Suitability of Image Classification Method for Hydrologic and Hydraulic Applications

Present methodologies of image classification give moderate accuracy due to, intermixing of the pixels and classification of the pixels into any other class. In hydrologic modeling, curve number are mainly dependent on land use and land cover (LULC) classification and in hydraulic modeling, mostly Manning’s coefficient are assigned based on the image classification hence its important to classify the geospatial image more accurately. Present study addresses the best methodology to be adopted for the water resources engineering problems. Comparison of Isocluster (IC), Generalized Likelihood Image Classification Method (GLIM) and Index based image classification method were used for the study area. Present study focused on the level I classification viz. Forest, Water, Urban, barren and Agricultural land. Accuracy assessment of the image classification was carried out by using Commission Error, Omission Error, Producer accuracy, User accuracy, Kappa coefficient and Overall Accuracy. All the errors were estimated by forming confusion matrix for the respective methods of image classification. Kappa coefficient for the GLIM method is greater than 0.63 indicates moderately agreement with ground reality data. Kappa coefficient for IC method was less than 0.5, hence indicating poor agreement with the ground truth data of class. The accuracy of Index based classification is greater than 0.8. Hence it indicates that, there is less number of pixels being classified to another class. The present methodology comes up with the strong conclusion that Index based image classification is best suitable for water resource application.