Land cover accuracy assessment in Okitipupa, Ondo State, Nigeria; application of atmospheric correction and machine learning algorithms.

Abstract

The interaction of solar radiation with the atmosphere causes changes in the solar radiation that the Earth's surface reflects satellite sensors. Therefore, by eliminating air influences from satellite images, applying an atmospheric correction aid in determining genuine surface reflectance values and retrieving physical properties of the Earth's surface, including surface reflectance. Perhaps the most crucial step in pre-processing data from satellites that have been remotely detected is atmospheric correction. Here, we assessed Okitipupa, Ondo State, Nigeria's land cover classification using Landsat 8 image. The acquired Landsat image was subjected to Quick Atmospheric Correction (QUAC), Dark Object Subtraction (DOS), and Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) algorithm. The corrected image was applied to create the land cover classification using random forest (RF) and support vector machine (SVM) techniques. Four different classes were used in this study: built-up, shrubs, vegetation, and wetland/river. The land cover classification accuracy was in the following order: 0.98 and 0.96 > 0.97 and 0.95 for SVM_FLAASH and SVM_QUAC. This was followed by SVM_QUAC with an overall accuracy of 0.97 and a kappa coefficient of 0.95. Quick Atmospheric Correction (QUAC), Dark Object Subtraction (DOS), and Fast Line-of- Sight Atmospheric Analysis of Spectral Hypercubes are the three atmospheric correction algorithms (FLAASH).