Calibration of broad- and narrow-band spectral variables for rangeland cover component quantification

Eight broad-band (BB) and 52 narrow-band (NB) spectral variables collected on single sampling dates in June, July and August of 1996 were used, either individually or over multi-temporal periods (June to July, June to August, and July to August), to evaluate their potential use in estimating key cover components (e.g. bare soil, rock, litter, lichen, moss, total live vegetation, shrubs, herbs, forbs and grasses) within a sagebrush steppe rangeland. Regression correlation coefficients ranged from a lour of 0.36 and 0.42 for moss to 0.80 and 0.87 for total live vegetation, for the leading BE and NE spectral variables respectively. In general, more specific plant growth forms (e.g. grass and forb) had a poorer explicability (i.e. lower regression correlation coefficient, r) than the more general components (e.g. herbs and total live vegetation). Several cover components, however, had improved explicability with NE spectral parameters using simple regression, relative to the leading BE variable. The greatest increases in r from using simple NE variables were for grass, shrub and total live vegetation cover. Slope-based (e.g. derivative) NE variables resulted in increased r values relative to the leading BE variable for forb and herb. In addition to the type of spectral variable, the date of spectral sampling was also found to be important. Although most cover components were best represented by June spectral data (e.g. at peak plant growth), shrub (live and dead) and grass cover were better estimated by August data (e.g. at senescence), and moss and rock cover by July data, indicating that the time of spectral sampling affects the ability to monitor these components. When multiple stepwise regression was used to isolate a subset of best spectral variables for each cover component, the model R-2 (coefficient of determination) ranged from 0.18 to 0.65 for BE data and from 0.11 to 0.75 for NE data. Finally, multi-temporal spectral sampling may improve explicability in this environment, but only for the herbaceous components. The greatest improvements from using changes in spectral indices over time were found using NE data: increases in r ranged from 0.005 for grass cover (July to August) to 0.082 for forb cover (June to July) and 0.106 for herb cover (June to August). In conclusion, it appears that if various detailed rangeland cover components are to be quantified with remotely sensed data, the use of NE spectral variables and multiple sampling dates may be beneficial with the greatest potential improvements limited to live vegetative components.