Evaluation of the impact of spatial accuracy of satellite imagery and supervised classification algorithms on the accuracy of land cover maps - Case study: Latakia Governorate.

Abstract

Remote sensing technology is considered one of the most effective methods for obtaining land cover maps, which are produced by applying satellite image classification techniques that are influenced by factors.

In this research, the impact of the type of classification algorithm and the spatial resolution of the satellite image used on the accuracy of the land cover map for a part of Latakia Governorate was studied. The supervised classification was applied with Maximum Likelihood, Mahalanobis distance, and minimum distance algorithms on Landsat 8 satellite image and Sentinel A image after resampling at resolutions of 15m and 10m, respectively.

The results showed that the Maximum Likelihood algorithm is the most accurate in classifying the Sentinel 2A satellite imagery for land cover mapping, with an overall accuracy of 88.2% and a kappa coefficient of 0.81. As for Landsat 8 imagery, the Minimum Distance algorithm was the most accurate, with an overall accuracy of 83.3% and a kappa coefficient of 0.79. On the other hand, the results indicated an improvement in the overall accuracy values for all algorithms with increasing spatial resolution of the classified imagery. The overall accuracy increased from 80.1% to 88.2% for the Maximum Likelihood algorithm, from 78.2% to 80.9% for the Mahalanobis Distance algorithm, and from 83.3% to 85.3% for the Minimum Distance algorithm. We also observed an improvement in the kappa coefficient values from 0.75 to 0.81 for the Maximum Likelihood algorithm, from 0.71 to 0.75 for the Mahalanobis Distance algorithm, and from 0.79 to 0.80 for the Minimum Distance algorithm. Finally, the results indicated that with increased image spatial accuracy, there is less overlap between class ranges, resulting in a reduction in misclassification of certain pixels.