Lab VII: Evaluation of recreational impacts on eelgrass using UAS and virtual ground truth data

Last updated 2/24/2024

Objective: This lab will involve the use of imagery acquired using a multispectral camera carried by an Unoccupied Aerial System (UAS) to evaluate the impact of recreational activity on eelgrass (Zostera sp.). This will involve the use of regression analysis and “virtual” ground truth data.

Eelgrass (Zostera sp.) provides a variety of important ecosystem services including stabilizing sediments, buffering storm surge and nutrient filtration. It also provides critical habitat for salmon, crab, shellfish and seabirds. Recreational impacts from boat traffic, docks and mooring have been shown to have adverse effects on eelgrass. Boat propellers can cut or uproot the vegetation and turbulence from propellers can resuspend sediments generating turbidity and shading of the vegetation and smothering as the sediment settles. Chemical pollution from fuel, lubricants and anti-fouling paints can also have impacts. Docks can impact eelgrass due shading and altered hydrodynamics resulting in erosion and sediment translocation. Anchoring and mooring create mechanical damage to vegetation and also stir up sediments. Two species of eelgrass occupy the inland waters of the Pacific Northwest: Zostera marina and Zostera japonica. Z. marina, the native and most abundant species, is found in mid to sub tidal regions and Z. japonica, a non-native species, is less abundant and is found in upper to mid-tidal ranges. Z. japonica is believed to have reached North America from Japan in the packing materials of clam exports.

Traditional methods for monitoring eelgrass involve the use of ground-based surveys that are quite time consuming and limited in spatial extent. Imagery from satellites and traditional aircraft have also been used but these approaches can be expensive, lack sufficient spatial resolution and timing image acquisition to tidal cycles can be problematic. More recently, the use of imagery acquired using low-cost, unoccupied aerial systems (UAS) has been used. The advantage of UAS imagery is the low cost, very high spatial resolution (a few centimeters) and ease of timing image acquisition to tidal cycles.

Previous effort to monitor eelgrass using UAS imagery has primarily been limited to simply mapping the presence or absence of eelgrass without quantifying percent coverage or biomass. One recent study was successful in distinguishing between Z. marina and Z. japonica but only broke out four broad percent cover categories.

The objective of this exercise to evaluate the use of simple vegetation indices, derived from multispectral imagery acquired using UAS, in conjunction with a novel approach for obtaining ground truth data, to quantify variation in the percent cover of eelgrass and to evaluate the effect of recreational boat launch activity on eelgrass.

Study Area: Our study area for this exercise is Wildcat Cove. Wildcat Cove is part of Larabee State Park (just south of Bellingham) and it includes a boat ramp that is heavily used during the summer months, particular during the recreational crabbing season that typically opens in mid-July. During low tides, much of the cove is exposed mudflat and vehicles typically back down across the mudflat to get boat trailers to the water’s edge. Those launching kayaks and other human-powered boats may also drive across the mudflat to the water’s edge. This activity has an impact on eelgrass.

Preliminaries: As usual, go to the J-drive and to the ESCI442 folder. Grab the entire “eelgrass” folder. This folder includes imagery from two different date; July 14 was the Friday before the opening of recreational crabbing season and July 17 was the Monday after.

The data: We will be using data collected using the Micasense Dual Camera system. This is a 10-band camera. Here is how the bands on this camera compare to the sensors on Landsat 8 and Worldview 2.

And here are the specifications for each band.

Table 1: Band names, numbers, center wavelengths, and wavelength ranges of the Micasense Dual Camera System. The names, center wavelengths, and numbers will be used in further writing in the format Band Name _{Center Wavelength} (Band Number).

Band Name	Band Number	Center Wavelength (nm)	Wavelength Range (nm)
Coastal Blue	1	444	430-458
Blue	2	475	459-495
Green	3	531	524-538
Green	4	560	547-574
Red	5	650	642-658
Red	6	668	661-675
Red Edge	7	705	700-710
Red Edge	8	717	711-723
Red Edge	9	740	731-749
Near Infrared	10	842	814-871

I’ve done quite a bit of prep work for you. Each flight generated several thousand images. The “eelgrass” folder includes two PDF files that describe each flight and the processing that was done that I did to generate an orthomosaics that cover all of Wildcat Cove. I also georectified the imagery with several ground control points located around the perimeter of the cove.

View the imagery: Start by opening “Wildcat_7_14_mica_60m_GCP_5cm_clip” in ENVI. This is a 10-band tiff file and, as the filename suggests, the pixel size is 5 cm and the imagery was obtained by flying at an altitude of 60 m above ground level (AGL). There is an equivalent file for July 17, “Wildcat_7_17_mica_40m_GCP.” This also has a pixel size of 2.5 cm (I neglected to include this in the file name; sorry) but I flew at 40 m AGL instead of 60 m.

To view a color-IR image, you will need to go to Change RGB bands and load bands 10,6,4 into the red, green and blue color guns. You may need to try a few enhancements to brighten it up a bit. It will look something like this.

You could also display a true color image using one of the red, one of the green and the blue band (6,4,2). And, you can also view the image for July 17. Note that, in addition to the high-resolution images for each date, there is are also versions that have been resampled to generate images with a pixel size of 50 cm. More on this below.

You will note the tide was quite low when I did this flight and there is an extensive exposed mudflat covering much of the cove. The red stuff on the mudflat, and in the water at the mouth of the cove, is mostly eelgrass but there is also some algae as well. You will also note a bare path from the boat ramp to the water’s edge that is caused by people backing cars and boat trailers down to the water. Our goal here is to assess the impact of this activity on the eelgrass after a weekend of high activity associated with the opening of crabbing season.

Virtual Ground Truth: The three rows of white dots (or squares if you zoom in) are ground control panels that I laid out prior to my flight. The panels are spaced ~5 m apart and the spacing between each row is ~15 m. These panels were used to generate “virtual” ground truth data that we will use for the analysis. Prior to my flight to obtain this imagery, I also flew quite low, just 5 m AGL, over these panels an took a standard color image of each panel with a 20-megapixel RGB camera. I then brought these images into Powerpoint and superimposed a 4 by 4 grid adjacent to each panel. The panels are 45 cm by 45 cm and I resized the 4 by 4 grid to match the size of the panel. This ensures that I was sampling a consistent area adjacent to each panel. The cover type was visually assessed at the corner of each grid cell, resulting in data for 25 points within this 45 cm by 45 cm sample grid. Percent cover for each cover type was then calculated for each sample grid based on these sample points. The four cover types included eelgrass (Zostera sp), algae (Ulva sp., mostly Ulva intestinalis), bare and detritus. I did not distinguish between the two species of eelgrass. Detritus is mostly composed of dead eelgrass.

Here is an example of one of the “virtual” ground control panels taken from just 5 m AGL with 4 by 4 sample grid superimposed on the image. The cover type at the corner of each grid cell, was recorded and the percent coverage of each cover type was calculated for the grid. The four cover types included eelgrass (G), algae (A), bare (B) and detritus (D). Note that detritus was not present at this location.

A typical approach for obtaining vegetation ground truth data could involve walking to individual points in the field and laying down a 50 cm by 50 cm sampling frame constructed from PVC tubing. This sampling frame would include a grid of string that generates 25 grid intersections identical to what is depicted in the image above. This ground-based sampling is quite time consuming and difficult to complete within the narrow window provided by low tide events. The use of the virtual ground truth data described above is faster and generates data that is equivalent to the ground-based approach.

Image Processing: The “eelgrass folder includes a number of other files. In addition to the 5 cm resolution images, I also generated files that were resampled to yield a pixel size of 50 cm. I did this to match the area sampled by my virtual ground truth plots.

You will need to start by generating a Normalize Difference Index (NDI) with the 50 cm resolution images using this equation:

B8 is one of the red edge bands and B6 is one of the red bands. This index is similar to the widely used Normalized Difference Vegetation Index (NDVI). NDVI is generated using the same equation but uses a near-IR band instead of the red edge band. NDI has been shown to be a useful predictor of vegetation cover. If you take a look at the image above, you will note that it can only result in values ranging from -1 to +1. If B6 has a value of zero and B8 is something other than zero, NDI=1.0. If B8 has a value of zero and B6 is anything other than zero, NDI= -1.0. Any other combination of values for B6 and B8 will result in decimal values between -1.0 and +1.0

You will generate NDI images for each date by using the Band Math tool that we’ve used in previous labs. But before opening this tool, click on one of the 50 cm resolution images and look at the metadata. Take a look at the Data type. Note that it is “UInt.” This indicates that all of the data values in this image are unsigned integers. This is a bit of an issue since the NDI values that we will calculate will be floating point values (decimal values, not integers).

This will require a new trick in the Band Math tool. Open it up and let’s generate an equation. Based on the equation above, you might be inclined to write something like this:

This seems reasonable……but if you do this, you will end up with an image that consists entirely of 0 and 1 instead of the decimal values we want. This is because you are starting with an image with a Data type of “unsigned integer” and this will result in an output image with the same data type and all decimal values will be rounded to 0 or 1.

Bummer.

So you need to learn a new trick. You need to convert you unsigned integer values to floating point values within the Band Math tool so your output image will have the floating point values we want. So you need to write an expression like this: (float(b1)-float(b2))/(float(b1)+float(b2))

In the Variables to Bands Pairings dialog box, be sure to select Band 8 for “b1” and Band 6 for “b2.” Run this calculation for both the July 14 and the July 17 50 cm resolution images.

After doing this, and for each date, I loaded both the 5 cm image and the 50 cm NDI image in ENVI. I viewed the 5 cm image to enable me to precisely locate each panel, and then toggled to the 50 cm NDI image and recorded the NDI value adjacent to the panel. I’ve done this for you for all panels and all dates, including two other dates that we will not use for this analysis.

The NDI values and the percent vegetation coverage (including both eelgrass and algae) obtained from the virtual ground truth panels is provided in the Excel file that is in the “eelgrass” folder.

Modeling Eelgrass Coverage: Open the Excel file and take a look at these data. We will be using regression analysis to use NDI to predict the percent coverage of eelgrass and algae. You should be familiar with regression analysis from your statistics course, but you can also google “regression analysis” to review. We will do this regression analysis in Excel but if you are more comfortable using the “R” stats package you can go this route.

In Excel, go to Insert-Charts and select the scatter plot and just the version with a bunch of dots but no connection of the dots. In the middle of your chart area, right-click and go to Select Data. In the Select Data Source window, go to Legend entries-Add. In the Edit Series window, give your chart a Series name. Then in the Series X values click on the little up arrow and drag over all of the NDI values. Then for the Series Y values click on the little up arrow and select all of the “green stuff” values. Then click OK, the OK again in the Select Data Source window. You should get something that looks like this.

You can pretty this chart up by adding axis labels.

This data looks pretty darned good. There seems to be a strong relationship between the NDI values and the percent coverage of eelgrass and algae. But we’d like to come up with an equation to describe this relationship and we’d like to quantify the strength of this relationship using an R-squared value. And what is an R-squared value? Look it up.

To generate this equation in Excel, all you need to do is right-click on the points and go to Add trendline. This brings up the Format trendline dialog box. You can choose between linear, logarithmic, polynomial or moving average. And down at the bottom you should check the Display Equation on chart and Display R-squared value on chart boxes. Try each of the trendline options to see what gives you the best fit to your data. After deciding on the best one, you should take a screenshot of the figure to include it in your lab report. And, you will need the equation for the next step!

Using your model: Back in ENVI, load the July 14 NDI image that you generated from the 50 cm resolution image. Now go to Band Math and use the equation that you generated in Excel to create a percent vegetation cover layer. After doing so, you can create a raster color slice to define a few % cover categories. I’d suggest perhaps 6 categories like:

<10%

10-20

20-40

40-60

60-80

80-100

For July 14, my result looks like this:

Do the same thing for your July 17 NDI layer.

OK now we are really only interested in the veg cover on the mudflat and the adjacent subtidal areas. So, I’ve created a mask to enable you to eliminate all of the upland from your veg cover layer. This is the “mud_clip” layer in the “eelgrass” folder. Open this layer and take a look. You will note that all of the upland has a value of 0 and the mudflat and adjacent subtidal areas have a value of 1. You know this trick. Go into Band math and multiply each of your vegetation cover layers by the mud_clip layer. However, as you do so, you might want to convert your floating point vegetation cover values to signed integer values. To do so in Band math, use this equation:

Int(b1*b2)

Fix(b1*b2) (correction added 3/5)

With your result as integer, rather than float (decimal values) it is easy to go to Quick Stats and get the number of pixels with each value. You can then export this to Excel and calculate the % of the study area in each % cover category in your image.

After doing so, my July 14 veg cover layer (with the same raster color slice applied) looks like this.

Do the same mud_clip for the July 17 vegetation cover layer.

You can use the data from Quick stats (exported to Excel) calculate the % of the mudflat in each of the % cover categories in your figure.

Finally, we want to quantify the change in vegetation cover between these two dates to see the impact of a busy weekend (associated with the first weekend of recreational crabbing) at this boat ramp. You know how to do this. Just use Band math and subtract the July 14 veg cover layer from the July 17 veg cover layer (and used the versions clipped to just the mudflat). Areas that lose veg cover will have negative values and areas with an increase in veg cover will have positive values. I used these categories

My result looked like this

As you can see, there was quite a bit of loss of vegetation cover during this busy weekend.

That’s it: Prepare a lab report for this exercise. In addition to the stuff above, you might want to use quick stats to prepare tables showing the % of the mudflat that is in each of the categories that you have illustrated in your figures.

Do I REALLY need to write another lab report? Well, maybe not. This is the sixth lab exercise for which you could write a lab report. Your grade for the lab portion of the class will be based on your best FOUR lab report grades. So if you are happy with your grades so far, you do not need to write a lab report for this exercise. But if you’d like to try to bring up your lab report average grade a bit, this is a chance to do so. It is entirely up to you.

As always, the report for this lab, if you choose to write one, will be due on March 12, if you are in the TR lab or 13 if you are in the WF lab..