Brian Bunker and Dr. Ryan Jensen, Geography Department

Main Text

Urban forests are trees and other woody vegetation growing in an urban environment. Urban forests have been shown to have economic, aesthetic, and ecological benefits (Gatrell & Jensen, 2002; Nowak & Crane, 2002). Monitoring the health of Urban Forests is important to maintain a healthy environment for humans in their principal habitat. Our research focused on developing methods to monitor urban forest biomass using remote sensing data.

During the study, biomass was measured and correlated to previously collected hyperspectral imagery data. Hyperspectral imagery can be thought of as a digital camera capable of capturing an image in 248 bands which cover electromagnetic wavelengths from 391nm to 965nm. A typical digital camera image can be thought of a 3 bands of blue, green, and red. These bands would cover wavelengths from 400nm to 700nm. Since the hyperspectral image has so many narrower bands, we are able to more accurately measure electromagnetic reflectance from the vegetation in an urban environment when compared to a standard digital camera image or similar multispectral image.

We collected the field biomass data over 5 days during the summer of 2010. We collected data during similar times for each field day. In preparation to collect the sample, we defined the study area and identified the number of trees to include in the sample. The study area was bounded on the west by University Avenue, the east by 900 East, the south by Center Street. The northernmost limit was the Smoot Administration building on BYU campus. This study area provided a good mix of large apartments, row houses, single family units, parks, and campus geographies. We determined that a sample size of 100 trees was sufficient to build and test our predictive model – 75 trees were used to build the model and 25 trees were used to test the predictive model for goodness of fit.

We calculated individual tree biomass values by randomly selecting a tree and identifying the species. An allometric biomass equation for the species or a similar species, we would measure the diameter of the tree at breast height, a measurement known as DBH. We also recorded the geographic coordinates of the tree, its species, and marked its location on an aerial photograph for future reference. The DBH measurement was input into previously published allometric equations to calculate above ground biomass of each tree.

Next spectral reflectance values were extracted from the hyperspectral imagery using the GPS coordinates of each tree. There were two problems with this approach: 1) the image contained random distortions which placed the geographic coordinate away from the correct tree, and 2) the GPS unit used to collect the geographic coordinates varied in its accuracy during use. To overcome these challenges, we used the annotated aerial image that was marked with each tree in the sample as a reference. Once all trees were correctly identified, the spectral characteristics of all 100 trees were exported to a text file. The exported data contained a spectral reflectance value for each of the 248 bands for each of the 100 trees in the sample. We were then able to import the data into Excel and associate calculated biomass values for each tree. Then we separated the 75 trees to build the model and the 25 trees to test it.

We added vegetation indices and spectral curve calculations for each tree. The indices are standard vegetation calculations which were created by performing basic mathematical operations on reflectance values at pre-determined bands. Indices used were Vogelmann (VOG) VOGa = (R740 /R720), VOGb = [(R734-R747)/(R715+R726)], VOGc = [(R734-R747)/(R715+R720)], red-edge spectal parameter (RESP) RESP = (R750/R710), Gitelson Merzylak index (GMI) GMI = (R750/R700), and Carter Index (CI) CI = (R795/R760). The Rλ notation means the reflectance value at λ wavelength. The spectral curve calculations were obtained by measuring the various areas under each reflectance curve for each tree. The areas under the curve (AUC) used were total AUC391 – 465, wide red-edge AUC (WRE) WRE670 – 770, narrow red-edge (NRE) NRE690 – 740, wide green-bump AUC (WGB) WGB500 – 670, and narrow green-bump AUC (NGB) NGB550 – 600. The subscript denotes the upper and lower bound wavelengths for AUC calculation. Maximum red-edge slope (MaxRES), maximum red-edge slope wavelength (MaxRESλ) and the red-edge inflection wavelength (REIλ) were also calculated and examined. Regression analysis was performed using the calculated biomass values as the dependant variable. Results from each of these cases are found in the following table:

Most of the indices and other reflectance derived calculations were not significantly correlated, however using the bands from 800nm to 965nm did result in a significant R2 value of .826, or 82.6% of the variation in biomass can be explained by the electromagnetic reflectance from 800nm to 965nm.

This project does not end here. I plan on continuing the study and trying more band combinations and reflectance derived variables to improve on the biomass prediction model. This will ultimately result in publishing of the results in an academic remote sensing journal. This project has also encouraged me to pursue remote sensing research at the graduate school level. In fact, I have recently been offered a research assistantship at the University of Arkansas. This offer may not have been made without this ORCA opportunity.

Sources

• Gatrell, J.D. & Jensen, R.R. (2002). Growth through greening: developing and assessing alternative economic development programmes. Applied Geography, 22(4), 331-350.
• Nowak, D.J. & Crane, D.E. (2002). Carbon storage and sequestration by urban trees in the USA. Environmental Pollution, 116, 381-389.