An automated procedure for cluster analysis of multivariate satellite data

Area of Science:

• Remote Sensing
• Data Science
• Geospatial Analysis

Background:

• Multivariate satellite data analysis requires effective clustering techniques.
• Existing methods like PCA, K-means, SOM, and ART have limitations in practical application.
• Assessing the physical relevance of clustering results is crucial for scientific interpretation.

Purpose of the Study:

• To evaluate the applicability and usefulness of Principal Component Analysis (PCA), K-means, Self Organizing Maps (SOM), and Adaptive Resonance Theory (ART) for multivariate satellite data.
• To develop and present a novel automated cluster analysis procedure.
• To demonstrate the capability of the new method in achieving meaningful data partitions for physical analysis.

Main Methods:

• Comparative analysis of four clustering algorithms: PCA, K-means, SOM, and ART.
• Development of a combined Self Organizing Maps (SOM) adaptive dynamic K-means procedure.
• Focus on the relevance and utility of clustering results for physical interpretation of satellite data.

Main Results:

• The study assesses the effectiveness of PCA, K-means, SOM, and ART on multivariate satellite data.
• A new combined SOM-ART-K-means method is presented for automated clustering.
• The proposed method demonstrates utility in partitioning previously unstudied multivariate satellite data.

Conclusions:

• The combined SOM-ART-K-means procedure offers a valuable tool for automated cluster analysis of multivariate satellite data.
• This novel approach enhances the physical analysis of satellite data by providing relevant partitions.
• The study highlights the importance of evaluating clustering methods based on their practical utility in scientific applications.