Pansharpening Remotely Sensed Data with R

What is data fusion, and What is pansharpening?

Data fusion, simply put, is a method of combining different modes of data (whose degrees of heterogeneity can vary) into a single output. While with the purpose of data fusion varies from project to project, the unifying goal is to extract information from the fused result that could not be easily obtained from either source alone. Bear in mind the quote, "The whole is greater than the sum of its parts."

Pansharpening is one common method of data fusion. Pansharpening is performed to increase the spatial resolution of remotely sensed imagery by fusing an image's panchromatic band with its multispectral bands. Due to the way in which a sensor captures an image, the panchromatic band always has a higher spatial resolution than its multispectral counterparts. With Landsat images, the panchromatic band has a spatial resolution of 15 meters, while the multispectral bands have a spatial resolution of 30 meters. ​

Why use R for Raster Analysis? Why not use Python or some other software?
​
There is nothing wrong with using Python. You can certainly perform the same tasks in that language. However, performing raster analysis and geoprocessing is not a huge leap away from R's natural ability to work with matrices; R's primary function has traditionally been as a statistical, scientific, and data driven programming language. Since raster data is simply matrix data that has been given geographic context, it should be no surprise that R also works well as an alternative to other raster GIS software like TerrSet, ERDAS, or ENVI.

The data and the methods:

The R scripts below illustrate some simple pansharpening methods in data fusion.

The first attempts to follow the Intensity-Hue-Saturation (IHS) method described in the TerrSet/Idrisi help guide.

The other is a Brovey transformation method described in Jixian Zhang's 2010 publication, "Multi-Source Remote Sensing Data Fusion: Status and Trends" in the International Journal of Image and Data Fusion.

The sample Landsat data I used can be obtained from: https://clarklabs.org/download/

IHS Transformation

Brovey Transformation

Comparing Outputs:

True Color Composite - Multispectral Bands

TerrSet IHS Method

Panchromatic Band

IHS Method in R - Attempt at Emulating TerrSet Methodology

Brovey Method

These images are zoomed in so you can more clearly compare methods. You can see that noise tends to increase from the spatial sharpening and there is some spectral distortion, which can make quantitative measurements of spectral change over time more difficult. These methods are generally most suitable for visual interpretation.

Also note the difference between the TerrSet output and the R IHS output. The R IHS output shows more visual contrast between features, but there is also more noise. If you want to perform this IHS method in R and want less noise, multiply the differenced values by a smaller value. However, be aware that this will also reduce visual contrast between features.

---

Works Referenced:
​
Jixian Zhang (2010) Multi-source remote sensing data fusion: status and trends, International Journal of Image and Data Fusion, 1:1, 5-24, DOI: 10.1080/19479830903561035