Vegetation Types Mapping Using Multi-Temporal Landsat Images in the Google Earth Engine Platform.

Area of Science:

• Remote Sensing
• Ecology
• Geospatial Analysis

Background:

• Vegetation Types (VTs) are critical for land cover management and conservation.
• Accurate identification of sparse VTs, particularly in arid/semiarid zones, remains a challenge for Earth observation.
• Multi-temporal datasets offer potential for enhanced VTs classification.

Purpose of the Study:

• To identify optimal multi-temporal datasets for improving Vegetation Type (VT) classification accuracy.
• To assess the effectiveness of Landsat 8 multi-temporal data for VT detection in a heterogeneous landscape.
• To evaluate the utility of Google Earth Engine (GEE) for selecting optimal imagery and periods for classification.

Main Methods:

• Analysis of Normalized Difference Vegetation Index (NDVI) temporal profiles for VTs from 2018-2020.
• Selection of optimal time-series images based on phenological patterns.
• Comparison of single-date versus multi-temporal Landsat 8 data using the Random Forest classifier in GEE.

Main Results:

• Multi-temporal Landsat 8 data achieved a median Overall Kappa of 74% and Overall Accuracy of 81%.
• Single-date classification yielded significantly lower results (median Overall Kappa 51%, Overall Accuracy 64%).
• GEE facilitated efficient identification of optimal periods and imagery for VTs classification.

Conclusions:

• Exploiting multi-temporal satellite datasets substantially enhances the accuracy of Vegetation Type classification.
• Cloud-based platforms like GEE are valuable tools for optimizing remote sensing-based land cover analysis.
• Improved VT identification supports effective land cover management and conservation strategies.