Remote sensing of wetland tree species in the iSimangaliso Wetland Park, KwaZulu-Natal, South Africa.
The impact of global change is expected to result in changes in the distribution and composition of species. Coastal swamp and mangrove forests are some of the most threatened forest types in the world. Remote sensing is a suitable tool for monitoring species distribution and varying condition because of its spatial extent and repeatability. The ability of remote sensing to separate between species can be attributed primarily to its capability to quantify the absorption features in the electromagnetic spectrum which relate to plant biochemical and biophysical properties such as pigments, nutrients (proteins and starch), leaf water content, leaf angle distribution, leaf area index and foliage biomass. For some species, these phenological variations are extreme, as in the case of deciduous tree species, thus enhancing the ability to differentiate between species, whereas others are less pronounced, such as with evergreen tree species, making spectral distinction between species much more challenging. Few studies have assessed the pigment and nutrient phenology of evergreen tree species in subtropical forested wetlands, let alone their spectral differences. This study assesses whether multi-season data across a number of phenological phases of evergreen wetland tree species will improve their classification accuracy when compared to a single season and single phenological event. The objectives were to (i) assess whether tree species had unique seasonal profiles of foliar biochemicals; (ii) ascertain the spectral bands of plant properties which remain important across phenological phases for species classification; (iii) determine whether leaf reflectance spectra from multiple seasons would improve species classification when compared to a single season; and (iv) whether multi-season imagery would improve species discrimination when compared to a single season. Thus, the study made use of leaf level and canopy level spectra collected using a handheld spectrometer and spaceborne RapidEye imagery, respectively. Six dominant evergreen tree species from forested wetlands in the subtropical region of KwaZulu-Natal, South Africa, were sampled across four seasons (winter, spring, summer and autumn). Differences in foliar biochemical concentration were assessed for two pigments, including carotenoids and chlorophylls, as well as two nutrients, nitrogen and phosphorous. The results showed that the majority of species had no significant changes in foliar pigments across the four seasons. Foliar nitrogen showed a significantly higher variability in the spring, summer and autumn seasons compared to the winter, whereas foliar phosphorus also varied across the seasons but to a lesser degree. The highest percentage of species pairs was separable using foliar nitrogen, compared to the pigments and phosphorus, emphasizing the importance of nutrients such as leaf proteins for species discrimination. The study found a changing relationship between leaf spectra and foliar nutrient concentration across the four seasons for the six evergreen tree species. Twenty-two spectral bands which are related to known absorption features of plant properties were identified across the four seasons as important for tree species discrimination. The relationship between leaf spectra and foliar nitrogen was highest during the spring, summer and autumn seasons for narrow bands associated with absorption features of proteins compared to the red-edge region. The spectra band combination 2130 nm and 2240 nm yielded the highest coefficient of determination between leaf spectra and foliar nitrogen across three of the four seasons. Season-specific prediction models were found to be more accurate in predicting foliar nitrogen than prediction models from across all seasons. The twenty-two bands were effective for the data reduction of the hyperspectral data and yielded a similar overall accuracy compared to 421 bands. Multi-seasonal data improved tree species classification for multispectral sensors with a few bands. The classification, in which multi-season leaf spectra or canopy data from RapidEye imagery was used, resulted in higher overall and user’s accuracies when compared to the single-season classifications. In contrast, the use of multi-season data for the classification of leaf spectra with 22 narrow bands, showed no statistical significance of differences compared to the classification results of the single season in which the highest overall accuracy of all single seasons had been obtained. The value of an increased classification accuracy should however be measured against the increase of cost when using images from multiple seasons. The study concludes that although seasonal profiles of foliar biochemicals overlap, multi-season information do improve species discrimination at foliar biochemical, leaf-spectra and canopy-spectra levels.