The first objective of the study provides an overview of progress in satellite remote sensing for mapping the spread of AIPs and associated challenges and opportunities. Serge Kiala Zolo for his necessary assistance in the production of the second article (chapter three), Mr.
Introduction
On the other hand, numerous human health risks have been reported due to Parthenium invasion (McConnachie et al. 2011). However, the limitations of using such methods in monitoring alien invasions are well documented in the literature (Ismail et al. 2016).
Aims and objectives
Research questions
Hypothesis
Chapter outline
This work therefore provides an overview of advances in satellite remote sensing for mapping and monitoring AIPs distribution and associated challenges and opportunities. Therefore, the use of high spatial and hyperspectral dataset is prohibitive for long-term monitoring of AIPs which is a requirement for effective management of AIPs distribution.
Introduction
Similarly, the repeat coverage possible with satellite remote sensing approaches allows the detection of plant phenology which is necessary for detecting the spread of AIPs (Flood 2013). Finally, the paper highlights the challenges of using remote sensing in the detection and mapping of AIPs and suggests directions for future research.
The ecology and spatial distribution of AIPs
Furthermore, the literature indicates that AIPs are unpalatable to game animals and pastures (Everitt et al. 1995; Pyšek 1998). The South American miconia (Miconia calvescens) for example was deliberately introduced to the island of Tahiti in 1937 for decorative reasons (Lowe et al. 2000).
Spectral properties of AIPs for remote sensing techniques
According to Kalusová et al. 2013) and Pyšek (1998), highly invasive plant species originating in Europe. People may deliberately introduce exotic plants into new regions for the control of other problematic species, to improve agricultural productivity or for ornamental reasons (Goodwin et al. 1999).
Multispectral remote sensing of alien invasion
In some cases, AIPs obscure the background of natural vegetation, especially in the early stages of their growth (Peerbhay et al. 2015). For example, Ngubane et al. 2014) reported improved detection (91.67% overall accuracy) of Bracken fern in Durban, South Africa, using high spatial resolution WordView-2 rather than medium spatial resolution SPOT-5 (72.22% overall accuracy ). Other studies that have reported improved discrimination of vegetation types using multispectral sensors with high spatial resolution include Oumar (2016), Peerbhay et al.
Despite the improved spatial discrimination of features, literature shows that the usefulness of multispectral sensors in vegetation monitoring is still hampered by the poor spectral resolution (Ngubane et al. 2014).
Hyperspectral remote sensing of AIPs
The use of multisource data for detection of AIPs
However, the full potential of data fusion for optimal detection and mapping of AIPs has not been sufficiently explored. While studies have shown the success of this method in detecting tree species from grassland environments (Ghosh et al. 2014; Naidoo et al. 2012; Cho et al. 2012), the high-performance computing power required to process fused remotely sensed data makes the approach is expensive, especially for large-scale mapping purposes (Huang and Asner 2009). Furthermore, data fusion for detection of AIPs has been mainly dominated by the integration of either hyperspectral or multispectral dataset with LiDAR, which is expensive, thus not a viable alternative for large-scale and continuous monitoring of AIPs.
On the other hand, Peerbhay et al. 2016a) suggest that the influence of the Bidirectional Reflectance Distribution Function (BRDF) still needs to be addressed to minimize false classifications that could arise, potentially, due to differences in solar and sensor geometry when using multisource data.
Parametric and non-parametric image classifiers for invasive alien plants
In Africa, the plant has become widespread in the eastern and southern parts of the continent (Zuberi et al. 2014). The literature indicates that annual germination and growth of Parthenium is limited by soil moisture (Goodall et al. 2010). However, the success of such methods in monitoring alien invasion has been largely limited to small-scale applications (Peerbhay et al. 2016b).
Dominant vegetation includes tropical shrubs and savannas, as well as tropical coastal forests (Wigley et al. 2009). -red/NIR+red Matongera et al. 2017) Difference Vegetation Index (DVI) NIR-red Dube et al. Moreover, the Random Forest is stable and faster (Chan and Paelinckx 2008), and can be easily implemented and interpreted (Odindi et al. 2014).
Odindi et al. 2014) was then used to classify all the image pixels based on the trained parameters. According to literature, growth and spread of Parthenium is determined by the available soil moisture (Goodall et al. 2010).
The use of vegetation indices for detection of AIPs
Relationship between classification algorithms and remote sensing dataset
Based on research, there is limited literature that clearly demonstrates synergies between a range of remotely sensed data used in conjunction with a chosen image classification algorithm. 2014), point out that many image classification algorithms perform well on medium-resolution multispectral data. 2016), for example, detected invasive Mesquite. Although sensor resolutions, especially spectral and spatial, are important factors for differentiating vegetation types (Oumar 2016), image classification algorithms also allow the estimation of accurate separation between different plants, even when using averaged spectral data and poor spatial resolution data. Obviously, the choice of the image classifier selected for a particular image classification process is crucial to improve the performance of the applied remote sensing sensor data.
The application of robust and advanced non-parametric image classification algorithms can also significantly improve the performance of these newly launched multispectral sensors.
Challenges in remote sensing of alien invasion
For example, Matongera et al. 2017), compared the performance of two different sensors with various spatial and spectral characteristics (i.e. high spatial WorldView-2 and medium spatial Landsat 8 OLI) in detecting Bracken fern using Discriminant Analysis. To demonstrate the role of non-parametric image classifiers over parametric classifiers, Gil et al. 2011), tested the performance of two parametric (Maximum Likelihood and Mahalanobis distance) and non-parametric (artificial neural network and support vector machine) algorithms in assessing the potential of high-resolution satellite images in vegetation mapping. Also, because AIPs often grow in a mix of coexisting vegetation, their detectability can be significantly compromised, especially with average spectral and lower spatial resolution data sets (Matongera et al. 2016a; Huang and Asner 2009).
According to Peerbhay et al. 2016c), the uniformity and comprehensiveness required in detecting the distribution of AIPs, especially in datasets with poor spatial and average spectral resolution, is not always achievable in the newly invaded landscapes.
Possible directions of future research
Moreover, integrating data from different sensors does not handle the small bandwidth dataset provided by high spatial resolution and hyperspectral imaging data. 2016), also detected the Lantana niche in the highlands of KwaZulu-Natal, South Africa using SPOT-6 and Random Forest data with an overall accuracy of 75%. Furthermore, Support Vector Machine and Random Forest provided good classification accuracies (91.80% and 93.07%, respectively) when mapping the patterns and spatial distribution of land use/cover types in a heterogeneous landscape of KwaZulu-Natal, South Africa , using RapidEye data (Adam et al. However, alien invasion studies using improved resolution data and nonparametric classification algorithms are mostly based on single-date image scene.
With increased free provision of multispectral data with improved resolution combined with the value of non-parametric algorithms to facilitate classification accuracy, timely and large-scale updates regarding the spatial and temporal distribution of AIPs are achievable.
Conclusions
2016), also detected the Lantana camara in field areas of KwaZulu-Natal, South Africa using SPOT-6 data and Random Forest with an overall accuracy of 75%. Tracking the spatial and temporal distribution of AIPs such as Parthenium is essential for facilitating management and mitigation of spread. This study attempted to determine the spatial and temporal distribution of Parthenium using multi-temporal SPOT series data, Random Forest and Land Change Modeler (LCM).
This study has demonstrated the value of cost-effective multispectral SPOT series data in conjunction with the robust and advanced nonparametric Random Forest classifier in detecting and mapping the spatial and temporal spread of AIPs.
Introduction
Traditionally, data for monitoring and controlling Parthenium have been collected using manual methods such as field surveys (McConnachie 2015; McConnachie et al. 2011). Alternatively, remote sensing has emerged as a reliable approach to map the distribution of AIPs scattered at the landscape scale (Matongera et al. More so, the robust and advanced non-parametric image classifiers such as Random Forest have been valuable in detecting AIPs scattered (Kganyago et al. 2017; Abdel-Rahman et al. 2014; Pal 2005).
Some studies (e.g. Matongera et al. 2017; Oumar 2016) have shown that the incorporation of vegetation indices is instrumental in improving classification accuracy, i.e. the calculation of eight vegetation indices.
Methods and Material
- Description of the study site
- Field data collection
- Image acquisition and preprocessing
- Vegetation indices retrieval
- Random Forest algorithm
- Image classification and Random Forest optimization
- Accuracy assessment
- Post-classification and change detection
- Annual rainfall distribution
For example, NDVI has been successful in estimating biomass and crop yields (Oumar 2016; Matongera et al. 2017). The advantages of Random Forest are that it is non-parametric (distribution free) and does not suffer from the Hughes phenomenon of overfitting (Abdel-Rahman et al. 2014). The overall accuracy is a percentage ratio between the number of correctly classified classes and the number of test data (Abdel-Rahman et al. 2014).
Research shows that this procedure is valuable for separating multitemporal image classification from image comparisons (Otunga et al. 2014).
Results
- Assessment of classification accuracies
- Land use/land cover transformation and Parthenium distribution
- Spatial and temporal variability in Parthenium distribution
- Comparing the spread of Parthenium between agricultural and non-agricultural dominated
However, in 2012, the area of bare land increased significantly (47%), taking over most of what was previously covered by Parthenium and grassland (Figure 3.9(a)). In 2016, there was a recovery in all vegetation classes occupying most of the area of previously vacant or bare land. In addition, the highest Parthenium infestations, covering 29% and 28% of the study area, were recorded in 2006 and 2009, respectively.
In the year 2012, the area occupied by Parthenium dropped significantly to 12% of the total land area within the study site.
Discussion
Despite the effect of low rainfall, Parthenium distribution showed a direct relationship with land use/cover transformations in the study area. For example, the sharp reduction in the area of land occupied by Parthenium and grassland in the year 2012 (Figure 3.2 and 3.3) can be attributed to the increase in the area of bare land (Figure 3.9(a)). During this period, grasses and Parthenium are believed to have declined due to low rainfall (Figure 3.7) while most of the study area's landscape was exposed by vegetation.
Despite the literature advocating for improvement in classification accuracy, the selected vegetation indices performed poorly in distinguishing parthenium from forest and grassland and were therefore not reported in the study.
Conclusion
Introduction
Evaluation of research objectives
- Objective one
- Objective two
The main objective of the study was to assess remote sensing applications for mapping the spatial and temporal distribution of Parthenium in Mtubatuba Municipality in KwaZulu-Natal, South Africa. Detecting subcanopy invasive plant species in tropical rainforest by integrating optical and microwave (InSAR/PolInSAR) remote sensing data. Using canopy reflectance models and spectral angles to assess the potential of remote sensing to detect invasive weeds.
Methodology to map the spread of an invasive plant (Lantana camara L.) in forest ecosystems using Indian satellite remote sensing data. An overview of remote sensing of invasive weeds and early spot detection example. Land use Land cover change at the boundary of eThekwini Municipality: Implications for urban green spaces using remote sensing.
