Remote sensing based forest canopy density mapping in the Białowieża Forest reserve

Here is a paradox: Europe’s last primeval lowland forest—protected for centuries as a royal hunting ground and later as a nature reserve—is losing canopy density at its margins even as its core remains among the most structurally intact forests on the continent. This research mapped forest canopy density (FCD) across the Białowieża Forest Reserve using Landsat time series data (2000-2020) and validated the results against field measurements at 86 ground-truth plots. The Forest Canopy Density Model, which combines four spectral indices (Advanced Vegetation Index, Bare Soil Index, Shadow Index, and thermal index), was applied to five image dates at four-year intervals. Results showed that the core reserve maintained FCD above 0.79 throughout the period, while buffer zones declined from 0.74 to 0.58 and transition zones from 0.66 to 0.38. Overall, 22.3% of the mapped area shifted to a lower density class between 2000 and 2020. Dense forest (FCD > 0.7) occupied 31.4% of the total area in 2020, down from 42.8% in 2000. Canopy loss was concentrated along the reserve’s eastern and southern boundaries, correlating with proximity to settlements (r = -0.71, p< 0.001) and logging concession areas. Classification accuracy reached 86.7% overall (kappa = 0.83) against field measurements. These findings confirm that the Białowieża’s core is resilient but its buffer and transition zones are degrading at rates requiring immediate management attention.

Illustrations explain the paper; charts are replotted from the paper's own table values.

AI-generated illustration

↗ Enlarge

Zonal Degradation ConceptClose ×

AI-generated conceptual illustration (Białowieża Forest canopy). It explains the idea and does not show study data.

Open full-size image ↗

Replotted from source data

↗ Enlarge

FCD 2000 and FCD 2008 and more: Core Reserve vs Buffer North vs Buffer South vs Transition E vs Transition WClose ×

Replotted from Table 1 (page 2) of the original paper. Every bar is the paper's own table value.

Open full-size image ↗

Replotted from source data

↗ Enlarge

2000 (%) and 2010 (%) and more: Dense (>0.7) vs Moderate (0.5-0.7) vs Sparse (0.3-0.5) vs Open (0.1-0.3) vs Non-forest (<0.1)Close ×

Replotted from Table 2 (page 2) of the original paper. Every bar is the paper's own table value.

Open full-size image ↗

Introduction

The paradox of Białowieża is that it is simultaneously one of Europe's most protected and most contested forests. Straddling the Poland-Belarus border, its 1,500 km² of lowland temperate forest includes stands that have never been commercially logged, hosting tree species assemblages and deadwood volumes found nowhere else on the continent [1,8] . Yet the forest's periphery faces pressures from selective logging, bark beetle management interventions, and encroaching agricultural land [2] . Canopy density-the proportion of ground area covered by the vertical projection of tree crowns-serves as a proxy for forest structural integrity, carbon storage capacity, and habitat quality [3,4] . Remote sensing offers the only practical means of mapping canopy density across large forested areas at regular intervals, and the Landsat archive provides a 20+ year baseline for change detection [5,10] . Earlier mapping of Białowieża used single-date imagery or focused on species composition rather than canopy structure [8] . No multi-temporal canopy density analysis covering the full reserve and its buffer zones had been published. This research filled that gap using the Forest Canopy Density Model applied to five Landsat dates between 2000 and 2020, validated against 86 field plots.

Model Description and Parameterisation

The Forest Canopy Density (FCD) Model developed by Rikimaru and colleagues [3] estimates canopy density from four biophysical indices derived from multispectral satellite data. The Advanced Vegetation Index (AVI) captures green biomass using red and nearinfrared bands. The Bare Soil Index (BSI) detects exposed ground using shortwave infrared and visible bands. The Shadow Index (SI) measures canopy shadow proportion from visible bands. The thermal index (TI) uses land surface temperature as an inverse indicator of canopy closure. These four indices are combined through principal component analysis and scaled to a 0-1 canopy density value. The model was calibrated for Białowieża conditions using 42 training plots with hemispherical photography-derived canopy closure values as the reference standard.

Material and Methods

Material and Methods

Material

Material

Methods

Each Landsat scene was atmospherically corrected using the FLAASH module in ENVI 5.6. The four spectral indices were computed following Rikimaru and colleagues [3] . PCA integration generated FCD maps at 30 m resolution for each date. FCD values were classified into five density classes: dense (>0.7), moderate (0.5-0.7), sparse (0.3-0.5), open (0.1-0.3), and non-forest (<0.1). Change detection compared sequential FCD maps to identify areas of density gain or loss. Accuracy assessment used the 86 field plots, with 42 used for training and 44 for independent validation. Correlation analysis tested relationships between FCD change and distance to settlements, roads, and logging concessions in ArcGIS 10.8.

Results

Results

Comprehensive Interpretation

The core reserve maintained FCD above 0.79 throughout the period, but buffer and transition zones declined sharply (Table 1, Figure 1). Dense forest shrank from 42.8% to 31.4% of total area, while sparse and open classes expanded (Table 2, Figure 2). The transition zones lost the most density (Δ = -0.24 to -0.25), with over 40% of their area shifting to a lower class. Distance to settlements correlated negatively with canopy retention (r = -0.71, p< 0.001).

↗ Enlarge

Figure 2Close ×

Distribution of canopy density classes across the Białowieża Reserve (2020)

Open full-size image ↗

Original paper figureSource PDF · page 2 ↗

↗ Enlarge

Figure 1Close ×

Forest canopy density values by zone and year (2000-2020)

Open full-size image ↗

Original paper figureSource PDF · page 2 ↗

Density Class	2000 (%)	2010 (%)	2020 (%)	Change
Dense (>0.7)	42.8	37.1	31.4	-11.4
Moderate (0.5-0.7)	28.7	29.4	28.7	0.0
Sparse (0.3-0.5)	14.6	17.8	19.8	+5.2
Open (0.1-0.3)	8.4	10.1	12.6	+4.2
Non-forest (<0.1)	5.5	5.6	7.5	+2.0

Table text extracted from the original PDF; superscript letters are shown inline. The original table image is below.

View original table image

↗ Enlarge

Table 2Close ×

Canopy density class distribution (% of total area)

Open full-size image ↗

Original paper tableSource PDF · page 2 ↗

Zone	FCD 2000	FCD 2008	FCD 2020	Change	Δ Class (%)
Core Reserve	0.87	0.84	0.79	-0.08	4.2
Buffer North	0.74	0.67	0.58	-0.16	28.7
Buffer South	0.78	0.73	0.64	-0.14	23.4
Transition E	0.62	0.51	0.38	-0.24	41.8
Transition W	0.66	0.55	0.41	-0.25	38.6

Table text extracted from the original PDF; superscript letters are shown inline. The original table image is below.

View original table image

↗ Enlarge

Table 1Close ×

Forest canopy density by zone (2000-2020)

Open full-size image ↗

Original paper tableSource PDF · page 2 ↗

Overall classification accuracy was 86.7% (kappa = 0.83).

Discussion

The core reserve's resilience (FCD still 0.79 in 2020) confirms that strict protection maintains structural integrity even as surrounding areas degrade [8] . But the 0.08-point decline even in the core suggests edge effects or climatedriven stress-possibly linked to the 2015-2018 bark beetle outbreak that affected spruce stands across the region [1,6] . The buffer zone losses of 0.14-0.16 FCD units over 20 years are alarming because these zones are meant to shield the core from external pressures. If degradation continues at this rate, buffer zone FCD will fall below 0.5 within a decade, effectively eliminating the buffering function [9,11] . The transition zone is already below 0.4 in most areas, functioning more as open woodland than closed forest.

Sensitivity Analysis

FCD model sensitivity was tested by varying the PCA weighting of the four input indices. Removing the thermal index reduced overall accuracy from 86.7% to 79.3%, confirming its contribution to discriminating canopy closure levels. Substituting Sentinel-2 data (10 m resolution) for the 2020 date improved spatial detail but did not change density class assignments at the zone level, suggesting that Landsat's 30 m resolution is adequate for monitoring at this scale [12,13] . Seasonal timing sensitivity was tested using September imagery for 2016, which produced FCD values 0.03-0.06 lower than July imagery due to early leaf senescence in deciduous species.

Conclusion

This research confirmed that Białowieża's core reserve retains high canopy density, but buffer and transition zones are degrading at rates that threaten the reserve's ecological integrity. Dense forest declined from 42.8% to 31.4% of total area between 2000 and 2020. Management efforts should focus on halting canopy loss in the buffer zones through stricter logging controls and natural regeneration support, while the FCD model provides a cost-effective monitoring tool for tracking future changes.

Acknowledgement

Acknowledgements Funding SourcesSupported by institutional research allocations for forest geography and remote sensing. Institutional SupportSchool of Geography and Environment, Pune, provided image processing facilities. Contributions Not Qualifying for AuthorshipThanks to field assistants who conducted ground-truth surveys in the reserve.

1. Franklin JF, Spies TA, Van Pelt R, Carey AB. Disturbances and structural development of natural forest ecosystems. Forest Ecology and Management. 2002;155(1-3):399-423.
2. Hansen MC, Potapov PV, Moore R, Hancher M. Highresolution global maps of 21st-century forest cover change. Science. 2013;342(6160):850-853.
3. Rikimaru A, Roy PS, Miyatake S. Tropical forest cover density mapping. Tropical Ecology. 2002;43(1):39-47.
4. Roy PS, Ravan SA. Biomass estimation using satellite remote sensing data. Journal of Biosciences. 1996;21(4):535-561.
5. Townshend JRG, Justice CO, Skole D, Malingreau JP. The 1 km resolution global data set. International Journal of Remote Sensing. 1994;15(17):3417-3441.
6. Lu D, Li G, Moran E. Current situation and needs of change detection techniques. International Journal of Image and Data Fusion. 2014;5(1):13-38.
7. Gao BC. NDWI: a normalized difference water index for remote sensing of vegetation liquid water. Remote Sensing of Environment. 1996;58(3):257-266.
8. Jaroszewicz B, Cholewinska O, Gutowski JM, Samojlik T. Natural processes and anthropogenic factors jointly influence primeval forest structure. Forest Ecology and Management. 2019;449:117467.
9. Congalton RG, Green K. Assessing the accuracy of remotely sensed data, second edition. CRC Press. 2009;1-183.
10. Wulder MA, Masek JG, Cohen WB, Loveland TR. Opening the archive: how free data has enabled Landsat science. Remote Sensing of Environment. 2012;122:210.
11. Foody GM. Status of land cover classification accuracy assessment. Remote Sensing of Environment. 2002;80(1):185-201.
12. DeFries RS, Houghton RA, Hansen MC, Field CB. Carbon emissions from tropical deforestation and regrowth. Proceedings of the National Academy of Sciences. 2002;99(22):14256-14261.
13. Kerr JT, Ostrovsky M. From space to species: ecological applications for remote sensing. Trends in Ecology and Evolution. 2003;18(6):299-305.
14. Asner GP, Keller M, Pereira R, Zweede JC. Canopy damage and recovery after selective logging in Amazonia. Forest Ecology and Management. 2004;191(1-3):97-108.
15. Lefsky MA, Cohen WB, Parker GG, Harding DJ. Lidar remote sensing for ecosystem studies. BioScience. 2002;52(1):19-30.
16. Goetz SJ, Baccini A, Laporte NT, Johns T. Mapping and monitoring carbon stocks with satellite observations. Carbon Balance and Management. 2009;4:2.
17. Turner W, Spector S, Gardiner N, Fladeland M. Remote sensing for biodiversity science and conservation. Trends in Ecology and Evolution. 2003;18(6):306-314.
18. Pettorelli N, Vik JO, Mysterud A, Gaillard JM. Using satellite-derived NDVI to assess ecological responses to environmental change. Trends in Ecology and Evolution. 2005;20(9):503-510.
19. Myneni RB, Hall FG, Sellers PJ, Marshak AL. The interpretation of spectral vegetation indexes. IEEE Transactions on Geoscience and Remote Sensing. 1995;33(2):481-486.