Enhancing vegetation formation classification: Integrating coarse-scale traditional mapping knowledge and advanced...

Area of Science:

• Ecology
• Environmental Science
• Remote Sensing

Background:

• Accurate mapping of vegetation formation types is essential for ecological and environmental research.
• Distinguishing similar vegetation types is challenging with current methods, especially with limited training data.
• Existing methods often rely on environmental variables and remote sensing spectral data.

Purpose of the Study:

• To develop an improved predictive vegetation mapping method.
• To enhance the accuracy of vegetation formation type classification.
• To reduce the dependency on extensive field samples for training.

Main Methods:

• Proposed a vegetation mapping method integrating machine learning and existing vegetation map knowledge (VMK).
• Utilized the random forest algorithm, incorporating an early vegetation map as an auxiliary feature (VMF).
• Incorporated knowledge of vegetation spatial distributions and hierarchies to refine classifications and correct unreasonable types.

Main Results:

• The VMK method achieved an overall accuracy of 67.7%–76.8%.
• VMK showed significant accuracy improvements (10.9%–13.4% and 3.2%–6.6%) over methods without early maps (NVM) and with early maps as features (VMF).
• The VMK method produced more spatially reasonable vegetation maps.

Conclusions:

• Integrating knowledge from existing vegetation maps significantly improves mapping accuracy at the vegetation formation level.
• The VMK method effectively addresses the challenge of distinguishing similar vegetation types.
• This approach substantially lowers the requirement for field samples, proving valuable for remote and challenging environments.