Harmful Algal Blooms – Part 6: Finally, a Project

This semester, I’ve been working with Dr. Kenton Ross, the national science advisor at NASA DEVELOP, to understand the spectral properties of Alexandrium monilatum. I am using aspects of this work as my term project for GIS 295 and GIS 255.

In Harmful Algal Blooms – Part 5: Trouble in Data Land, I discussed the challenge of completing a mapping project when the data has lost its geospatial reference.  However, we were able to obtain approximate locations for each hyperspectral scene by estimating pixel size and lining up the time field in the hyperspectral data  with the time field in the the flight path data set. You can see the scene center lines in black on the image below.

Scene Center lines

The water in this map looks unusual. NDVI is a band ratio index used to indicate vegetation. I used ((Landsat 8 band 5 – band 4)/ (band 5+band 4)) to mask out the land  in ArcGIS. I then clipped out the unmasked water and used false color imagery (bands 6, 5 and 4) to enhance different features of the water.  The water on the right side looks thick and yellow because it is very turbid.

Once I knew where each scene was located, I analyzed each hyperspectral image individually. This is York scene 5.

20150817_YK5

The first thing I noticed was that any detail was hard to see. That’s because one side is in shadow. So, I clipped out the black edges and the bright, overly illuminated right side of the image. I also reduced my number of bands from 283 to 13 carefully selected bands in order to reduce noise and maximize the spectral signal. The resulting image is in the slide below.

Slide7

The deep red swirl is the algal bloom.

I used bands 146 (710 nm), 128 (665 nm) and 85 (559 nm) to create a NIR, red, green false color composite to highlight areas with high chlorophyll.  That’s the second image in the slide.

High chlorophyll should appear red in these images, but the areas of intense blooms appeared bright yellow-green. This is because there is also a red pigment in the blooms that gives them their color.

The third image is estimated chlorophyll. I used the  ratio of band 146 (710 nm) over band 128 (665 nm) as a proxy for chlorophyll-a. Red and yellow indicate high chlorophyll, while dark blue indicates low chlorophyll. You can see that despite the red color, the algal bloom has plenty of chlorophyll.

What about the third image? I used ENVI to run an unsupervised classification. This means that ENVI sorted the pixels in the image into five groups according to their spectral properties. I then used class statistics to obtain the spectral signature of each group. That’s the big image on the left of the slide.

I repeated this procedure on seven images from the York River, four images from the James River, and one image from Mobjack Bay. The signatures in the York River were distinctive enough that we believe we are on the right track to find the spectral signature of Alexandrium monolatum.

For my GIS 295 class project, I created an web app which shows the imagery and spectral signatures at each scene. If you are a member of Northern Virginia Community College GIS group, you can access the app here. Everyone else will have to wait until we are ready to make the app public.

References:

Moses, Wesley J.; Gitelson, Anatoly; Perk, Richard L.; Gurlin, Daniela; Rundquist, Donald C.; Leavitt, Bryan C.; Barrow, Tadd M.; and Brakhage, Paul, “Estimation of chlorophyll-a concentration in turbid productive waters using airborne hyperspectral data” (2012). Papers in Natural Resources. Paper 313. http://digitalcommons.unl.edu/natrespapers/313

Advertisements

Harmful Algal Blooms – Part 5: Trouble in Data Land

If you’ve read Harmful Algal Blooms, Part 4, you know that I had developed a plan to obtain the spectral signature of the Alexandrium monilatum, a toxic dinoflagellate that causes harmful algal blooms in the Chesapeake Bay watershed, from the hyperspectral data that was collected August 17, 2013. I wanted to use spectral signatures to map the extent of harmful algal blooms in the James and York Rivers. However, lots of data doesn’t always mean good data.

The hyperspectral data was collected using a sensor that was mounted on a NASA airplane. The angular cone of visibility detected by a sensor at a given time is called the Instantaneous Field of View (IFOV). The size of the IFOV determines the resolution or minimum size of a pixel.

In this image from Natural Resources Canada, area A is the IFOV and area B is the area on Earth’s surface that that can be seen at a given time (B=A*C).

IFOV2

Area B, the area that can be sensed at any time, depends on many factors, including the altitude of the plane and the angle of the sensor. This illustration from Natural Resources Canada illustrates the effect of angle of view.

IFOV 1

 

During the August 17 data collection flight, the part of the sensors internal navigation system that measures the plane’s attitude or angle failed. This meant that we could not determine pixel size. It also meant that location data was not available for the hyperspectral data.

GIS stands for Geographic Information Systems. The term “geographic” refers to location.  Without geographic coordinates, we could not accurately place our data on a map. What could we do?

Normally, one would georectify the data using by lining up ground control points. Ground control points are known locations on the ground. they must be small, unchanging and easy to recognize. But, how do you find ground control points in a picture of water?

YK6

Fortunately, we had a .kml file of the flight path, which listed geographic coordinates and times, and a few images with features other than water like this image with a large Navy dock.

20150817_YK7

We were able to use measurements of the dock to calculate pixel size. We calculated the pixels to be about 2 meters long (a long track) and 3 meters wide (across track).

Dr. Kenton Ross, the national science adviser for NASA DEVELOP,  was also able to use time signatures from the flight path file to determine where the plane was at a given time. He matched these times to the time variable of the hyperspectral images and was able to estimate approximate geographic coordinates for each of the images. The seven hyperspectral image sites for the York River are shown below.

Path

However, I still had to give up a big part of my project design. When planning the project, I had forgotten one very important fact: water flows. Unlike ground control points, water does not stay in place over time.

On the image above, you can see a black squiggle. This is the path of the data flow cruise. It overlaps two  of the hyperspectral images shown above in space. However, since the chlorophyll samples were not collected at the exact same time (although it was within a few hours) as the hyperspectral images, the two data sets do not overlap in time. Because there is a time difference, the water moved. This might not make a big difference at 30 meter resolution, but at 2 to 3 meters, it could be a big deal.

There was another problem. The shape file with the locations lost its headers during processing. The ASCII file had to be edited in order to move the locations into ENVI.

Stay tuned for the final post of this series, Harmful Algal Blooms, Part 6, to learn how I (hopefully) resolve these issues and finish my project.

Harmful Algal Blooms – Part 4: What is a Spectral Signature?

The human eye is a light sensor. We can see because the objects around us emit or reflect light at wavelengths that our eyes can detect.

Remote sensing detectors work in the same way. Sunlight is reflected from Earth’s surface. Satellite sensors detect that light and create images. But, satellite and aerial sensors are a lot more sensitive than our eyes.

69904main_RemoteSnsg-fig2

Our eyes can only detect light with wavelengths from about 390 nanometers to 700 nanometers. This range is called visible light and is shown as a rainbow in the NASA image below. But, satellites can detect a much wider range of wavelengths, depending on the sensor. Landsat satellites detect light from visible blue (450 nm)  to the thermal infrared (1251 nm).ems_length_final

All objects reflect, absorb and transmit light. But, some types of materials reflect and absorb certain wavelengths of light in very characteristic ways. So, these types of materials can be distinguished from each other based on the differences in reflectance, or the differences in the pattern of wavelengths that are detected by the sensor. The pattern is called a spectral signature. The following image from NASA shows spectral signatures for some common Earth materials.

(If you click on the picture, it will take you to a NASA site that explains light and remote sensing in more detail.)

spectral-signatures.png

Harmful algal blooms also have spectral signatures. This is the spectral signature of Cochlodinium polykrikoides, one of the species that causes HABs in Virginia (Simon and Shanmugam, 2012).

Cochlodinium

C. polykrikoides is known to absorb light at 555 nanometers and reflect it at 678 nanometers.  These wavelengths correspond to the high and low peaks in the image above.

There is a lot of research about C. polykrikoides, but very little is known about the spectral signature of Alexandrium monilatum. I decided that I would use the hyperspectral data that was collected on our Golden Day of Data Collection, August 17, 2015 to identify blooms of C. polykrikoides and I would try to decode the spectral signature of Alexandrium blooms.

I had a plan. I would map the chlorophyll measurements from the data cruises on the Landsat image to identify the areas with the most chlorophyll. I believed this would correspond to the blooms. I would join these areas to the hyperspectral images in order to obtain spectral signatures. Then, I could map the extent of each bloom.

It was a plan, but you know they say about plans….

Stay tuned for Harmful Algal Blooms Part 5: Trouble in Data Land.

 

 

Harmful Algal Blooms – Part 2: Monitoring Harmful Algal Blooms

In Harmful Algal Blooms – Part 1, I discussed what a harmful algal bloom is and why we care about blooms.  I wrote about the dangers that HABs pose to public health and the economy and explained why it is important to monitor and study HABs.

Members of the Virginia Harmful Algal Bloom Task Force use a combination of fixed stations, continuous sampling, and periodic dataflow cruises to monitor water quality in the Chesapeake Bay watershed. This map shows monitoring stations in Virginia.

Sorce: Virginia Institute of Marine Science
Sorce: Virginia Institute of Marine Science

Data for each of these stations is available at the Virginia Estuarine and Coastal Observing System (VECOS) website.  It looks as if Virginia’s rivers are well-covered, but if you go to the site and take a look at this data, you will notice that many of these sites are no longer active due to funding cuts.

Current monitoring consists of fixed stations and periodic dataflow cruises operated by the Virginia Institute of Marine Science and Old Dominion University. Most of this monitoring focuses on the James River and the York River. Sanitation districts, like the Hampton Roads Sanitation District, do automated continuous sampling in their service areas. This monitoring is on-going, but it doesn’t provide a complete picture of HAB activity.

The first obstacle is that processing all these samples takes time. Harmful algal bloom species have to be separated out of water samples that contain hundreds or thousands of other microorganisms through a complex series of DNA tests. Information isn’t available until weeks or months after a bloom occurs. By then, it may be too late to determine what factors contributed to the bloom.

The second problem is that using dataflow cruises for real-time monitoring is expensive. So, boat cruises are restricted to areas of high concern. This means that blooms in other parts of the Chesapeake Bay watershed may be overlooked.

This makes it difficult to get the environmental and water quality information needed to understand and predict HAB occurrences. It also makes it difficult to get real time information about HAB activity in the Chesapeake Bay watershed.

What if there was a less expensive option?

Last summer, I participated in the NASA DEVELOP program at the Patrick Henry Building in Richmond. Our team, Cassandra Morgan and I, worked on a method to monitor harmful algal blooms using satellite data.

Sara Lubkin (NASA DEVELOP), Todd Egerton (ODU), Wolfgang Vgelbein (VIMS), Kimberly Reece (VIMS), Cassandra Morgan (NASA DEVELOP)
Sara Lubkin (NASA DEVELOP), Todd Egerton (ODU), Wolfgang Vgelbein (VIMS), Kimberly Reece (VIMS), Cassandra Morgan (NASA DEVELOP)

Satellites can take pictures of HABs like this Landsat 8 true color image.

study_area_81715
Landsat 8 08/17/2015

While, you can see the bloom in this picture, it’s hard to determine exactly which areas are in the bloom. However, phytoplankton uses chlorophyll-A to harvest the energy of the sun. VIMS and ODU detect harmful algal blooms by measuring levels of chlorophyll-A. What if we could detect chlorophyll-A in the water using remote sensing? We could then use chlorophyll measurements as a proxy for HABs.

NASA’s MODIS Aqua satellite collects information about water, including chlorophyll-A levels. Chlorophyll-A maps are available at no charge from NOAA CoastWatch’s East Coast Node.

This is a MODIS chlorophyll map for the same day.MODIS

You can see that there are high levels of chlorophyll in the James River, Upper Chesapeake, Potomac and Mobjack Bay.

The problem is that MODIs aqua chlorophyll products have a 1.4 kilometer resolution. So, they don’t give a lot of detail – especially in narrow rivers like the York.

Landsat 8 has a 30 meter resolution. But, there are no publicly available Landsat chlorophyll-A products. It was Cassandra and my job to create this product.

We started by  downloading Landsat 8 images (Path 14, Row 34) from May through September 2011-2014. We looked for images without too much cloud cover. We masked out the land and the clouds and  filtered these images through a 1.4 km moving window to match the MODIS resolution. We chose the day with the best overlap between MODIS and Landsat and joined MODIS the chlorophyll-a values to each smoothed, masked Landsat band. We also added bathymetric measurements

Once we had this data in one table, we were able to export it to “R” and run a series of regressions. We tested 78 separate equations. The best five equations were used to create tools using ArcGIS model builder (r-squared values .57 to .62). Here is an image of chlorophyll in the Bay created with one of those tools.ChlJuly3

We were now able to show chlorophyll-A estimates at 30 meter resolution.

Cassandra and I created a preliminary model for chlorophyll-A during our summer term. This term, NASA DEVOLOP teams at Langley Research Center and Wise County are testing the model on additional dates and validating the equations using VECOS water quality data. They are also creating easy-to-use ArcGIS tools that will allow VIMS and ODU to quickly assess the extent of algal blooms in the Chesapeake Bay.

These tools will save time and money by allowing VIMS and ODU to better target their monitoring efforts. The tools will also allow researchers to collect information about environmental factors associated with HABs.

To learn more about NASA DEVELOP and this project, check out our video or our story map.

Next Week: Harmful Algal Blooms – Part 3: The Golden Day of Data Collection