Building Good Spectral Vegetation Backgrounds in the VNIR

It has been said that one person´s signal is another person´s noise, or alternatively, one engineer´s ground clutter is another engineer´s object of interest. Vegetation observed with spectral imagers can be regarded either way. Some users of spectral data (multispectral or hyperspectral) want to study the plants in the pixels, and others want to use characteristics of the plants to move them out of the way of more interesting things. This is a difficult problem in either case in that vegetation is not time-static (spectral reflectance changes with seasons) and vegetation in general has smaller spatial scale than spectral data sets do. One approach is to model the reflectance or other properties of plants at pixel scale, using various observed data and mathematical models to span both wavelength and spatial dimension. Data collection at the leaf/branch level is known to be costly in labor, travel, and equipment and fraught with pitfalls for the field operator. Models for scaling up from plant data to larger spatial dimensions also have assumptions and built-in limitations. One big problem is to identify and avoid sources of error and variance which can cause a collection´s data to be suspect. Remotely-sensed spectral data have been around for decades, and research on uses of these data continues even as commercial and government agencies use multispectral and hyperspectral sets to identify species cover in forests, monitor the health of crops, seek mineral deposits, track pollution, and monitor the condition of roadways. Those who use spectral images get the data from sources like LANDSAT, ASTER, and SPOT, or newer systems such as AVIRIS, MODIS, and HYPERION. The technical challenge is to aggregate, upward, the knowledge collected at the leaf level, to enable people to make inferences at the pixel level. I will discuss some of these published efforts in more detail. There are several issues at each level in this aggregation, including calibration, drift, illumi- - nation, and sample integrity. In measuring leaf reflectance, there are several protocols dependent on the measuring equipment. The instrument also needs to be calibrated and controlled for drift. Deterioration of the sample being measured will occur. Illumination, if solar, will vary as measurements proceed. The same issues occur in the greenhouse/laboratory setting. We will explore several of these sources of non- representative spectra. Researchers are attempting to bridge the gap from leaf to pixel using canopy reflectance models. A few widely accepted models are SAILH and PROSPECT, and their variants. (Canopy models are also used for estimating biometric properties over large areas).