Vegetation type and land cover mapping in a semi-arid heterogeneous forested wetland of India: comparing image classification algorithms

Deval, Kundan , Joshi, P. K.

2022-03-01 ENVIRONMENT DEVELOPMENT AND SUSTAINABILITY 2022   24(卷), 3(期), (3947-3966页)

查看原文

The present study evaluates and compares performance of three supervised classification algorithms namely Maximum Likelihood (MXL), Artificial Neural Network (ANN) and Support Vector Machine (SVM), using very high resolution WorldView-2 satellite imagery for vegetation type/land cover (VT/LC) mapping in Keoladeo National Park (KNP), India. We mapped 16 (8 gregarious VT and 8 LC) classes, and used Bootstrap (with 100 iterations) method for accuracy assessment. All three algorithms produced high overall accuracy (OA) (67-85%) and kappa (K) (65-83) values. Visual comparison of the predictions revealed that SVM (OA = 85.12% (K = 83.9) with 3.85% width of confidence interval) performed the best followed by ANN (69.72% (67.32) with 4.43%) and MXL (67.37% (65.22) with 4.33%). This research provides insight for selection of classification algorithm for detailed VT/LC mapping of wetland associated systems using very high resolution satellite data. The findings of this research are useful for environmental management, restoration and conservation planning of KNP, India. The database will be of high value for future development and sustainability issues in the park.