Supervised Land Cover Classification
This lab applied supervised classification to Landsat imagery to map land cover types using training data and spectral analysis. Training polygons were manually digitized in QGIS to represent different land cover classes, and spectral reflectance values were used to train a classification model in R Studio The resulting land cover map was evaluated using a confusion matrix to assess classification accuracy.
Workflow
Note: Image classification and accuracy assessment were conducted using a provided R-based workflow developed for the course; my work focused on preparing training data, evaluating spectral separability among land cover classes, and interpreting classification results.
Step 1 – Define Training Data for Land Cover Classes
Polygons were digitized in QGIS representing seven land cover classes (e.g., coniferous forest, water, developed areas). Then polygons were split into training (70%) and validation (30%) datasets for model development and accuracy assessment.
Step 2 – Analyze Spectral Signatures of Land Cover Classes
Landsat pixel values were extracted from training polygons across six spectral bands. Mean reflectance and variability for each class was calculated to visualize spectral signatures and assess class separability.
Step 3 – Perform Supervised Classification
A Maximum Likelihood Classification (MLC) algorithm was applied to classify the Landsat image into seven land cover classes based on the training data. Then, a final land cover classification map representing spatial distribution of classes was generated.
Step 4 – Evaluate Classification Accuracy
Predicted land cover classes were compared with validation data using a confusion matrix. Metric were calculated including overall accuracy, producer's accuracy, and user's accuracy to evaluate classification performance.
Tools & Technologies
R Studio
QGIS