Charting the quality of forage : measuring and mapping the variation of chemical components in foliage with hyperspectral remote sensing

Rangeland carrying capacity is determined by both the quantity, as well as the quality, of forage. The estimation of the quantity of rangeland foliage and its spatial distribution has been studied and reliable results achieved. Traditional methods to determine quality involve collection of field samples and their analysis in a laboratory using wet chemical techniques. These analysis methods are time consuming and expensive. Previous studies have demonstrated that laboratory spectroscopy can be used to determine the concentration of leaf chemical compounds.<o:p></o:p></span>In this study I demonstrate a number of methods to measure the chemical composition of foliage using hyperspectral remote sensing techniques, and try to predict the<span class=SpellE>interspecific</span>variation of chemical composition in vegetation.<o:p></o:p></span>The results show that soil nitrogen and phosphorus content affect the spectral reflectance signatures of a tropical tree species (<span class=SpellE>Colophospermum</span>mopane ), grown in a greenhouse. Although several studies have indicated potential effects of herbivory on plant-chemical composition, this could not be confirmed. In an attempt to find better spectral indices to detect foliar nitrogen concentration across multiple species, it was shown that there are differences between species for the optimal spectral regions used to measure nitrogen concentration. For foliage trace-elements I found large differences between species thereby confounding the ability to generalize prediction models.<o:p></o:p></span>Variation of mopane foliar chemistry and forest structure in the north of<st1:place w:st="on"><st1:City w:st="on">Kruger National Park</st1:City>,<st1:country-region w:st="on">South Africa</st1:country-region></st1:place>, was mainly dependent on fire history, which is closely related to the parent material of the soil on which the plants were located.<o:p></o:p></span>This thesis concludes by combining the ecological and remote sensing knowledge generated in previous chapters into a neural network model to predict the concentration of Condensed Tannins and Polyphenols in a stretch of mopane woodland in<st1:place w:st="on"><st1:PlaceName w:st="on">Kruger</st1:PlaceName><st1:PlaceType w:st="on">National Park</st1:PlaceType></st1:place>from hyperspectral data. With an accuracy of approximately 60-70% this model proved to be successful. This is a step towards mapping the chemical composition of foliage, which may subsequently be used to analyze herbivore distribution patterns in space and time, and test optimal foraging theories