Help with Landsat Data

695
6
12-06-2013 04:35 PM
EricBrossman
New Contributor III
Hello:

I am attempting to do a Barn Owl habitat selection study utilizing Landsat data.  I have the flight paths and locations of the owls plotted.  I want to clip and buffer the flight paths out of the Landsat image/data.  I want to be able get a percent coverage of each habitat type within the owls' buffered flight paths.  The problem is I cannot simply get the information from just the landsat image.  I need to be able to get the information on the land cover when I click on the portion of the map inside the buffer zone(using the Identify Tool).  I'm hoping this will also let me get the information on an attribute table and be able to calculate the percent coverage for further analysis.  Has anyone ever utilized Landsat data for a similar purpose?
0 Kudos
6 Replies
JosephineHorton
New Contributor
I'd like to help you with this, but I have a couple of things I need you to clarify for me first.  I am familiar with working in areas of the US, if you are studying an area outside the US some of my suggested sources below may not be as useful.

1.  You say you are doing a " Barn Owl habitat selection study utilizing Landsat data".  Does this mean you are trying to determine what type of land cover/land use (crop field, grassland/prairie, woodland, developed/residential, etc.) the Barn Owls utilize or prefer?  If you are not looking at the land cover/land use, what characteristic(s) are you evaluating (proximity to roadways/development, size of range, etc.)?

2.  If you are looking at land cover, you are not going to be able to get that information directly from Landsat imagery, you will need a processed land cover product (either created by you or you can use a publicly-available product like the National Land Cover Database (although it's a few years out of date) or check to see if your state/country/region has a land cover map available.  The Landsat imagery will just have a digital number (indicating brightness) for each pixel in the standard "raw" format you would get from the USGS, not any specific vegetation information.

3.  If you are not looking for land cover information, can you clarify a bit more what exactly it is you are trying to extract from the Landsat data?  There is a lot of potential derived information that can be made from Landsat images, but often it requires processing the images vs. being available in the original 'raw' data.

Good luck!  Studying natural habitat and species selection processes is challenging but very fun and rewarding in my opinion.  (Although as map nerds our ideas of 'fun' can get us some pretty weird looks at parties...lol).
0 Kudos
EricBrossman
New Contributor III
FYI, I agree. If you can help me, you would be very popular if you and I were at this party.  I copied your questions and added my answer to hopefully get the ball rolling.


1. You say you are doing a " Barn Owl habitat selection study utilizing Landsat data". Does this mean you are trying to determine what type of land cover/land use (crop field, grassland/prairie, woodland, developed/residential, etc.) the Barn Owls utilize or prefer? If you are not looking at the land cover/land use, what characteristic(s) are you evaluating (proximity to roadways/development, size of range, etc.)?

Yes, I am interested in determining what types of landcover the Barn Owls appear to be selecting.  I want to clip the owls positions and flight paths and buffer those clips at different scales(30 m, 300m, 1,000m, etc) to determine proximity/avoidance to certain landcover types(Ex. grassland vs. urban developed areas)


2. If you are looking at land cover, you are not going to be able to get that information directly from Landsat imagery, you will need a processed land cover product (either created by you or you can use a publicly-available product like the National Land Cover Database (although it's a few years out of date) or check to see if your state/country/region has a land cover map available. The Landsat imagery will just have a digital number (indicating brightness) for each pixel in the standard "raw" format you would get from the USGS, not any specific vegetation information.

Yes, I found out after banging my head against the wall that I cannot get the information directly from the image.  I have been trying to use the National Landcover Database since the data collected from the database occurred during the same time the data on the owls' positions were collected.  I just don't know how to go about getting finding the data and getting it into GIS.  I want to find a way to get the percent coverage of each landcover type within the clipped section of the owls' flight paths but I have had no luck in finding a way to do this.  My advisor said I needed data associated with the pixels(so when you click on the landcover with the Identify tool, it identifies the landcover type).  I found some data from the 2006 Landcover Database in the ArcGIS online data but of course I was not able to make the raster/vector conversion with that data source.  He suggested looking at GLOVIS but it seems like those are simply just images.  Also, I'm unfamiliar how to create a processed landcover product per your suggestion since I have had little luck finding what I need from the National Landcover Database.

3. If you are not looking for land cover information, can you clarify a bit more what exactly it is you are trying to extract from the Landsat data? There is a lot of potential derived information that can be made from Landsat images, but often it requires processing the images vs. being available in the original 'raw' data.

See above response.  It seems like there is a lot of raw data, I'm just confused/overwhelmed in trying to figure out how to refine it.
0 Kudos
JosephineHorton
New Contributor
Alright, lets see if we can get you started on this.  For future reference, you are correct that GLOVIS is generally just the repository of 'raw' satellite imagery that the USGS makes available, there aren't any land cover files out there that I'm aware of.

-You can get a raster version of the NLCD 2006 from http://www.mrlc.gov/nlcd06_data.php.  You can download the file (I think it's compressed and needs to be unzipped) and then bring it into ArcGIS.  This will be a national map, so you'll have to clip it to your study area(s), but it sounds like you were going to be doing something like that anyway so not a big issue.  This file is annoying for several reasons.  First of all, the download is an .img file type, which ArcGIS doesn't clip to, so you need to convert it to a GRID file (if I remember correctly, it's been a while since I had to download this one.  Tried to test it to refresh my memory and it was taking forever, so didn't finish that, sorry).  Also, the file does not have the land cover classification built in to the attributes in a logical way (in my opinion).  The attribute table lists 255 OID values, most showing a 0 for the "Count".  The ones that have "Counts" listed coincide with the Land Cover classification codes in the legend (found at http://www.mrlc.gov/nlcd06_leg.php).  For example: OID 11 has a count of 464840229 (or something like this), indicating that 464840229 pixels of the map are classified as Class 11-Open Water.  This file is also at a 30-m resolution, so keep that in mind when you are calculating the area of each class present in your areas of interest and determining your buffer sizes.

-Once you have this file you can add it to your working ArcGIS map document that has your owl locations/flight paths/roosting sites/etc. (I'm assuming you have those as shapefiles), create your desired proximity buffers, and clip the NLCD to those buffers.  You should then be able to access the attribute table for the newly clipped files to see how many pixels are present in each land cover class as well as how many pixels are present in the entire buffer/clip/etc.  For example, you create a 3000m buffer around the roost site of Owl A, clip the NLCD to that buffer and look at the attribute table.  You see that there were a total of 4000 pixels in that buffer (these numbers are just made up, I didn't actually create any buffers)m 351 pixels of which were in class 41-Deciduous Forest, which allows you to determine the square meters of that habitat type present in that buffer (351 pixels that are each 30 square meters) and by extension the percent of that buffer that is that habitat type. 

This is just a generalization of the process, it's been a while, sorry if I miss a step here and there.  Let me know if you have problems getting it to work or if I misunderstood what you were trying to do.

-In regards to making your own land cover product, that's quite a bit more involved and is best done by someone with experience in aerial imagery interpretation/remote sensing.  If you are confident that you can identify the different habitat types in your study area from satellite imagery (for example, if you have traveled in your study areas extensively to the point where you are confident you can identify grassland vs. hay field vs. deciduous forest) and want to try that, ArcGIS does have some tools for image classification of Landsat imagery.  I, however, have not used any of them.  I have another software package that I use for my land cover mapping and image interpretation.

Good luck and keep me posted if you have more questions.
0 Kudos
JeffreyEvans
Occasional Contributor III
I think that it would be prudent to take stock of your question and what data you require. It is fairly clear that you should focus on existing landcover classifications and not try to produce your own from raw satellite imagery. Keep in mind that NLCD is fairly coarse in its classification schema (very broad classes that may not be relevant to the scale of your analysis). There are a few alternatives that you can explore including the USGS-GAP vegetation classifications which represent considerably more detail in the defined classes (down to species level). Since you have not defined a specific hypotheses that you are testing or a geography it is difficult to make more specific data recommendations. 

One issue that I see in your proposed analysis is that intensity (percentage) only explains so much. Configuration is an important consideration in understanding ecological process and habitat use. To illustrate this think of a simple circular buffer and a binary habitat [0,1] indicator. If you assign 50% of the habitat data to one side of the buffer it will tell you something very different than if that same 50% is randomly distributed through the area. In the first case the pattern is consolidate whereas the second example indicates a perforated habitat. Species respond in very different ways to these two differing configurations. Incorporating configuration into your analysis will provide you with much more insight into habitat utilization and connectivity. You can calculate landscape metrics using the software FRAGSTATS. I would recommend calculating moving window metrics which will produce rasters for each metric at the specified scale. You can then mask the results based on a vector buffer of your owl data, then calculate zonal statistics on each metrics. This will allow you to look statistical variation within each buffer while avoiding edge effect.
0 Kudos
EricBrossman
New Contributor III
Thank you for your input.  I am attempting to do a compositional analysis which compares the available habitat to the owls compared to the habitat they utilize.  The hypothesis is that the owls will utilize each habitat they encounter in proportion to the next(ie, they do not select one habitat over the other) compared to the alternate hypothesis in which they select particular habitats disproportionately compared to the others(selecting one or more habitat more than others)not sure if you have any experience doing this type of analysis from a GIS standpoint?  Are you suggesting that this type of analysis cannot be done within the buffers at different scales without incorporating the configuration method you described?  I'll admit, I'm very new to this and I am struggling greatly with this project.
0 Kudos
JeffreyEvans
Occasional Contributor III
I am not saying that you cannot perform your analysis with a buffer approach. If you just want proportion of a given habitat then it is a trivial matter to convert each of your landcover classes to binary rasters and then use zonal statistics to calculate associated proportion for each buffer. 

The point that I am trying to convey is that there is an entire field in ecology that addresses questions of pattern and process. Compositional analysis is often misleading, and not commonly used in ecological and wildlife modeling, because configuration (spatial arrangement) and other spatial characteristics of habitat either independently drive or interact with proportion to drive habitat utilization. With avian species, amount and complexity of habitat edge are often a critical component that defines characteristics such as forage availability. Metrics that indicate habitat homogeneity and spatial arrangement are also quite useful. I would highly recommend reading a basic book on landscape ecology and performing a literature review on landscape metrics.            

I have modeled and published work on Strix occidentalis habitat utilization and can say that configuration and edge is a critical component for Owls. Another observation is that, given your stated hypothesis, the landcover classes defined in NLCD will be woefully inadequate to describe a use verses availability question. You will need a detailed vegetation classification.

There are very few datasets that come to mind with this level of detail. As I mentioned previously the GAP program performs regional vegetation classifications. If working on US Forest Service lands the common stand exam databases represent polygons based on field surveys and aerial photo interpretation. There are sometimes regional datasets that are available (e.g., CoMAP Colorado vegetation classification) but you will need to ferret these out on a case-by-case basis and they may not entirely cover your study area.

It is critical to match a given classification to your actual hypotheses and not just use whatever data is conveniently available. This is how many studies such as this go off track and present nonsensical results. I have rejected many manuscripts because of unsupported inference based on mismatched scales.              

Here is an overview of landscape metrics, by Kevin McGarigal, that describes what the different types of metrics are intended to describe.
http://www.umass.edu/landeco/teaching/landscape_ecology/schedule/chapter9_metrics.pdf
0 Kudos