Final Project: Quantifying Tree Canopy Coverage Using Supervised Image Classification

Quantifying Tree Canopy Coverage Using

Remotely-Sensed Imagery for Urban Forestry Management

Background

In recent years, Central Florida’s population has increased at an alarming rate. In fact, in an article published by Forbes earlier this year entitled, “Full List: America’s Fastest Growing Cities 2017,” Central Florida (which includes the Orlando-Kissimmee-Sanford Metropolitan area) was ranked second in the country with a population growth of 2.95% in 2016, and a projected population growth of 3.14% in 2017 (Sharf, 2017). This increase affects community and housing development, which involves the construction, mass grading, and excavation of vegetative lands. These rapid changes to the landscape has brought awareness to environmentalists and city planners of local governments and municipalities, of the need to implement an urban forestry management plan.

Urban forestry is defined as the caring, planning, maintenance, and management of tree populations in urban infrastructures, while promoting their health and economic benefits in the human environment (Wikipedia, 2017). Some of these evidence-based benefits include, but are not limited to: cleaner air, improved water quality, reduced utility bills, and thermal comfort – which are directly associated with reduced stormwater runoff, irradiance absorption, transpiration cooling, reduced pollution, increased infiltration, and more (Livesley et. al, 2016).

Most recently, the Parks and Public Lands Department at Osceola County Board of County Commissioners (located in Kissimmee, Florida) met with consultants to begin establishing an urban forestry management plan. I work as a GIS Technician at this organization. When I found out about the project I became immediately intrigued as I was simultaneously learning about image processing techniques for land use/land cover (LULC) classification in our Photo Interpretation and Remote Sensing graduate class. The Urban Forester Specialist at Osceola County is currently using an open-source platform from the U.S. Department of Agriculture (USDA) Forest Service and other contributors called “i-Tree Canopy” to estimate tree coverage in the county’s urban growth boundary. The tool is incredibly neat, useful, and free; it creates outstanding statistics such as atmospheric carbon value estimates and error scores (USDA, 2017a). However, there are some drawbacks in using the platform. For example, it uses Google Maps aerial photography for the user to manually assess cover type for thousands of generated points in the study area; and, integration of i-Tree Canopy with ArcGIS products for further spatial analysis and creation of cartographic outputs does not currently exist.

For my final project, I wanted to challenge myself by attempting to use my newly acquired skills from class by applying automated image processing techniques in a real-world scenario at work.

Project Goal

The goal of this project is to perform a supervised image classification on remotely-sensed imagery within Osceola County’s urban growth boundary for the purpose of quantifying tree canopy coverage to assist in a department’s urban forestry management plan efforts. Additionally, this project aims to showcase the benefits of using a) higher-resolution satellite data, b) automated processing, and c) the world’s leading geospatial data system: ERDAS Imagine.

Imagery Data

The remotely-sensed data used for this project is a Landsat 8 OLI/TIRS (Operational Land Imager/Thermal Infrared Sensor) C1 Level-1 from Landsat 8 satellite. The imagery is dated December 2, 2017, and was downloaded through the U.S. Geological Survey’s Global Visualization Viewer (GloVis). It has a total of 11 bands (coastal aerosol, blue, green, red, near infrared, short-wave infrared 1, short-wave infrared 2, panchromatic, cirrus, TIRS 1, and TIRS 2). The OLI multispectral bands have a 30-meter resolution, the OLI panchromatic band has a 15-meter resolution, and the TIRS thermal bands have a resampled 30-meter resolution.

Pre-Processing Techniques

After the download, the data needed to be reformatted and converted from a compressed format (.tar.gz) to TIFF, and then to .img, before applying pre-processing techniques. To do this, I unzipped the .tar.gz file first using the “7-zip” option, and then unzipped the file it created again. The second unzipping created a new folder containing the 11 TIFF images, one for each bandwidth. Then, I imported TIFF bands 1 through 7 only into ERDAS Imagine and converted them to .img. Then, using the Layer Stack tool (under Raster tab > Resolution group > Spectral) I brought in the 7 .img bands to create a multispectral image. Finally, I created a subset of an area within Osceola County’s urban growth boundary using the Inquire Box tool.

Processing Techniques

My project’s goals were to create a supervised image classification (in ERDAS Imagine). To begin this, I created spectral signatures and AOI (Area of Interest) features by using the Grow tool and Polygon tool under the Drawing tab, but first I needed to make sure to add a new, blank AOI layer to do so. I decided to create 5 signatures: Trees, Urban, Water, Wetlands, and Other Vegetation.

While creating my signatures, I made sure to adjust the Spectral Euclidean Distance and Neighborhood settings when necessary so that the P, I, H, and A columns were all checked off. During this process, I also recognized and eliminated spectral confusion between signatures in bands by using both the Histogram and Mean Plot tools and kept track of the bands that had the greatest separation among signatures. I found that bands 2, 3, 4, and 6 had the greatest separation. To create my final image, I ended up using bands 3, 4, and 6.

To run the classification process using these bands, I did the following: on the Signature Editor window > Edit > Colors > Approximate True Colors > Set Signature Colors. The final steps included: applying a Parametric Rule of "Maximum Likelihood" to create both an Output File and an Output Distance File; merging classes to new, unique values using the Recode tool under Raster tab > Thematic, as well as adding class name and acreage columns to the attribute table. After all changes were saved, I imported my final working image into ArcMap to finalize my cartographic design product.

Outcomes

I was semi-successful in conducting a supervised image classification from high-resolution satellite imagery using a reputable and powerful software. From the subset I created, I can quantify the acreage of tree canopy coverage versus the other classifications. Unfortunately, where I failed was in not being able to combine two imagery downloads comprising of the urban growth boundary in Osceola County. The satellite image scene cuts across the county, which made me unsuccessful in acquiring the canopy coverage measurement as intended. But that’s okay. I’m close. I now know what to focus on the next round. Also, I should have created more signatures than I did because I noticed that there are areas in my classification that shouldn’t be what they were classed.

Choosing a topic of my choice was challenging but rewarding because it made me revisit all my notes, readings, process summaries, lab instructions, and blog posts to remember the sequence of steps. It’s important to do things systematically to avoid doing double work.

Map

The map below depicts a land cover classification intended to highlight tree canopy coverage in Kissimmee, Florida – which was done using an automated processing technique called “supervised image classification” with remotely-sensed satellite imagery. This map offers a glimpse of what could become a larger project in establishing an urban forestry management plan in Osceola County, Florida.

  

Additional Screenshot

 

References

Fulfrost, B. (2017). Lab instructions 6, 7, and 10 [Lecture notes]. Retrieved from https://elearning.uwf.edu/

Livesley, S. J., McPherson, E. G., & Calfapietra, C. (2016). The urban forest and ecosystem services: Impacts on urban water, heat, and pollution cycles at the tree, street, and city scale. Journal of Environmental Quality, 45, 119-124. Retrieved from https://www.fs.fed.us/psw/publications/mcpherson/psw_2016_mcpherson001_livesley.pdf

Sharf, S. (2017). Full list: America’s fastest-growing cities 2017. Forbes. Retrieved from https://www.forbes.com/sites/samanthasharf/2017/02/10/full-list-americas-fastest-growing-cities-2017/#73ac1c783a36

USDA Forest Service. (2017a). i-Tree Canopy (v6.1) [Web software application]. Retrieved from https://canopy.itreetools.org/

USDA Forest Service. (2017b). i-Tree: Learn about i-Tree. Retrieved from https://www.itreetools.org/

USGS. (2017a). Earth Resources Observation and Science (EROS) Center Archives. Retrieved from https://www.usgs.gov/centers/eros/science/usgs-eros-archive-products-overview?qt-science_center_objects=0#qt-science_center_objects

USGS. (2017b). What are the best Landsat spectral bands for use in my research?. Retrieved from https://www.usgs.gov/faqs/what-are-best-landsat-spectral-bands-use-my-research?qt-news_science_products=0#qt-news_science_products

Wikipedia. (2017). Urban forestry. Retrieved from https://en.wikipedia.org/wiki/Urban_forestry