Ecological Archives A017-086-A1

Natalie T. Boelman, Gregory P. Asner, Patrick J. Hart, and Roberta E. Martin. 2007. Multi-trophic invasion resistance in Hawaii: bioacoustics, field surveys, and airborne remote sensing. Ecological Applications 17:2137–2144.

Appendix A. Detailed methods of airborne imaging spectroscopy, LiDAR, bioacoustic recordings, and bird surveys.

(1) Airborne Imaging Spectroscopy

In February 2005, we used the Jet Propulsion Laboratory (JPL) Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) to collect imagery over a 150 km2 area including our field sites. AVIRIS measures upwelling radiance from 400–2500 nm in 224 channels or 10 nm full-width at half maximum (1). The instrument was mounted on DASH-10 turboprop that flew 3.2 km a.g.l. during data acquisition, resulting in a ground instantaneous field-of-view (GIFOV) of 3.2 m at nadir.

We used the ACORN-5 atmospheric radiative transfer model (ImSpec, Palmdale, California) to convert radiance data to apparent surface reflectance and to mask wavelengths from 1344–1408 and 1793-2008 nm, which are dominated by atmospheric water absorption. This resulted in 194 channels of useable data spanning the 400–2500 nm wavelength range. All ACORN simulations used a tropical atmosphere, water vapor retrieval using the 940 and 1140 nm absorption features, and a 250 km visibility (aerosol) setting. Preliminary geo-registration of all data was performed using the inertial navigation system data taken onboard the aircraft. The images were further geo-rectified using digital orthophoto quadrangle maps. Geo-location uncertainty of the final image products was 0.43 m.

The fractional cover of three major surface materials – photosynthetic vegetation (PV), non-photosynthetic vegetation (NPV), and bare substrate – was quantified in each AVIRIS pixel using a fully automated spectral unmixing algorithm designed for use with high-fidelity imaging spectrometer observations (2, 3, 4). PV is comprised of live green foliage; NPV is a combination of surface litter, standing senescent material and woody surfaces exposed to the field-of-view of the imager. Bare substrate is a third surface constituent representing bare soil, ash, and lava. Each of these component endmembers contributes to the pixel-level AVIRIS reflectance


where is the reflectance of each land-cover endmember at wavelength . C is the fraction of the pixel composed of e, and is the error of the solution at wavelength . The second equation indicates that the endmembers are forced to sum to unity. For this study, endmember bundles for PV, NPV and bare substrate (pv(), npv(), s()) were constructed from general field spectral databases collected in ecosystems throughout Hawai'i (4).

Following the canopy cover analyses, we extracted statistics for each 300 m study transect and the surrounding 10 ha area. The transect data were compared directly to field-based measurement of canopy cover (for validation), and the 10 ha areas were then used in comparisons with field-based measures of ecosystem structure and bird community composition (Figs. 1–3).

(2) Airborne Scanning Light Detection and Ranging (LiDAR)

In March 2006, we flew a scanning discrete-return LiDAR system (Optech ALTM-2025) over the region, and collected laser returns at ~0.5 m spacing. The LiDAR first-returns and last-returns were binned in 3.0 m grid cells, and the difference in ranges was used as an estimate of vegetation height. The PV output from the airborne spectrometer data (previous section) was then used to ensure that all derived vegetation heights were associated with live tree or shrub cover. The LiDAR data was then co-registered to the AVIRIS imagery for coordinated analysis of vegetation cover and height of each field site and surrounding 10 ha area.

(3) Bioacoustic Recordings

Recordings were made using a digital audio recorder (722, Sound Devices, LLC, Reedsburg, Wisconsin), two microphones (MKH-30 and MKH-40, Sennheiser Electronic GmbH and Co. KG, Wedemark, Germany), a blimp windscreen (MZW-201, Sennheiser Electronic GmbH and Co. KG), and a microphone tripod. The 722 recorder is capable of recording sound in frequencies between 20 Hz and 22 kHz, and has a dynamic range of 110 decibels (dB). We employed a sampling rate of 44.1 kHz and stored recordings as uncompressed audio files (WAV format). The MKH-30 microphone has a figure-8 (or bi-directional) pickup pattern, and is thus most sensitive to sound to either side of the microphone. The MKH-40 microphone has a cardioid (heart shaped) pickup pattern that offers maximum rejection at the rear of the microphone. By physically co-mounting these two microphones, one on top of the other, and recording with both simultaneously, the resultant pickup pattern is nearly omni-directional. Both microphones have a frequency response of 40 Hz to 20 kHz, a signal to noise ratio of 81 dB, and dynamic range of 121 dB. In combination, the recorder and microphones provided a frequency response of 40 Hz to 20 kHz and a dynamic range of 110 dB.

Prior to collecting audio recordings, a recording of white noise was played at a sound level of 65 dB while the 722 Recorder was programmed to measure the 65 dB of white noise as -30 dB (digital recordings have negative dB values). In this way, the amplitude of our bioacoustic recordings could be compared to a reference sound amplitude, thus allowing for the conversion of amplitude to dB. Decibels are a measure of the difference in intensity of two sounds (e.g., the reference sound and the bioacoustic recording). The conversion of sound level amplitude to dB was calculated using the following equation:

dB = 20*log10(abs(a/ao))

where a is the amplitude of the target sound and ao is the amplitude of the reference amplitude. When a = ao, the dB value = 0, which is the maximum amplitude that can be measured by the recording system without distortion. By forcing the 722 Recorder to convert 65 dB of white noise to -30 dB, the maximum sound level that could be accurately measured by our recording system was 95 dB. 95 dB is equivalent to the sound level of a heavy truck operating 1 m away from the recording device and is thus well above the upper limit of the bioacoustic recordings measured in the field. In this study, since a >> ao, the dB values are always negative, so that the louder the bioacoustic signal compared to the reference white noise, the closer the dB value is to zero.

At each of our eight field sites, recordings were made at three stations spaced at 150 m intervals along a straight line transect. At each station, recordings were made for a minimum of 9 min and a maximum of 45 min, depending on wind conditions. Established ornithological field survey methodologies in Hawaii consider 8 min as the required amount of time within which most individuals at a given location will vocalize and can be accurately recorded (5,6). Therefore, we obtained 8 min of wind-free recordings at each station, since wind adds unwanted sound, or noise. Recordings were not made during precipitation events, directly after precipitation events when drops of water were falling from the canopy, or during periods of consistent wind.

Recordings were made two times per day: (i) 07:00–09:00 am and (ii) 12:00–14:00 pm local time between 22 February and 9 April 2006. These sampling periods were aimed at recording mainly bird vocalizations, although at the woodland site on the Exotic Gradient, crickets were audible during the morning recording. All recordings were made at least once at each site. To determine whether sampling date affected our results, a second morning recording was made at the forest, woodland and savanna sites on the Native Gradient, and a second evening recording was made at the forest and woodland sites on the Exotic Gradient. These sites were randomly chosen. We found no significant differences between recordings made on two separate dates.

The following pre-processing steps were employed prior to data analyses of bioacoustic recordings:

1. Recordings were first edited using free audio editing software (Audacity 1.2.4, available at Each audio file was shortened to 8 min by removing recordings of wind gusts or water drips from the canopy.

2. Each 8 min audio file, which contained information from 40 Hz to 20 kHz, was reduced to a frequency range of 2000 Hz to 8000 Hz, which preserves the frequency range of avian vocalizations for all but one rarely observed bird species (kalij pheasant, Lophura leucomelanos) present at our sites. By removing audio content below 2000 Hz, low frequency noise caused by auto traffic, as well as light wind gusts, were removed. Similarly, high frequency noise above 8000 Hz, from more distant auto traffic and rustling canopy foliage, was removed. We then repeatedly applied a fast Fourier transform filter until no sound remained outside of the frequency range of interest (2000–8000 Hz).

3. The edited and filtered audio files were then imported into Matlab (The Mathworks, Natick, Massachusetts, USA), and the ‘spectrogram’ function was applied to each file. This function computes the short-time Fourier transform of a signal, allowing user-defined FFT length and sampling frequencies for each audio file. The function outputs three vectors: time, frequency and power. Values in the power vector were then converted to decibels using Eq. (2).

Following these pre-processing steps, the three station recordings were averaged for each site. Decibels were then plotted as a function of frequency (Fig. 6). This allowed us to compare relative sound levels and frequency band usage among stations and sites. We also calculated the area under each of the bioacoustic spectra. The area under each curve included all frequency bands associated with the dB value that was greater than the minimum dB value for each curve. The area values are thus a function of both the sound level and the number of frequency bands used by the avifauna at each station and site. As all of the avifauna species in HAVO are vocal in the spring, this measure served as our bioacoustic index of avian abundance (Figs. 1, 2c, 3a).

(4) Traditional Ornithogical Surveys

Variable circular plot (VCP) counts are a form of sampling in which stations set at intervals along transects serve as the centers for estimating radial distances to birds for a period of time (Reynolds et al. 1980). VCP counts were conducted at the same three 3 stations for each site as were the bioacoustic recordings. For each site, all VCP-derived measures of avian community characteristics (total avian abundance (Fig. 2c), native and exotic avifauna abundance (Figs. 1, 3b,c,d,), ratio of native to exotic community sizes (Fig. 4), and individual species counts (Fig. 5) represent the average of the three station values. The VCP method is often used to estimate the population size or to model the density of a species within a particular habitat (7). In our study, the goal was to compare the total number of individuals and species of birds seen or heard by an observer to the total number of individuals and species estimated from bioacoustic recordings. VCP counts were conducted once at each station between 7 April and 9 May 2006. Counts were performed between 07:00 and 10:30 am local time by a single observer using 10 × 42 Leica binoculars during favorable conditions (light or no rain, and wind < 25 kph). Observers recorded all birds seen or heard during an 8 min period, and estimated the distance to each individual bird. Additional variables recorded at each station included time, and categories for wind, rain, cloud cover.

(5) Field-based Measurements of Biophysical Structure

Detailed field measurements were collected along the 300 m line transects at each field site. Fractional canopy cover and species composition were recorded in 1 m intervals using a point-intercept methodology (8). Tree and shrub heights were also surveyed using either a range pole or a digital-laser rangefinder (Impulse200; Laser Technology, Los Angeles, California). These field data were used to validate the remotely sensed canopy cover and vegetation height estimates.


Asner, G. P., and D. B. Lobell. 2000. A biogeophysical approach for automated SWIR unmixing of soils and vegetation. Remote Sensing of Environment 74:99–112.

Asner, G. P., and K. B. Heidebrecht. 2002. Spectral unmixing of vegetation, soil and dry carbon in arid regions: Comparing multi-spectral and hyperspectral observations. International Journal of Remote Sensing 23:3939–3958.

Asner, G. P., A. J. Elmore, F. M. R. Hughes, G. S. Warner, and P. M. Vitousek. 2005. Ecosystem structure along bioclimatic gradients in Hawaii from imaging spectroscopy. Remote Sensing of Environment 96:497–508.

Canfield, R. H. 1941. Application of the line interception method in sampling range vegetation. Journal of Forestry 39:388–304.

Green, R. O., M. L. Eastwood, C. M. Sarture, T. G. Chrien, et al. 1998. Imaging spectrometry and the airborne visible/infrared imaging spectrometer (AVIRIS). Remote Sensing of Environment 65:227–248.

Reynolds, R. T., J. M. Scott, and R. A. Nussbaum.1980. A variable circular-plot method for estimating bird numbers. Condor 82:309–313.

Scott, J. M., and F. L. Ramsey. 1981. Length of count period as a possible source of bias in estimating bird numbers. Studies in Avian Biology 6:409–413.

Scott, J. M., S. Mountainspring, F. L. Ramsey, and C. B. Kepler. 1986. Forest bird communities of the Hawaiian Islands: their dynamics, ecology, and conservation. Studies in Avian Biology. 9:1–431.

[Back to A017-086]