A New approach to assessing the accuracy of the LULC classification using Remote Sensing & GIS in Greater Mumbai, Maharashtra, India

This study focuses on developing an advanced accuracy assessment of supervised LULC-classified remote sensing images in the Greater Mumbai area. The commonly used Kappa index method was found to be problematic in terms of the accuracy of the assessment and interpretation of classified images. Therefore, this research aims to review the commonly used methods, such as the Pixel-based error, and create a new method, such as the unbiased area error matrix. The study also determines which method is better suited for assessing the quality of the classification. The new statistical method involves an unbiased area estimation error matrix that provides an overall assessment of the classification's accuracy, including the User’s accuracy, the Producer's accuracy, and a 95% confidence interval. The nature and magnitude of the errors, their source, and the reasons for errors within each class are identified for both the User’s and the Producer's accuracy. Comparing the unbiased area estimates error matrix to the traditional pixel-based error matrix, this study finds that the unbiased area matrix is more accurate for assessing image classification. Additionally, this paper suggests correcting the identified errors for further improvement in the classification.