Model answer

Supervised classification — you teach the computer what each class looks like, then it sorts every pixel.

How it works

1. Find places you know — forest patches, water bodies, urban blocks.
2. Sample those — give the computer color signatures of each class.
3. Pick a decision rule — Max Likelihood is the most common.
4. Run it — every pixel in the scene gets a class label.
5. Check accuracy — compare to ground-truth points you didn’t use for training (confusion matrix, kappa).

Strengths

• Higher accuracy than unsupervised when your training data is good.
• You control the class scheme — useful when you have a specific question.
• Works well when classes are spectrally distinct.

Weaknesses

• Training-sample collection is slow — fieldwork, photo interpretation, ground truth.
• Bias risk — if you forget a class, the computer can’t predict it.
• Max Likelihood assumes Gaussian classes — fails on weird-shaped distributions.

🔬 Show the science / technical version

Procedure

1. Define training sites (AOIs) for each known class.
2. Compute class signatures (mean + covariance per band).
3. Apply a decision rule — Min Distance, Mahalanobis, Max Likelihood, Parallelepiped — to every pixel.
4. Accuracy assessment (confusion matrix, kappa).

Advantages

• Analyst controls the class scheme — matches the question.
• Higher accuracy than unsupervised when training data is good.
• Works well when classes are spectrally distinct + well-sampled.

Disadvantages

• Labor-intensive — good ground truth takes time.
• Bias risk — unrepresented classes won’t be predicted.
• Max Likelihood assumes multivariate-normal class distributions.