Model answer

Unsupervised classification — let the computer find natural groupings, label them after.

How it works

1. Tell the computer how many clusters to look for (say, 30).
2. ISODATA (or K-means) iteratively groups similar pixels.
3. You inspect the result and label each cluster — “cluster 7 looks like wetland.”
4. Often merge redundant clusters with the Recode tool.

Strengths

• No training data required — fast first look at unfamiliar imagery.
• Reveals natural groupings the analyst didn’t predict.
• Useful as a starter before doing supervised classification, or for stratified sampling.

Weaknesses

• Clusters don’t always match meaningful categories — one class may split across multiple clusters, one cluster may contain several classes.
• Sensitive to K and initial seeds — different runs give different results.
• Post-classification cleanup is still required — labeling, merging, re-running with different K.

🔬 Show the science / technical version

Procedure (ISODATA / K-means)

1. Choose number of clusters.
2. Iterate: assign pixels to nearest centroid → recompute centroids → repeat until convergence.
3. Analyst labels each cluster post-hoc from reference data.

Advantages

• No training data required — fast first pass.
• Reveals natural spectral groupings the analyst might not have predicted.
• Useful as preprocessing for supervised classification or for stratified sampling.

Disadvantages

• Clusters don’t always map to meaningful classes (one class may split; one cluster may contain several classes).
• Sensitive to K and initial seeds — different runs give different results.
• Post-classification labeling/merging (often via Recode) still required.