I contributed substantially to the study design, fieldwork, analysis, interpretation and discussion of results, and overall manuscript preparation, and was therefore the corresponding first author of both manuscripts. Invasive alien plant species (IAP) affect a range of ecosystem types in different regions of the world. First, the study attempted to determine the optimal subset of bands from canopy hyperspectral data for discriminating P .
Although the higher spatial resolution of SPOT 6 was useful for better characterization of the distribution and patch sizes, the study found that the spectral configuration of OLI was more important in identifying potential sites contaminated with P. Overall, the study showed that fewer spectral bands were selected by the proposed hierarchical approach has the greatest potential for reliably distinguishing IAP types using airborne and satellite hyperspectral sensors. The study also showed that current information needs about IAPs can be addressed using accessible multispectral data, valuable for effective land management, site-specific weed management and site prioritization.
I would also like to express my sincere gratitude to all the SANSA staff who in one way or another contributed to my professional development, thus the completion of this study. I would like to thank KZN Wildlife for their permission to conduct the study in Ndumo Game Reserve.
- Background
- The potential of Remote Sensing for IAP species discrimination
- Hyperspectral remote sensing of IAP species
- Multispectral remote sensing of IAP species
- Classification algorithms and vegetation indices for discriminating IAP species . 7
- Research problem
- Aim and Objectives
- Aim
- Objectives
- Research questions
- Scope of the study
- Study area
- Chapter outline
Furthermore, information on the spatial patterns of IAPs is important for establishing ecological links with diversity, underlying ecosystem structure and processes, and habitat changes (Turner et al., 2003). Hyperspectral data contain electromagnetic energy reflected from an area of interest in hundreds of contiguous narrowband intervals (see Figure 1) (Jensen et al., 2007). The increased power of spectral resolution allows the discrimination of subtle intra- and inter-species reflectance differences (Narumalani et al., 2009).
These properties allow species discrimination based on their absorption of specific regions of the electromagnetic spectrum (Jia et al., 2011, Chun et al., 2011). In addition, main factors affecting the spectral signature over 400nm- 2500nm regions of the electromagnetic spectrum are shown (Adam et al., 2010, Adjorlolo et al., 2012b). For example, Fernandes et al. 2013) note that optimal spectral bands selected from hood hyperspectral measurements are essential for selecting the most suitable satellite images for mapping.
Multispectral sensors capture the reflected and emitted energy from an area of interest in multiple broadband intervals (i.e. ~4 to 36) of the electromagnetic spectrum (see Figure 1) (Jensen et al., 2007). Although several techniques have been proposed to overcome these challenges, none of them has proven superior (Adam and Mutanga, 2009, Jia et al., 2011).
- Introduction
- Materials and methods
- Species description
- Data collection
- Pre-processing and analysis
- SVM classification and validation
- Results
- Kruskal-Wallis ANOVA
- Inter-band correlation and AUC-ROC variable importance
- SVM-RFE
- SVM Classification and validation
- Discussions
- Conclusions
The chapter provides a general background to Parthenium hysterophorus, including its description, distribution and impact, the benefits of using remote sensing data and algorithms to distinguish IAP species. Furthermore, research objectives, description of the study area and the scope of the study are detailed.
Introduction
Data and materials
- Data description
Methods
- Pre-processing
- Support Vector Machines (SVM) Classification
- Distribution and patch sizes of P. hysterophorus
- Accuracy assessment and map comparisons
Results
- Parameterisation of SVM classifier
- SVM classification results
- Accuracy assessment and map comparisons
- Distribution and patch sizes of P. hysterophorus
Discussions
- The capability of multispectral data for mapping P. hysterophorus
- Addressing information needs for optimising control mechanisms
Conclusions
Introduction
These are produced in just four to six weeks after germination, followed by the production of approximately 25,000 easily dispersible seeds per plant (Dogra et al., 2011). As a result, the effectiveness of physical, chemical and biological control mechanisms increases during the juvenile growth stage (Reddy et al., 2009, Khan et al., 2012). Thus, for early detection and optimization of extinction mechanisms, it becomes important to explore the potential of hyperspectral and multispectral remote sensing data to distinguish and map P.
The results in this study showed the potential of remote sensing for the discrimination and mapping of a problematic IAP, P. The results were significant for understanding the spectral differences between species and identifying the optimal bands valid for operational mapping using airborne and satellite sensors, choosing suitable sensor for mapping and early detection of IAP species (Chapter 2). OLI and SPOT 6 and the ability to provide spatial distribution and patch sizes of P.
The results demonstrated the potential of OLI and SPOT 6 spectral and spatial configurations for reliably mapping P distribution and patch sizes.
Improving classification accuracy through feature subset selection and dimensionality
Optimal spectral bands for discrimination of P. hysterophorus
The utility of red-edge, NIR and SWIR regions for distinguishing species has been reported in the literature (Schmidt and Skidmore, 2003, Mutanga and Skidmore, 2007, Adam and Mutanga, 2009). In addition, all selected spectral bands (in this study) were different from those selected in other recent studies (see Table 14), reinforcing the claim that there is no single technique that is useful for all species types. Previously selected bands for species discrimination separated by wide spectral ranges proposed by (Fernandes et al., 2013).
Reliable information for effective management of P. hysterophorus
Conclusions and Recommendations
Mapping insect (solanum mauritianum) infestations in pinus patula plantations using hyperspectral imagery and support vector machines. Classification of hyperspectral remote sensing data with primitive SVM for small-size training dataset problem. Monitoring the invasion of an exotic tree (Ligustrum lucidum) from 1983 to 2006 with Landsat TM/ETM+ satellite data and Support Vector Machines in Córdoba, Argentina.
Effect of spatial resolution on classification errors of pure and mixed pixels in remote sensing. A hybrid model for classification of remote sensing images with linear SVM and support vector selection and adaptation. Impact of Feature Selection on the Accuracy and Spatial Uncertainty of Per-Country Crop Classification Using Support Vector Machines.
Remote sensing as a tool for monitoring plant invasions: Testing the effects of data resolution and an image classification approach on the detection of a model plant species Heracleum mantegazzianum (giant hogweed). Support vector machine-based feature selection for land cover classification: a case study with DAIS hyperspectral data. Distribution map of Crofton weed (Eupatorium adenophorum spreng) in southwest China using time series remote sensing data.
Application of unmixing machine and spectral support vector to airborne hyperspectral imagery for detection of giant reeds. Landsat remote sensing approaches for monitoring long-term tree cover dynamics in semi-arid forest lands: Comparison of vegetation indices and spectral mixture analysis.
