Laserfiche WebLink
<br />2. Vegetation texture is an important factor in vegetation map accuracies. Texture was <br />found to increase mapping accuracies by 20-30 absolute percent over accuracies <br />obtained from vegetation mapping that used just color retlectance information. <br />Mapping accuracy is the percentage of picture elements for a particular vegctation <br />alliance or collection of alliances that are correctly identilied. Texture is much better <br />defined within the vegetation at higher spatial resolutions and is best derived from <br />imagc resolutions near 20 cm or higher. <br /> <br />3. Calibrated, 9-band multispectral image data (I-m resolution) produced higher map <br />accuracies than higher resolution (II-cm), un-calibrated CIR film. For corridor-wide <br />inventory studies, radiometric calibration is critical. Reducing the number of <br />multispectral bands used to map vegetation to only four wavelength bands did not <br />greatly reduce mapping accuracies over those obtained using the full 9-band set. The <br />four most useful bands for mapping Ihe CRE riparian vegetation were centered near <br />the wavelengths 0.53-54 I'm. 0.066-0.67 jlm. 0.70 I'm, and 0.79-0.82 jlm. These are <br />close to the wavelengths selected for the early Landsat Multispectral Scanner System <br />(Landsat MSS). which was primarily designed for vegetation monitoring. <br /> <br />4. The use ofa Global Positioning System (GPS) and an Inertial Measuring Unit (IMU) <br />that provides 30-cm positional accuracies should be used with DEM data to <br />onhorectify the image data in order to provide accurate area and volume estimates. <br /> <br />5. Phase angle during image data collection should be kept to a narrow range (:'0 10l) <br />because canopy retlectance values can change at different solar incidence angles and <br />sensor viewing angles. <br /> <br />6. Subsequent field work is required to reach an 80% mapping accuracy, but the tield <br />work would be much less intensive and invasive than current field survcys for <br />vegetation. <br /> <br />7. Remote sensing will not eliminate the need for the current random sampling of <br />vegetation within the CRE bccause understory is a key component in these field <br />surveys and remote-sensing data even at 6-cm rcsolution cannot idcntify, and in most <br />cases see, the understory. <br /> <br />We have not yet found an imaging system that can provide thc four wavelength bands <br />listed in i1c;n 3 above with acceptable spatial resolutions. We did locate a 4-band imaging system <br />operated by ISTAR Americas that provides three of the four desired wavelength bands (not the <br />band centered at 0.70 I'm), in addition to a blue wavelength band, at 44-cm spatial resolution. <br />This system was used to collect the 2002 annual image and DEM data for the entire CRE. The <br />positional accuracy of the onhorcctified image data was found to be about 30 cm, which is bettcr <br />Ihan most other image data acquired by GCMRC to date and acceptable for biologic monitoring <br />purposes. The 44-cm spatial resolution was less than desired, but our preliminary vegctation <br />analyses using these data show that texture derived from this 44-cm data still added 10-25% to <br />the c1assitication accuracies for various vegetation alliances. We did cncounter a problem with <br />saturation in the critical NIR image data, which means that the recorded NIR brightness for some <br />vegetation alliances were similar even though their NIR ground rcllectance spectra shuw that they <br />are distinctly different. This is a calibration issue and requires contractors to properly set detector <br />gain settings in order to capture the full retlectance rangc of all vegetation species. <br /> <br />16 <br />