The analysis assessed imagery from March–May, representing the pre-monsoon season, and October–December, representing the post-monsoon season, for two years. Its machine-learning framework used spectral indices derived from Sentinel-2 data, including the Normalized Difference Vegetation Index and Normalized Difference Moisture Index, alongside elevation, slope and gray-level co-occurrence matrix texture features. The same Random Forest procedure was applied independently to all four season-year datasets, with reference samples balanced across NDVI-informed canopy-density strata.

Pre-monsoon comparisons showed a pronounced redistribution among the mapped classes. From 2016 to 2024, the areas assigned to non-forest and open canopy decreased by 39.62% and 7.15%, respectively. At the same time, mapped moderate and dense canopy increased by 231.64% and 240.43%. The post-monsoon pattern differed: moderate canopy increased by 26.01%, while dense canopy declined by 9.46%. These results describe changes in classes produced by the classification workflow, rather than direct measurements of biological canopy growth or loss.

Transition analysis within a common comparison area of 4,533.54 hectares also revealed substantial persistence in canopy-density classes alongside movement between neighboring categories. The researchers reported particularly notable redistribution from open to moderate canopy and from dense to moderate canopy. Spatial changes occurred in riparian, interior and peripheral portions of the sanctuary, indicating that mapped canopy differences were not confined to a single landscape zone. Overall, the study demonstrates how repeated satellite observations and machine-learning classification can provide a replicable means of comparing seasonal and interannual patterns across a protected landscape.

The authors cautioned that the two years were represented by different Sentinel-2 product levels: 2016 imagery came from Level-1C top-of-atmosphere data, whereas 2024 imagery came from Level-2A surface-reflectance products. The interannual differences should therefore be interpreted as changes in mapped canopy-density classes, not as precise biophysical changes in canopy structure. Despite this limitation, the framework offers a way to examine where seasonal and longer-term canopy redistribution appears most pronounced and to support spatially targeted monitoring and conservation planning in Orchha Wildlife Sanctuary.

Subject of Research: Remote-sensing assessment of seasonal and interannual forest canopy-density changes in Orchha Wildlife Sanctuary, India

Article Title: Monitoring seasonal and long-term forest canopy dynamics using Sentinel-2 imagery and Random Forest on the Google Earth Engine platform

Article References: Kumar, P., Kumar, R., Sahu, S., Dobriyal, M., Lavania, P., Srivastav, M., Pandey, P. C., & Turyasingura, B. (2026). Monitoring seasonal and long-term forest canopy dynamics using Sentinel-2 imagery and Random Forest on the Google Earth Engine platform. PLOS One, 21(10), e0361024. https://doi.org/10.1371/journal.pone.0361024

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0361024

Keywords: Sentinel-2, Google Earth Engine, Random Forest, forest canopy, NDVI, NDMI, remote sensing, seasonal monitoring, wildlife sanctuary, machine learning, Monitoring, seasonal