Mapping of temperate upland habitats using high-resolution satellite imagery and machine learning

Area of Science:

• Ecology
• Remote Sensing
• Geospatial Analysis

Background:

• Upland habitats offer crucial ecological services but face significant threats from natural and anthropogenic stressors.
• Effective conservation requires accurate monitoring of upland habitat distribution and condition.
• Traditional remote sensing methods struggle with the fine spatial scales and heterogeneous, mosaic nature of upland habitats.

Purpose of the Study:

• To assess the impact of varying spatial resolutions on mapping accuracy for Irish upland habitats.
• To develop and evaluate a complementary remote sensing approach for mapping complex upland habitat mosaics.
• To utilize high-resolution Pleiades satellite imagery and Random Forest (RF) machine learning for habitat classification.

Main Methods:

• Employed high-resolution (2m) Pleiades satellite imagery for habitat mapping at two Irish upland sites.
• Utilized Random Forest (RF) machine learning algorithms for habitat classification.
• Investigated the influence of different spatial resolutions on classification accuracy and employed fuzzy classification for mosaic representation.

Main Results:

• Classification accuracy generally improved with finer spatial resolution data.
• The highest accuracy (80.34% and 79.64%) was achieved using the 2m resolution Pleiades imagery.
• RF-based fuzzy classification effectively represented complex habitat mosaics and gradual transitions.

Conclusions:

• High-resolution satellite imagery combined with RF machine learning significantly enhances the accuracy of upland habitat mapping.
• The proposed fuzzy classification approach provides a more realistic representation of complex upland ecosystems than traditional methods.
• This methodology offers a valuable tool for improving the spatiotemporal understanding and conservation of large-scale upland habitats.