Evaluating the Impact of Drought Using Remote Sensing Methods

Histogram stretching, sometimes referred to as contrast stretching, “changes the distribution and range of the digital numbers assigned to each pixel in an image" (Sahidan et al., 2008, p. 1). Since the majority of images do not utilize the "total range of the output device," the contrast between features within an image can be weaker than if the image was enhanced to allow for the pixel value range to reach its full potential (Sahidan et al., 2008; Xue, 2022, p. 34). The two camps of contrast enhancement include linear and non-linear mechanisms (Jensen, 2016).

Linear contrast enhancement "linearly expands the original digital values of the remotely sensed data into a new distribution. By expanding the original input values of the image, the total range of sensitivity of the display device can be utilized. Linear contrast enhancement also makes subtle variations within the data more obvious" (Al-amri et al., 2010, p. 27).

The linear contrast enhancement methods used in this analysis are Minimum Maximum,

Percent Clip, and Standard Deviation histogram stretches. “Non-linear contrast enhancement "often involves histogram equalizations through the use of an algorithm" (Al-amri et al., 2010, p. 27). The non-linear contrast enhancement method used in this analysis is the histogram equalization process which occurs as a part of the Dynamic Range Adjustment applied to the image.

Dynamic Range Adjustment (DRA) changes the underlying pixel values of an image only within the specific extent of the image being viewed through a program like ArcGIS Pro (Pitcher, 2022; Simularity, n.d.; W, 2011). DRA is an automated contrast enhancement of the image portion being viewed which utilizes an algorithm to calculate the color depth of the viewing area based on "the brightest value in the image" (Simularity, n.d., para. 2). Color depth is then implemented for histogram equalization. Histogram equalization is a form of nonlinear contrast enhancement in which the algorithm analyzes each band in the image and equally distributes the pixels to classes based on the "cumulative frequency histogram" (Al-amri et al., 2010; Jensen, 2016, p. 288). Next, "the image is converted back to RGB using the equalized Brightness value" (Simularity, n.d. para. 2).

Based on the pixel value distribution, histogram equalization can assign identical values to pixels that originally contained distinctive values or the proximity of originally adjacent pixel values (Jensen, 2016). The equalization stretch allows the use of the entire color range as it brings up the underrepresented tail ends of the data values to be equally represented throughout the image (Okin, 2022b). The equalization stretch increases contrast "the most populated range of brightness values of the histogram (or 'peaks')" and "reduces the contrast in very light or dark parts of the image associated with the tails of a normally distributed histogram" (Al-amri et al., 2010, p. 27). The advantage of DRA is seen in the difference between viewing an image entirely with DRA enabled and then zooming into a smaller subset of the image (Pitcher, 2022). In Dr. Pitcher's Technical Demonstration video "Image Histogram Stretching," he views the entire image with DRA enabled, and the contrast between the water features and the land features is lower than when he zooms in to the portion of the river. The DRA recalculates the histogram equalization based on the subset of the viewed image and produces a higher contrast between the river and the land features.

The Minimum Maximum Histogram Contrast Stretch and other forms of linear contrast enhancement are "best applied to remotely sensed images with Gaussian or near-Gaussian histograms" (Jensen, 2016, p. 282). The Lansat 9 image of the greater Las Vegas region sans the enabling of DRA or any stretch types has a near-Gaussian histogram (see Appendix Figure 1). It, therefore, is an ideal candidate for the Minimum Maximum Histogram Stretch. While this is "rarely the case" for images that "contain both land and water bodies," the image shown below likely presents with a near-Gaussian histogram due to the relatively few pixels which contain water features in comparison to many pixels which contain land features (Jensen, 2016, p. 283). The water features' presence in the image can be seen in the lowest value region of the histogram, presenting a typical spectra profile response to the bands (see Appendix Figure 2). The Minimum Maximum stretch manipulates and alters pixel values in images using the calculation of Equations 1 (see Appendix) (Jensen, 2016).

The effect of enabling DRA in conjunction with the Maximum Minimum linear stretch can be seen in Figure 3. It is evident that the pixel values were adjusted, or recalculated, by the Minimum Maximum contrast enhancement. Figure 4 (see Appendix) shows the original pixel values for each band in grey behind each new histogram generated by the Minimum Maximum stretch. The image has become brighter as a whole, evidenced by Figure 4 with the shiting of the histogram peak (Okin, 2022a). It is also evident in Figure 4 that the contrast between the bright soil, concrete, and land features and the much lower-valued water features is increased by the distancing of their values.

For ease of viewing and to compensate for the DRA's impact on the zoomed-in image seen in Figure 3, Figure 5 (see Appendix) provides a comparison of the entire images seen in Figure 1 and Figure 3. Appendix Figure 5 shows an increased brightness of pixels in the image with DRA enabled and Minimum Maximum stretch applied (on the right) in comparison to the image without DRA enabled and no stretch applied. The enhanced contrast between the water and land features seen in the right image shows the power of the DRA and Minimum Maximum stretch. Additionally, the enhancements also provided a greater contrast between soil and vegetation.

The Percent Clip method of linear contrast enhancement uses the same formula as the Minimum Maximum contrast enhancement but redefines the mink and maxk vaues, effectively "trim[ming] extreme values from both ends of the histogram" based on a percentage (L3 Harris Geospatial, n.d., para. 2; Okin, 2022a). The Percent Clip contrast enhancement reassigns all pixels containing values that previously resided outside of the newly defined minimum and maximum values are given the minimum or maximum value on their respective side of the histogram (ArcGIS Desktop, n.d.). For example, as shown in Figure 6 (see Appendix), the Percent Clip Minimum and Maximum values, auto-defined by the implementation of the Percent Clip stretch in the toolbar, are 0.25. This indicates that ArcGIS Pro removed 0.25% of values residing at the lower and upper portions of the image value range and reassigned those values to the redefined maximum and minimum at the 0.25% cut-off, respectively (ArcGIS Pro, n.d.a).

Figure 7 (see Appendix) shows that the Percent Clip contrast stretch provided a much stronger brightening effect than the Minimum Maximum contrast stretch seen in Figure 4 (see Appendix). The amplitude of the brightening effect is evidenced by the drastic shift in the histogram peak (Okin, 2022a). For ease of viewing, Figure 8 (see Appendix) provides a comparison of the entire images resulting from the Minimum Maximum (left) and Percent Clip (right) contrast stretches. Figure 8 shows the increased contrast provided by the Percent Clip stretch compared to the Minimum Maximum stretch, evidenced by the stark distinction between water and land features and soil types, vegetation, and man- made materials.

For ease of viewing and to compensate for the DRA's impact on the zoomed-in image seen in Figure 6, Figure 9 compares the entire images seen in Figure 1 and Appendix Figure 6. Figure 9 shows the benefits of utilizing a Percent Clip stretch on an image to increase the accentuation of land and water features.

The Standard Deviation linear contrast enhancement stretch also utilizes the formula implemented by the Minimum Maximum contrast enhancement. However, the mink and maxk values of this stretch are defined by the "standard deviation value" (ArcGIS Desktop, n.d., para. 7; Jensen, 2016). Similarly to the Percent Clip stretch, any pixel values outside these newly defined minimum and maximum values are reassigned to the new minimum or maximum values, respectively. For example, in Figure 10 (see Appendix), the autogenerated number of standard deviations selected by ArcGIS Pro when selecting the Standard Deviation stretch type on the toolbar was 2.

Therefore, "the values beyond the second standard deviation" are assigned as the new minimum and maximum values, respectively, while anything in between is linearly stretched (ArcGIS Pro, n.d.b, para. 8).

In Figure 11 (see Appendix), it is evident by the drastic shift of the apex to the right that the Standard Deviation contrast stretch created the largest increase of brightness throughout the image compared to the Minimum Maximum and Percent Clip stretch histograms seen in Figures 4 and 7, respectively (see Appendix). The Standard Deviation stretch's slope is "greater than for a simple min-max contrast stretch" (Jensen, 2016, p. 286). For ease of viewing and to compensate for the DRA's impact on the zoomed-in image seen in Figure 10, Figure 12 provides a comparison of the entire images for all stretches conducted. Evaluating the images side-by-side while viewing the full images, it is debatable whether the Percent Clip or Standard Deviation stretches provides the best contrast between land and water features. The Standard Deviation stretch image contains highly contrasting pixels, with many being dark and many being very light. A quick glance at the four images side by side suggests that the Percent Clip is the easiest version to quickly identify the water feature. However, when zooming in on the water feature of the images, the Standard Deviation stretch with DRA enabled clearly provides the crispest and most contrasted differentiation between land and water features.