Haze Optimized Transform on LANDSAT 8 Imagery for Thin Cloud Detection and Removal

Abstract

Landsat images, being optically captured, contain one of the most fundamental remote sensing issues-- cloud and haze contamination. Various algorithms have been developed through the years to correct haze contamination and maximize the use of archived Landsat images since its launch in 1972. One such algorithm is the Haze Optimized Transform (HOT). HOT identifies spatially varying haze thickness on the premise that clear sky conditions can be characterized from the regression of highly correlated blue and red bands; and features diverging from this relation indicate thickness of haze using their orthogonal distance (HOT values) from clear sky function. A modified Dark Object Subtraction is then performed based on histogram matching per HOT value versus the clear sky case.

This paper presents modifications in applying the HOT algorithm considering the effects of increased radiometric resolution and new coastal blue band in Landsat 8 were tested on two separate images with different dates of acquisition from a test site in the Davao Oriental province of southern Philippines, which was selected due to its prevalent cloud cover condition throughout the year. The effects of the increased radiometric resolution and new coastal blue band in Landsat 8 were tested on two separate images with different dates of acquisition. Haze correction using the coastal blue band demonstrates noticeable difference in adjustment for certain land cover types. On the other hand, the increase in radiometric resolution shows exponential effects to HOT value ranges which translates to finer haze depth estimation but at the expense of performance. Moreover, applying the algorithm demonstrates a higher rate of over correction, which is then compensated by applying a clear aerosols fraction adjustment.

The corrected images are then further processed to compute Normalized Difference Vegetation Index and Supervised Classification to show the effectiveness of the HOT correction algorithm. This study shows that the HOT algorithm with the presented modifications can be efficiently and effectively implemented on Landsat 8 images, and obtain the desired results.

Keywords— Remote Sensing, Relative Radiometric Correction, Haze Removal, Landsat 8, Haze Optimized Transform