Fig. 3
From: Mapping 10-m Industrial Lands across 1000+ Global Large Cities, 2017–2023
Confusion matrices for industrial land classification, 2017–2023. Each matrix shows the true positive rate (correctly classified industrial land), false positive rate (non-industrial land incorrectly classified as industrial), false negative rate (industrial land incorrectly classified as non-industrial), and true negative rate (correctly classified non-industrial land). The overall accuracy (OA) for each year is shown above the matrices, highlighting the consistent performance and robustness of the classification model across years. Each matrix aggregates results from 43 clusters, weighted by the built-up area of the cities within each cluster. Validation sample sizes for both industrial and non-industrial land are 500 per cluster, per year.
