Comparison modeling for alpine vegetation distribution in an arid area.

Area of Science:

• Ecology
• Remote Sensing
• Geographic Information Systems (GIS)

Background:

• Predictive vegetation distribution modeling is advancing rapidly with new statistical techniques and GIS tools.
• Modeling alpine vegetation in arid, complex environments remains a significant challenge due to environmental heterogeneity.

Purpose of the Study:

• To discriminate vegetation groups and formations in the Qilian Mountains using various data and models.
• To evaluate the performance of Decision Tree (DT), Maximum Likelihood Classification (MLC), and Random Forest (RF) models for alpine vegetation mapping.

Main Methods:

• Utilized 70 variables from ASTER GDEM, WorldClim, and Landsat-8 OLI (albedo, vegetation indices).
• Employed Decision Tree (DT), Maximum Likelihood Classification (MLC), and Random Forest (RF) models.
• Validated models against field data and existing vegetation maps.

Main Results:

• Variable combinations effectively discriminated vegetation groups but not formations.
• Elevation was consistently the most important parameter for alpine vegetation modeling.
• The Random Forest (RF) model achieved higher accuracy (75% overall accuracy, kappa 0.64) compared to DT and MLC models.

Conclusions:

• The Random Forest model offers superior accuracy for alpine vegetation modeling in this region.
• Model accuracy is influenced by variable combinations and chosen modeling approaches.
• Further research is needed to refine methods for discriminating vegetation formations in complex alpine environments.