Geo-parcel-based land cover mapping and uncertainty analysis with airborne hyperspectral remote sensing images

Hyperspectral remote sensing images are effective for high-precision land cover (LC) mapping. However, data and information level issues accumulate during mapping, creating uncertainty. We propose a new geo-parcel-based mapping method and an information-entropy-based uncertainty calculation approach. To verify its accuracy and quantify uncertainty, we compare the classification accuracies of three different mapping units. This letter presents three quantitative uncertainty assessment indices and verifies their rationality. Results show that the geo-parcel-based LC mapping approach has the highest Overall Accuracy (OA) and Kappa coefficient. Variance analysis validates the reasonableness of the quantitative uncertainty indices. The three q-values of the uncertainty indices (0.6793, 0.5755 and 0.5812) effectively indicate the reliability of LC mapping at both pixel-levels and geo-parcel-levels. In conclusion, this letter highlights the accuracy advantages of aerial hyperspectral LC mapping at the geo-parcel-level and offers a quantitative way of describing uncertainty. The preliminary results highlight the potential for further research in this area.