FEA Prediction Performance\u003c/h2\u003e\n\u003cp\u003eFor each FEA, we selected a representative VI pair based on PI rankings (Section \u003cspan class=\"InternalRef\"\u003e2.5.2\u003c/span\u003e; Supplementary Figures S3, S5, S7, S9). This procedure ensured interpretability and comparability across structural zones and years while retaining ecological relevance. The selected pairs were NDRE\u0026thinsp;+\u0026thinsp;GARI (Degradation), NDRE\u0026thinsp;+\u0026thinsp;NDMI (Moisture Content), NDVI\u0026thinsp;+\u0026thinsp;NDRE (Site Type), and CVI\u0026thinsp;+\u0026thinsp;NDRE (Stand Age).\u003c/p\u003e\n\u003cp\u003eRepresentative PI results for each FEA are summarized in the main text, whereas the complete PI grids and alternative VI combinations are presented in the Supplementary Figures S4, S6, S8, and S10. This organization allows readers to follow the principal findings directly in the manuscript while providing full transparency and reproducibility through detailed supplementary materials.\u003c/p\u003e\n\u003cdiv id=\"Sec15\" class=\"Section3\"\u003e\n\u003ch2\u003e3.2.1. Degradation\u003c/h2\u003e\n\u003cp\u003ePI analysis (Supplementary Figure S3) identifies NDRE as the most influential vegetation index for predicting degradation, particularly in Rare zones, while GARI and GRNDVI also rank highly in fragmented forest contexts. PDPs for the NDRE\u0026thinsp;+\u0026thinsp;GARI pair (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e) reveal consistent spectral patterns associated with increasing canopy degradation. In these surfaces, higher response values correspond to intensified canopy stress, characterized by declining red-edge reflectance (NDRE\u0026thinsp;\u0026asymp;\u0026thinsp;0.35\u0026ndash;0.45) and reduced pigment-resistant greenness (GARI\u0026thinsp;\u0026asymp;\u0026thinsp;0.60\u0026ndash;0.70).\u003c/p\u003e\n\u003cp\u003eCore zones maintain relatively stable NDRE and GARI ranges through all years, indicating limited physiological stress and structural disturbance. Transitional zones exhibit a gradual shift toward lower NDRE\u0026ndash;GARI values by 2024, reflecting chlorophyll depletion and partial canopy thinning. Rare zones display the widest spectral spread, with pronounced low-NDRE regions (\u0026lt;\u0026thinsp;0.40) and depressed GARI values, signifying pigment loss, crown opening, and exposure effects typical of fragmented or repeatedly disturbed stands.\u003c/p\u003e\n\u003cp\u003eComparable spectral responses are observed for alternative indices such as GRNDVI and NDWI (Supplementary Figure S4), which likewise capture stress-related reductions in chlorophyll content and canopy water retention under conditions of increased structural openness and edge exposure.\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section3\"\u003e\n\u003ch2\u003e3.2.2. Moisture Content\u003c/h2\u003e\n\u003cp\u003ePI rankings (Supplementary Figure S5) highlight NDMI and NDRE as dominant predictors, especially under the drier 2020 conditions when moisture stress intensified in Rare zones. PDPs for NDRE\u0026thinsp;+\u0026thinsp;NDMI (Fig.\u0026nbsp;6) show clear spectral gradients corresponding to hydrological variability. Higher response values denote greater moisture stress, linked to lower near-infrared reflectance (NDMI\u0026thinsp;\u0026asymp;\u0026thinsp;0.20\u0026ndash;0.30) and weaker red-edge activity (NDRE\u0026thinsp;\u0026asymp;\u0026thinsp;0.40\u0026ndash;0.50).\u003c/p\u003e\n\u003cp\u003eCore zones remain relatively stable, maintaining high NDRE\u0026ndash;NDMI values consistent with fresh or moist soils. Transitional zones display moderate declines in NDMI by 2024, indicating partial drying and disrupted water balance. Rare zones exhibit the sharpest decrease (\u0026lt;\u0026thinsp;0.25 NDMI; \u0026lt; 0.45 NDRE), reflecting progressive desiccation and canopy-level moisture loss caused by fragmentation and exposure.\u003c/p\u003e\n\u003cp\u003eComparable responses are observed for moisture-sensitive indices such as NDWI and CI red-edge (Supplementary Figure S6), confirming these spectral traits as reliable indicators of hydrological stress in fragmented forests.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\n\u003ch2\u003e3.2.3. Site Type\u003c/h2\u003e\n\u003cp\u003ePI results (Supplementary Figure S7) identify NDVI and NDRE as the main predictors of site-type variability, with CI red-edge gaining importance in fragmented zones. PDPs for NDVI\u0026thinsp;+\u0026thinsp;NDRE (Fig.\u0026nbsp;7) reveal distinct spectral separations among habitat types. Higher response values correspond to fertile or moist forest sites with high NDVI (\u0026gt;\u0026thinsp;0.70) and stable red-edge reflectance (NDRE\u0026thinsp;\u0026asymp;\u0026thinsp;0.45\u0026ndash;0.55), while lower responses represent bog or swamp habitats characterized by reduced chlorophyll activity.\u003c/p\u003e\n\u003cp\u003eCore zones consistently retain high NDVI\u0026ndash;NDRE combinations, indicating stable fresh or moist broadleaf and coniferous stands. Transitional zones show intermediate values (~\u0026thinsp;0.55\u0026ndash;0.65 NDVI) by 2024, suggesting mixtures of moist broadleaf and swamp forests. Rare zones display pronounced spectral contrasts (NDVI\u0026thinsp;\u0026lt;\u0026thinsp;0.55; NDRE\u0026thinsp;\u0026lt;\u0026thinsp;0.45), marking shifts toward bog coniferous and riparian floodplain habitats associated with disturbance and altered water tables. Additional VI pairings for site-type prediction are presented in Supplementary Figure S8.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\n\u003ch2\u003e3.2.4. Stand Age\u003c/h2\u003e\n\u003cp\u003ePI analysis (Supplementary Figure S9) indicates CVI as the leading predictor of stand-age variation in Core and Transitional zones, with NDRE capturing younger regrowth dynamics in Rare areas. PDPs for CVI\u0026thinsp;+\u0026thinsp;NDRE (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e) show clear spectral relationships between canopy structure and forest maturity. Higher response values denote older stands, characterized by greater chlorophyll density (CVI\u0026thinsp;\u0026asymp;\u0026thinsp;3.0\u0026ndash;4.5) and stronger red-edge reflectance (NDRE\u0026thinsp;\u0026asymp;\u0026thinsp;0.55\u0026ndash;0.65), whereas lower responses correspond to younger, regenerating canopies with diminished pigment and structural complexity.\u003c/p\u003e\n\u003cp\u003eCore zones are dominated by high-CVI, high-NDRE combinations across all years, reflecting mature, undisturbed stands. Transitional zones exhibit mixed-age structures by 2024, with intermediate spectral ranges (CVI\u0026thinsp;\u0026asymp;\u0026thinsp;2.0\u0026ndash;3.0; NDRE\u0026thinsp;\u0026asymp;\u0026thinsp;0.45\u0026ndash;0.55) consistent with selective disturbance and regrowth. Rare zones predominantly show low-CVI (\u0026lt;\u0026thinsp;2.0) and low-NDRE (\u0026lt;\u0026thinsp;0.45) responses, indicative of young secondary forests and patchy regeneration occurring in structurally open, disturbance-prone zones. Additional VI pairings for stand-age prediction are provided in Supplementary Figure S10.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Ecological Drivers and Degradation Trends Across FAD Zones\u003c/h2\u003e\u003cp\u003eAcross FAD-based structural zones, NDRE, GARI, and GRNDVI consistently captured canopy stress, reflecting their sensitivity to pigment decline and early warning signals [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Treating degradation as a continuous gradient revealed subtle transitions preceding visible structural change, aligning with trait-based remote-sensing frameworks emphasizing anticipatory indicators [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. In Core areas, PDPs were notably stable\u0026mdash;consistent with buffered microclimates and structural continuity that reduce edge exposure and maintain physiological equilibrium[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. In contrast, Transitional and Rare zones showed steeper, more variable gradients, symptomatic of greater microclimatic variability, nutrient depletion, and wind exposure near structurally open edges[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. These patterns echo global evidence that canopy pigment decline accelerates once forest continuity drops below critical thresholds of ~\u0026thinsp;20\u0026ndash;30%[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The strong and recurring influence of NDRE and GARI in Rare zones underscores their value as early indicators of canopy stress under open or discontinuous cover. Interpreted through FAD-based stratification, these responses reveal how pigment-sensitive indices vary with local canopy density and connectivity, providing a transferable approach for detecting early degradation in temperate forest mosaics.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Moisture Dynamics and Spectral Predictability\u003c/h2\u003e\u003cp\u003eMoisture responses diverged clearly by structural context. NDMI and NDRE consistently ranked as the most informative indices for soil\u0026ndash;canopy interactions under variable canopy closure [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Core zones retained stable, fresh-moist conditions\u0026mdash;consistent with dense canopies and established root systems that buffer against drought [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Rare zones, in contrast, exhibited stronger moisture contrasts and higher variability, reflecting localized drying under reduced canopy shade and altered evapotranspiration\u0026mdash;hallmarks of structurally open forest patches [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. By 2024, slight brightening in Rare-zone PDPs may signal partial hydrological recovery following thinning or disturbance, a pattern also observed elsewhere[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. These contrasts highlight the diagnostic value of FAD-based stratification for isolating hydrological responses otherwise obscured in aggregate forest metrics. Together, they demonstrate how open-data, reproducible workflows can support climate-resilience assessment across structurally heterogeneous forest landscapes.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Site Type as a Landscape Filter\u003c/h2\u003e\u003cp\u003eSite-type variation reflected underlying fertility and hydrological gradients captured by NDVI and NDRE, with CIred-edge contributing additional pigment-based contrast. Core zones exhibited homogeneous PDP responses consistent with fertile, moist broadleaf\u0026ndash;conifer conditions and the relative stability of intact interiors [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Transitional and Rare zones displayed stronger heterogeneity, reflecting spatial variability in soils, drainage, and successional stage influenced by historical disturbance [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Modeling site type on a continuous scale enabled direct comparison among zones but may smooth over fine-scale heterogeneity, especially in Rare fragments combining pioneer regrowth with mature remnants. Integrating spectral indices with LiDAR-derived structural variables or soil data could sharpen habitat delineation in such mosaics [1]. Even so, the clear spectral separation across zones shows that commonly used indices such as NDVI and NDRE remain efficient proxies for fertility and site quality in large-area inventories. This integration of spectral and structural information underscores how FAD-based zoning can serve as a practical habitat-filter framework, linking trait-based ecology with reproducible monitoring at the landscape scale.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec23\" class=\"Section2\"\u003e\u003ch2\u003e4.4 Stand Age and Structural Maturity\u003c/h2\u003e\u003cp\u003eStand-age patterns linked CVI and NDRE with canopy development, capturing pigment and structural changes during forest succession. Core and Transitional zones displayed stable PDPs characteristic of mature, uniform stands, confirming that optical indices can reliably reflect stand-age variation in low-disturbance contexts [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Rare zones, however, presented greater spectral ambiguity: regenerating edges often exhibited reflectance signatures resembling older stands, risking overestimation of maturity\u0026mdash;a known limitation in structurally heterogeneous forests [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Incorporating LiDAR or GEDI-derived metrics alongside vegetation indices would enhance age-structure interpretation and ecological realism, ensuring that optical proxies reflect both pigment and physical canopy development[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Such integration would improve the precision of maturity mapping in complex forest mosaics, bridging spectral traits and structural dynamics within open-data, reproducible frameworks.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section2\"\u003e\u003ch2\u003e4.5 From Process Detection to Management Application\u003c/h2\u003e\u003cp\u003eZone-specific analysis offers a framework for tailoring management interventions to structural context. \u003cem\u003eCore\u003c/em\u003e zones, characterized by stable spectral and ecological conditions, remain priorities for strict protection to preserve carbon stocks and interior-dependent biodiversity [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. \u003cem\u003eTransitional\u003c/em\u003e zones, which exhibited intermediate and variable responses, could benefit from adaptive strategies such as selective thinning, mixed-species enrichment, and the establishment of ecological corridors to stabilize pigment and moisture dynamics [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. \u003cem\u003eRare\u003c/em\u003e zones, displaying the strongest stress signals, align with the need for restoration or passive rewilding aimed at rebuilding canopy continuity and mitigating desiccation risk.\u003c/p\u003e\u003cp\u003eThese interpretations are consistent with broader landscape evidence showing that small, well-managed forest fragments can significantly enhance connectivity and ecosystem function [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Spectral indicators such as NDRE, NDMI, and NDVI therefore serve as efficient tools for diagnosing stress and monitoring ecological recovery across structurally heterogeneous forests. Because the workflow relies entirely on open data, similar zone-based diagnostics can be readily applied to other temperate and boreal systems, supporting regionally adapted conservation strategies grounded in transparent and reproducible evidence.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec25\" class=\"Section2\"\u003e\u003ch2\u003e4.6 Limitations and Methodological Considerations\u003c/h2\u003e\u003cp\u003eSeveral methodological considerations qualify the interpretation of these findings. Treating categorical field attributes (\u003cem\u003edegradation\u003c/em\u003e, \u003cem\u003emoisture\u003c/em\u003e, \u003cem\u003esite type\u003c/em\u003e) as continuous variables improved detection of ecological gradients but may obscure abrupt thresholds\u0026mdash;particularly in edge-dominated Rare zones, where variability in composition and microclimate is high.\u003c/p\u003e\u003cp\u003eSpatial resolution presents an additional constraint: Sentinel-2\u0026rsquo;s 10 m pixel size can underrepresent microhabitat heterogeneity, especially in species-rich or structurally diverse stands. Integrating UAV or LiDAR data would offer finer structural context and improve ecological interpretation [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Transferability beyond temperate, pine-dominated forests may also require local calibration, as spectral overlap among canopy species can reduce generality [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDespite these caveats, combining ET with PI and PDP visualization effectively linked structural forest heterogeneity to ecological gradients. The open-data and code-based design enhances transparency and reproducibility, while the framework\u0026rsquo;s interpretability supports future extensions that integrate climatic, edaphic, and LiDAR-based covariates to refine resilience forecasting under different disturbance regimes.\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study examined how structural forest heterogeneity influences key ecological processes\u0026mdash;degradation, moisture dynamics, habitat quality, and structural maturity\u0026mdash;across Core, Transitional, and Rare zones in the Tuchola Forest Biosphere Reserve using open Sentinel-2 vegetation indices. Pigment-sensitive indices such as NDRE and GARI consistently acted as early indicators of canopy stress in Rare zones, where structural openness and edge exposure precede physiological decline and reduced habitat quality [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Moisture-sensitive indices like NDMI distinguished the hydrological stability of Core zones from the variability and drying trends observed in Rare areas, reflecting the cumulative effects of disturbance and shallow rooting typical of monodominant Scots pine stands [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Site-type predictions captured fertility contrasts between closed interiors and structurally mixed mosaics, while stand-age modelling emphasized the risk of misclassifying regenerating edges as mature stands\u0026mdash;highlighting the need for integrated spectral and structural interpretation.\u003c/p\u003e\u003cp\u003eBy combining Foreground Area Density (FAD)\u0026ndash;based stratification with interpretable ensemble learning (Extra Trees, LightGBM, Partial Dependence), the analysis establishes a reproducible framework linking spectral traits to ecological processes across structural gradients [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. The findings demonstrate that open, zone-aware approaches can reveal functional variation otherwise masked in aggregate forest metrics, offering a transferable template for ecological diagnosis.\u003c/p\u003e\u003cp\u003eAlthough developed for a temperate, pine-dominated landscape in northern Poland, the workflow\u0026rsquo;s modular design\u0026mdash;built entirely on open data, transparent preprocessing, and explainable modelling\u0026mdash;makes it readily adaptable to other forest systems. Incorporating LiDAR, GEDI, or climatic variables would further refine the framework\u0026rsquo;s capacity to trace recovery trajectories and anticipate resilience loss under disturbance and drought [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eMore broadly, integrating vegetation indices with FAD-based structural zoning advances remote sensing from descriptive mapping toward diagnostic, reproducible assessment of ecological function. Such frameworks can support scalable forest monitoring aligned with global biodiversity and climate objectives by identifying where action is most needed: safeguarding Core interiors, adapting management in Transitional areas, and restoring connectivity in Rare zones. In this way, the study contributes to the broader transition toward interpretable, open-data monitoring of forest resilience and provides a practical foundation for climate-smart, structure-aware conservation planning [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch3\u003eAcknowledgements\u003c/h3\u003e\n\u003cp\u003eThe authors thank Dr. Melaine Aubry Kientz and Prof. Gr\u0026eacute;goire Vincent for valuable advice and guidance during the internship at the AMAP Laboratory, Montpellier. We also thank the Statistical Analysis Center at Nicolaus Copernicus University for access to their computational cluster and expert statistical feedback. We acknowledge the Bank Danych o Lasach \u0026ndash; Lasy Państwowe (Forest Data Bank \u0026ndash; State Forests) for providing high-quality forest inventory data collected by professional foresters.\u003c/p\u003e\n\u003ch3\u003eFunding\u003c/h3\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003ch3\u003eAuthor contributions\u003c/h3\u003e\n\u003cp\u003eS.D. conceived the study, curated data, performed analyses, developed methods and software, validated results, prepared visualizations, and wrote the manuscript.\u003c/p\u003e\n\u003cp\u003eJ.W. contributed software, statistical analysis, visualization, validation, and manuscript review and editing.\u003c/p\u003e\n\u003cp\u003eD.J-A. provided supervision and manuscript review and editing.\u003c/p\u003e\n\u003cp\u003eT.K.R. contributed to manuscript review, interpretation of results, and critical revisions.\u003c/p\u003e\n\u003cp\u003eM.K. provided resources, supervision, map visualizations, and manuscript review and editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research did not involve any experiments on humans, human data, or live vertebrates/higher invertebrates. All analyses were based on open-access remote sensing datasets and publicly available forest inventory data from the Polish State Forests. Therefore, no ethical approval or informed consent was required.\u003c/p\u003e\n\u003ch3\u003eData availability\u003c/h3\u003e\n\u003cp\u003eDerived datasets and code to reproduce all results are archived at Zenodo (DOI: \u003cstrong\u003e10.5281/zenodo.17241364\u003c/strong\u003e). The archive includes: (i) tabularized vegetation and ground indices for 2016, 2020, and 2024 across FAD-classified zones (Core, Transitional, Rare); (ii) Python code for preprocessing, FAD zoning, and model replication; and (iii) documentation (README, metadata, data dictionary) and example panels. Sentinel-2 Level-2A imagery and Dynamic World \u0026ldquo;trees\u0026rdquo; probabilities are publicly available from the Copernicus Open Access Hub and Google Earth Engine. Ground attributes were obtained from the Polish Forest Data Bank (BDL) under open-data provisions. \u003cem\u003eA formal data citation is included in the reference list.\u003c/em\u003e\u003c/p\u003e\n\u003ch3\u003eAdditional information\u003c/h3\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eUse of AI tools\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGenerative AI tools (ChatGPT by OpenAI and Grok) were used \u003cstrong\u003eonly\u003c/strong\u003efor language editing and readability improvements. No text, figures, or analyses were generated automatically. All content was reviewed and approved by the authors, who take full responsibility for the work.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlonzo, M., McFadden, J. P., Nowak, D. J. \u0026amp; Roberts, D. A. Mapping urban forest structure and function using hyperspectral imagery and lidar data. \u003cem\u003eUrban Urban Green.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 135\u0026ndash;147. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ufug.2016.04.003\u003c/span\u003e\u003cspan address=\"10.1016/j.ufug.2016.04.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eArroyo-Rodr\u0026iacute;guez, V., Saldana-Vazquez, R. A., Fahrig, L. \u0026amp; Santos, B. A. Does forest fragmentation cause an increase in forest temperature? \u003cem\u003eEcol. Res.\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, 81\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11284-016-1411-6\u003c/span\u003e\u003cspan address=\"10.1007/s11284-016-1411-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBauer, L. P. L., Huth, A., Bogdanowski, A., M\u0026uuml;ller, M. \u0026amp; Fischer, R. Edge effects in Amazon forests: integrating remote sensing and modelling to assess changes in biomass and productivity. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e16\u003c/b\u003e, 501. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs16030501\u003c/span\u003e\u003cspan address=\"10.3390/rs16030501\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBlake, J. G. \u0026amp; Karr, J. R. Species composition of bird communities and the conservation benefit of large versus small forests. \u003cem\u003eBiol. Conserv.\u003c/em\u003e \u003cb\u003e30\u003c/b\u003e, 173\u0026ndash;187. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/0006-3207(84)90065-X\u003c/span\u003e\u003cspan address=\"10.1016/0006-3207(84)90065-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1984).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBreiman, L. Random forests. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 5\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/A:1010933404324\u003c/span\u003e\u003cspan address=\"10.1023/A:1010933404324\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2001).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBriant, G., Gond, V. \u0026amp; Laurance, S. G. Habitat fragmentation and the desiccation of forest canopies: eastern Amazonia. \u003cem\u003eBiol. Conserv.\u003c/em\u003e \u003cb\u003e143\u003c/b\u003e, 2763\u0026ndash;2769. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biocon.2010.07.024\u003c/span\u003e\u003cspan address=\"10.1016/j.biocon.2010.07.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBritton, J., Graham, G. \u0026amp; Woloszyn, M. Exploring relationships between drought indices and ecological drought impacts using machine learning and explainable AI. \u003cem\u003eJ. Appl. Serv. Climatol\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, e009001. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.46275/JOASC.2024.09.001\u003c/span\u003e\u003cspan address=\"10.46275/JOASC.2024.09.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrown, C. F. et al. Dynamic World, near real-time global 10 m land use land cover mapping. \u003cem\u003eSci. Data\u003c/em\u003e. \u003cb\u003e9\u003c/b\u003e, 251. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41597-022-01307-4\u003c/span\u003e\u003cspan address=\"10.1038/s41597-022-01307-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBudniak, P. \u0026amp; Zięba, S. Effects of forest fragmentation on the volume of wood resources in managed pine-dominated forests in Poland. \u003cem\u003eForests\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 590. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/f13040590\u003c/span\u003e\u003cspan address=\"10.3390/f13040590\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBurns, P., Hakkenberg, C. R. \u0026amp; Goetz, S. J. Multi-resolution gridded maps of vegetation structure from GEDI. \u003cem\u003eSci. Data\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e, 881. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41597-024-03668-4\u003c/span\u003e\u003cspan address=\"10.1038/s41597-024-03668-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDobor, L. et al. Post-disturbance recovery of forest carbon in a temperate forest landscape under climate change. \u003cem\u003eAgric. Meteorol.\u003c/em\u003e \u003cb\u003e263\u003c/b\u003e, 308\u0026ndash;322. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.agrformet.2018.08.028\u003c/span\u003e\u003cspan address=\"10.1016/j.agrformet.2018.08.028\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDutt, S., Batar, A. K., Sulik, S. \u0026amp; Kunz, M. Forest ecosystem on the edge: fragmentation susceptibility in Tuchola Forest, Poland. \u003cem\u003eEcol. Indic.\u003c/em\u003e \u003cb\u003e161\u003c/b\u003e, 111980. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecolind.2024.111980\u003c/span\u003e\u003cspan address=\"10.1016/j.ecolind.2024.111980\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFahrig, L. Effects of habitat fragmentation on biodiversity. \u003cem\u003eAnnu. Rev. Ecol. Evol. Syst.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, 487\u0026ndash;515. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/annurev.ecolsys.34.011802.1324199\u003c/span\u003e\u003cspan address=\"10.1146/annurev.ecolsys.34.011802.1324199\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2003).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFahrig, L. Ecological responses to habitat fragmentation \u003cem\u003eper se\u003c/em\u003e. \u003cem\u003eAnnu. Rev. Ecol. Evol. Syst.\u003c/em\u003e \u003cb\u003e48\u003c/b\u003e, 1\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/annurev-ecolsys-110316-022612\u003c/span\u003e\u003cspan address=\"10.1146/annurev-ecolsys-110316-022612\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFahrig, L. et al. Is habitat fragmentation bad for biodiversity? \u003cem\u003eBiol. Conserv.\u003c/em\u003e \u003cb\u003e230\u003c/b\u003e, 179\u0026ndash;186. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biocon.2018.12.026\u003c/span\u003e\u003cspan address=\"10.1016/j.biocon.2018.12.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFletcher, R. J. et al. Is habitat fragmentation good for biodiversity? \u003cem\u003eBiol. Conserv.\u003c/em\u003e \u003cb\u003e226\u003c/b\u003e, 9\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biocon.2018.07.022\u003c/span\u003e\u003cspan address=\"10.1016/j.biocon.2018.07.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGeurts, P., Ernst, D. \u0026amp; Wehenkel, L. Extremely randomized trees. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e63\u003c/b\u003e, 3\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10994-006-6226-1\u003c/span\u003e\u003cspan address=\"10.1007/s10994-006-6226-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2006).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGonz\u0026aacute;lez-\u0026Aacute;vila, S., Ortega, E. \u0026amp; Mart\u0026iacute;n, B. Potential priority areas for forest-dwelling species in Spain based on the degree of forest fragmentation. \u003cem\u003eJ. Maps\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 2223597. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/17445647.2023.2223597\u003c/span\u003e\u003cspan address=\"10.1080/17445647.2023.2223597\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGr\u0026ouml;mping, U. Variable importance assessment in regression: linear regression versus random forest. \u003cem\u003eAm. Stat.\u003c/em\u003e \u003cb\u003e63\u003c/b\u003e, 308\u0026ndash;319. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1198/tast.2009.08199\u003c/span\u003e\u003cspan address=\"10.1198/tast.2009.08199\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHaddad, N. M. et al. Habitat fragmentation and its lasting impact on Earth\u0026rsquo;s ecosystems. \u003cem\u003eSci. Adv.\u003c/em\u003e 1, e1500052 (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1126/sciadv.1500052\u003c/span\u003e\u003cspan address=\"10.1126/sciadv.1500052\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe, L. et al. Temporal dynamics of soil moisture in a northern temperate mixed successional forest after a prescribed intermediate disturbance. \u003cem\u003eAgric. Meteorol.\u003c/em\u003e \u003cb\u003e180\u003c/b\u003e, 22\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.agrformet.2013.04.014\u003c/span\u003e\u003cspan address=\"10.1016/j.agrformet.2013.04.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKe, G. et al. LightGBM: a highly efficient gradient boosting decision tree. \u003cem\u003eAdv. Neural Inf. Process. Syst.\u003c/em\u003e \u003cb\u003e30\u003c/b\u003e, 3149\u0026ndash;3157. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5555/3294996.3295074\u003c/span\u003e\u003cspan address=\"10.5555/3294996.3295074\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dl.acm.org/doi/\u003c/span\u003e\u003cspan address=\"https://dl.acm.org/doi/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKozak, J. et al. Forest-cover increase does not trigger forest-fragmentation decrease: Polish Carpathians. \u003cem\u003eSustainability\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 1472. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su10051472\u003c/span\u003e\u003cspan address=\"10.3390/su10051472\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLausch, A., Erasmi, S., King, D. J., Magdon, P. \u0026amp; Heurich, M. Understanding forest health with remote sensing\u0026mdash;Part I: spectral traits, processes and characteristics. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 1029. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs8121029\u003c/span\u003e\u003cspan address=\"10.3390/rs8121029\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLausch, A., Erasmi, S., King, D. J., Magdon, P. \u0026amp; Heurich, M. Understanding forest health with remote sensing\u0026mdash;Part II: approaches and data models. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 129. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs9020129\u003c/span\u003e\u003cspan address=\"10.3390/rs9020129\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eŁoś, H. et al. Evaluation of XGBoost and LGBM performance in tree species classification with Sentinel-2 data. In \u003cem\u003eIGARSS 2021 \u0026ndash; IEEE International Geoscience and Remote Sensing Symposium\u003c/em\u003e 5803\u0026ndash;5806 (IEEE, 2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/IGARSS47720.2021.9553031\u003c/span\u003e\u003cspan address=\"10.1109/IGARSS47720.2021.9553031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eŁuc\u0026oacute;w, D. et al. Pine forest management and disturbance in northern Poland: combining high-resolution paleoecology and remote sensing. \u003cem\u003eFront. Ecol. Evol.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 747976. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fevo.2021.747976\u003c/span\u003e\u003cspan address=\"10.3389/fevo.2021.747976\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMolnar, C. \u003cem\u003eInterpretable Machine Learning: A Guide for Making Black Box Models Explainable.\u003c/em\u003eLeanpub, (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://christophm.github.io/interpretable-ml-book\u003c/span\u003e\u003cspan address=\"https://christophm.github.io/interpretable-ml-book\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNguyen, H. T., Nguyen, T. D. \u0026amp; Kappas, M. Land cover and forest type classification by values of vegetation indices and forest structure of tropical lowland forests in central Vietnam. \u003cem\u003eInt. J. For. Res.\u003c/em\u003e 8896310 (2020). (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2020/8896310\u003c/span\u003e\u003cspan address=\"10.1155/2020/8896310\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNicodemus, K. K., Malley, J. D., Strobl, C. \u0026amp; Ziegler, A. The behaviour of random forest permutation-based variable importance measures under predictor correlation. \u003cem\u003eBMC Bioinform\u003c/em\u003e. \u003cb\u003e11\u003c/b\u003e, 110. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/1471-2105-11-110\u003c/span\u003e\u003cspan address=\"10.1186/1471-2105-11-110\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePedregosa, F. et al. Scikit-learn: machine learning in Python. \u003cem\u003eJ. Mach. Learn. Res.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 2825\u0026ndash;2830 (2011). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://jmlr.org/beta/papers/v12/pedregosa11a.html\u003c/span\u003e\u003cspan address=\"https://jmlr.org/beta/papers/v12/pedregosa11a.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePeters, J. et al. Random forests as a tool for ecohydrological distribution modelling. \u003cem\u003eEcol. Model.\u003c/em\u003e \u003cb\u003e207\u003c/b\u003e, 304\u0026ndash;318. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecolmodel.2007.05.011\u003c/span\u003e\u003cspan address=\"10.1016/j.ecolmodel.2007.05.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2007).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRiitters, K. \u0026amp; Wickham, J. Decline of forest interior conditions in the conterminous United States. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 653. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/srep00653\u003c/span\u003e\u003cspan address=\"10.1038/srep00653\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2012).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRossini, M., Panigada, C., Meroni, M. \u0026amp; Colombo, R. Assessment of oak forest condition based on leaf biochemical variables and chlorophyll fluorescence. \u003cem\u003eTree Physiol.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 1487\u0026ndash;1496. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/treephys/26.11.1487\u003c/span\u003e\u003cspan address=\"10.1093/treephys/26.11.1487\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2006).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchwartz, N. B., Budsock, A. M. \u0026amp; Uriarte, M. Fragmentation, forest structure, and topography modulate impacts of drought in a tropical forest landscape. \u003cem\u003eEcology\u003c/em\u003e \u003cb\u003e100\u003c/b\u003e, e02677. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/ecy.2677\u003c/span\u003e\u003cspan address=\"10.1002/ecy.2677\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVogt, P. \u0026amp; Riitters, K. GuidosToolbox: universal digital image object analysis. \u003cem\u003eEur. J. Remote Sens.\u003c/em\u003e \u003cb\u003e50\u003c/b\u003e, 352\u0026ndash;361. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/22797254.2017.1330650\u003c/span\u003e\u003cspan address=\"10.1080/22797254.2017.1330650\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, K., Franklin, S. E., Guo, X. \u0026amp; Cattet, M. Remote sensing of ecology, biodiversity and conservation: a review from the perspective of remote-sensing specialists. \u003cem\u003eSensors\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 9647\u0026ndash;9667. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s101109647\u003c/span\u003e\u003cspan address=\"10.3390/s101109647\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, Z., Han, L., Wang, L., Shi, H. \u0026amp; Luo, Y. Neighboring patch density or patch size? Which determines the importance of forest patches in maintaining overall landscape connectivity in Kanas, Xinjiang, China. \u003cem\u003eBiology\u003c/em\u003e \u003cb\u003e14\u003c/b\u003e, 881. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/biology14070881\u003c/span\u003e\u003cspan address=\"10.3390/biology14070881\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWei, X. et al. Forest harvesting and hydrology in boreal forests: under an increased and cumulative disturbance context. \u003cem\u003eEcol. Manage.\u003c/em\u003e \u003cb\u003e522\u003c/b\u003e, 120468. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2022.120468\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2022.120468\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWulder, M. A., White, J. C., Andrew, M. E., Seitz, N. E. \u0026amp; Coops, N. C. Forest fragmentation, structure, and age characteristics as a legacy of forest management. \u003cem\u003eEcol. Manage.\u003c/em\u003e \u003cb\u003e258\u003c/b\u003e, 1938\u0026ndash;1949. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2009.07.041\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2009.07.041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXue, J. \u0026amp; Su, B. Significant remote sensing vegetation indices: a review of developments and applications. \u003cem\u003eJ. Sensors\u003c/em\u003e 1353691 (2017). (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2017/1353691\u003c/span\u003e\u003cspan address=\"10.1155/2017/1353691\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYu, X. et al. Spatiotemporal variation and predictability of vegetation coverage in the Beijing\u0026ndash;Tianjin\u0026ndash;Hebei metropolitan region, China. \u003cem\u003eTheor. Appl. Climatol\u003c/em\u003e. \u003cb\u003e145\u003c/b\u003e, 47\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00704-021-03616-x\u003c/span\u003e\u003cspan address=\"10.1007/s00704-021-03616-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDutt, S. \u0026amp; Wojtasik, J. Tabularized datasets of Sentinel-2 vegetation indices (VIs) and ground indices (GIs) for the Tuchola Forest (2016, 2020, 2024). \u003cem\u003eZenodo\u003c/em\u003e. (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5281/zenodo.17241364\u003c/span\u003e\u003cspan address=\"10.5281/zenodo.17241364\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"structural heterogeneity, ecological processes, Sentinel-2, Foreground Area Density (FAD), interpretable machine learning","lastPublishedDoi":"10.21203/rs.3.rs-7923328/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7923328/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eStructural heterogeneity within forests strongly influences ecological function, yet zone-specific diagnostics remain limited. This study presents a reproducible, zone-aware framework that links forest configuration to ecological processes using open remote-sensing and inventory data. In the Tuchola Forest Biosphere Reserve (Poland), we assessed degradation, moisture stress, habitat quality, and structural maturity across Foreground Area Density (FAD)\u0026ndash;based zones\u0026mdash;Core (\u0026ge;\u0026thinsp;90%), Transitional (40\u0026ndash;60%), and Rare (\u0026le;\u0026thinsp;10%)\u0026mdash;by integrating multi-temporal Sentinel-2 imagery (2016, 2020, 2024) with field-observed ecological attributes. Interpretable ensemble models (Extra Trees, LightGBM) and partial-dependence analyses revealed consistent contrasts: Rare zones exhibited early-stage canopy stress linked to structural openness and edge exposure, while Core interiors maintained stable moisture regimes and mature canopy structure. Site-type and stand-age patterns showed that spectral similarity at open edges can mimic maturity, underscoring the importance of combining spectral and structural information in future monitoring. Validated against field observations, the workflow offers a spatially explicit, transferable benchmark for diagnosing ecological variability from open data. The results support zone-specific management strategies\u0026mdash;preserving Core interiors, stabilizing Transitional areas through adaptive corridors, and restoring connectivity in Rare zones.\u003c/p\u003e","manuscriptTitle":"Linking Structural Forest Heterogeneity and Ecological Processes Using Sentinel-2 and FAD-Based Zoning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-18 09:16:50","doi":"10.21203/rs.3.rs-7923328/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-07T20:18:31+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-05T11:58:56+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"291405571038608182190450660544101081816","date":"2026-05-03T05:07:52+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-02T05:43:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"45876248583474464365051071935012366044","date":"2026-05-01T13:29:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"160537341468167476670174339511638561622","date":"2026-05-01T08:31:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"139278117608760180864784691334571950777","date":"2026-04-29T08:40:36+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"195451823385128360269928013079656068776","date":"2026-04-24T09:13:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-31T11:47:00+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"111087260542636014611122184408900214912","date":"2026-01-21T07:52:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-07T11:09:56+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-11-03T13:26:02+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-25T09:36:14+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-25T09:34:28+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-10-22T11:55:58+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0ee0e037-550d-4d07-8e47-1e9a21603d36","owner":[],"postedDate":"November 18th, 2025","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-07T20:18:31+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[{"id":58106256,"name":"Biological sciences/Ecology"},{"id":58106257,"name":"Earth and environmental sciences/Ecology"},{"id":58106258,"name":"Earth and environmental sciences/Environmental sciences"}],"tags":[],"updatedAt":"2026-05-07T20:23:58+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-18 09:16:50","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7923328","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7923328","identity":"rs-7923328","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}