Upland vegetation mapping using Random Forests with optical and radar satellite data.

Area of Science:

• Ecology
• Remote Sensing
• Geospatial Analysis

Background:

• Uplands provide essential ecosystem services but face management pressures from diverse stakeholders.
• Accurate spatial mapping of upland vegetation is crucial for effective conservation, management, and impact assessment of environmental changes.
• Existing mapping methods may not fully capture the complexity of upland vegetation distribution.

Purpose of the Study:

• To evaluate the efficacy of medium spatial resolution optical and radar satellite data for mapping upland vegetation.
• To assess the utility of ancillary soil and topographic data in improving vegetation classification accuracy.
• To test the Random Forests (RF) algorithm for identifying and mapping diverse upland vegetation types.

Main Methods:

• Utilized medium spatial resolution optical and radar satellite data.
• Integrated ancillary soil and topographic datasets.
• Employed the Random Forests (RF) algorithm for classification and mapping.
• Calibrated and validated models using intensive field survey data from three Irish study sites.

Main Results:

• Overall classification accuracy ranged from 59.8% to 94.3% across study sites.
• Inclusion of soil and elevation data enhanced classification accuracy by 5% to 27%.
• The RF approach demonstrated consistent performance across different environmental contexts.

Conclusions:

• Medium resolution satellite data, combined with RF, is effective for mapping upland vegetation.
• Ancillary soil and topographic data significantly improve mapping accuracy, benefiting conservation and management efforts.
• The validated methodology is applicable across diverse upland environments.