1 of 23

The Myth of Enhance

Tips and Tricks for Storytelling With Satellite Data

Leanne Abraham • October 14, 2020 • NACIS 2020

2 of 23

San Francisco, California

September 11, 2019

©Planet Labs

SPECTRAL

RESOLUTION

RADIOMETRIC

RESOLUTION

SPATIAL

RESOLUTION

TEMPORAL

RESOLUTION

As cartographers we are constantly trying to wrangle data from a wide variety of sources. One of the sources that intimidated me the most before starting at Planet was raster data, and specifically satellite imagery. So today, I am hoping I can do a bit to “demystify” some basics and provide solutions for a few “gotchas” that trip me up to this day.

I am going to focus on data that has been collected by earth imaging satellites, but I acknowledge that there are many other sources of remote sensing data that will share many of the characteristics and cavates to the data sources I will discuss here.

I will focus on passive sensors, as different techniques are needed to analyze active sensors such as Iidar.

My goal here is to break down two questions

what do I need to know to use this data in a map
what are new maps that could make with this data?

Although this second one is really up to you

------

I will do this by breaking down imagery by *****,

I will define each of these as we go.

3 of 23

SPECTRAL RESOLUTION

4 of 23

Spectral Resolution

So you might be thinking, what do wavelengths have to do with it? I am used to thinking about satellite imagery in terms of bands!

So, a band is just specified wavelength interval in the electromagnetic spectrum.

What I am showing here are the spectral responses for different satellite constellations at different intervals across the visible and NIR portion of the electromagnetic spectrum.

There is a good amount of variation between what different systems consider a band, and even variation between the size of the slice of the spectrum that is collected depending on the system. So, for example GOES doesn’t have a visible green band, while MODIS has 3.

Tip: When creating an image composite it is important to double check and make sure the band you are using is the recommended one for the task! There are different characteristics between bands other than their location on the electromagnetic spectrum, which is to say if there are 3 NIR bands that doesnt mean they are all interchangeable.

5 of 23

Spectral Resolution

6 of 23

Spectral Resolution

7 of 23

Spectral Resolution

8 of 23

Santa Cruz, California - September 6, 2020

Sentinel-2

9 of 23

Red, Green, Blue

Santa Cruz, California - September 6, 2020

Sentinel-2

10 of 23

NIR, Red, Green

Santa Cruz, California - September 6, 2020

Sentinel-2

11 of 23

SWIR, NIR, Green

Santa Cruz, California - September 6, 2020

Sentinel-2

bands="$swir1 $nir $green"

gdalbuildvrt -resolution lowest -separate ${name}_sng.vrt $bands

In this combination, newly burned areas are reflective in swir and appear red darker red than other bare ground, and easier to distinguish than in the NRG combo. Plant health is easy to discern in places where the image is almost electric green. This is because plants are highly reflective in NIR

One thing I haven’t touched on yet is that the spatial resolution of different bands can vary even within the same constellation.

This image is a good example of where I got caught compositing different bands without checking the resolution first. The NIR and Green were 10m, but SWIR is 20. This means when I composited them the pixels did not align perfectly.

Tip: Using gdal’s flag resolution lowest resampled everything to 20m

Not touched on:

Band math and using indexes for things like NDVI, which would give you an approximation of vegetation distribution, or NDWI, which could be used to create a water mask.

12 of 23

RADIOMETRIC RESOLUTION

13 of 23

Common Bit Depths

gdal_translate -scale 0 4096 0 65536

14 of 23

Santa Cruz, California - September 6, 2020

Bit depth also becomes important when saving an image, because it has a direct impact on image size

A binary image (0 or 1) will be smaller than a single band 8 bit image, which will be smaller than a 3 band color image.

And, because as we now know, an image can have any number of bands. Decisions about radiometric resolution are compounded as you add more bands.

Therefore, when saving a file that you have worked on, you should be sure to give some thought to how much detail needs to be contained in the pixels in an image. For a color corrected image for illustrative purposes (think a base map) 8 bits is often more than enough, and will be half the size of a 16 bit image with no visible reduction in quality.

Tip: I won’t cover this here, but the compression you use when saving an image can also have a huge impact on file size. I tend to use LZW on tiffs when I save them from either photoshop or gdal.

Bit depth is also important when creating gifs, if you want to chat about that feel free to slack me after this presentation!

DO NOT SAY This doesnt apply only to saving an image but also when you convert into different formats. If you try to turn a series of 8 bit images into a 256 color gif it will like more similar to the 4 bit or 2 bit image than the 8 bit shown here because all the images in the sequence will have to fit within that color range (this is also why if anyone ever asks you to make a gif of a series of satellite images you will have a huge file on your hands). Classified data that uses the same color over and over however should look ok

15 of 23

SPATIAL RESOLUTION

16 of 23

Santa Cruz, California - August 4, 2018

National Agriculture Imagery Program (NAIP)

17 of 23

Santa Cruz, California - August 4, 2018 & September 6, 2020

18 of 23

Santa Cruz, California - August 4, 2018

National Agriculture Imagery Program (NAIP)

19 of 23

Pansharpening

20 of 23

TEMPORAL RESOLUTION

21 of 23

Different phenomena have different ideal temporal resolutions, and how often there is a useable image of a place can depends on clouds.

Currently landsat revisits about every 16 days

Sentinel-2 (Either A or B) revisits about every 5

Or, for example, GOES in in a geostationary orbit producing 1km pixels in the visible spectrum, but scans the continental US every 15 minutes.

Typically, the higher the spatial resolution, the lower the revisit time, but this rule of thumb can change depending on how many satellites are in the constellation

Latitude also plays a role, at higher latitudes the swath widths can overlap for an increased revisit rate, more but the viewing angles will be slightly different. This means if you are looking for an image at a higher latitude you potentially would have a few more options to choose from. But there will be other considerations such as the appearance of shadows.

For example, GOES in in a geostationary orbit producing 1km pixels in the visible spectrum, but scans the continental US every 15 minutes.

The ability to task a satellite (point it using on-board propulsion to capture a particular location) can alter temporal resolution. For example Planet Skysat’s have a theoretical revisit rate of multiple high resolution collects in a day, but only if an aoi has been tasked. Otherwise, there are locations on earth that have never been imaged by that sensor.

Tip: When choosing a constellation for your map, you will want to be conscious of how often imagery will be collected. You will also This means that you will likely want to pay attention to collection time since this can impact how colors appear (scenes at noon the light will be brighter, longer shadows early in the morning etc)

22 of 23

Remote Sensing of the Environment: An Earth Resource Perspective

John R. Jensen, 2007, p.37

23 of 23

Thanks!

Any Questions?