Corresponding author: Perushan Rajah ( perushanrajah@gmail.com ) Corresponding author: John Odindi ( odindi@ukzn.ac.za ) Academic editor: Andreas Huth
© 2019 Perushan Rajah, John Odindi, Onisimo Mutanga, Zolo Kiala.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Rajah P, Odindi J, Mutanga O, Kiala Z (2019) The utility of Sentinel-2 Vegetation Indices (VIs) and Sentinel-1 Synthetic Aperture Radar (SAR) for invasive alien species detection and mapping. Nature Conservation 35: 41-61. https://doi.org/10.3897/natureconservation.35.29588
|
The threat of invasive alien plant species is progressively becoming a serious global concern. Alien plant invasions adversely affect both ecological services and socio-economic systems. Hence, accurate detection and mapping of invasive alien species is valuable in mitigating adverse ecological and socio-economic effects. Recent advances in active and passive remote sensing technology have created new and cost-effective opportunities for the application of remote sensing to invasive species mapping. In this study, new generation Sentinel-2 (S2) optical imagery was compared to S2 derived Vegetation Indices (VIs) and S2 VIs fused with Sentinel-1 (S1) Synthetic Aperture Radar (SAR) imagery for detecting and mapping the American Bramble (Rubus cuneifolius). Fusion of S2 VIs and S1 SAR imagery was conducted at pixel level and multi-class Support Vector Machine (SVM) image classification was used to determine the dominant land use land cover classes. Results indicated that S2 derived VIs were the most accurate (80%) in detecting and mapping Bramble, while fused S2 VIs and S1 SAR were the least accurate (54%). Findings from this study suggest that the application of S2 VIs is more suitable for Bramble detection and mapping than the fused S2 VIs and S1 SAR. The superior performance of S2 VIs highlights the value of the new generation S2 VIs for invasive alien species detection and mapping. Furthermore, this study recommends the use of freely available new generation satellite imagery for cost effective and timeous mapping of Bramble from surrounding native vegetation and other land use land cover types.
Alien species invasions, Sentinel-1, Synthetic Aperture Radar (SAR), Sentinel-2, Vegetation Indices (VIs), American Bramble, Fusion, Support Vector Machine (SVM)
Global biodiversity is increasingly becoming susceptible to pressure from invasive species (
In South Africa, approximately two million hectares of land have been invaded by invasive alien plant species (
To develop optimal mitigation of spread and eradication approaches, determination of spatial cover and extent of Bramble infestation is paramount. Traditionally, surveys have been adopted for mapping and monitoring of invasive alien plant species (
Conventional remote sensing of invasive alien species utilises spectral wavelengths of absorbed and reflected light by distinguishing certain pigments in leaves and inflorescence (
This study was conducted at the uKhahlamba Drakensberg Park (UDP), a UNESCO proclaimed world heritage and nature conservation area. The area is situated along the western edge of the KwaZulu-Natal province of South Africa (Figure
Field data collection was conducted during spring and summer of 2016. A purposive sampling technique was utilised to record ground truth points of four major land cover classes (Bare rock, Bramble, Forest and Grassland). These seasons were chosen for field data collection as Bramble patches are most phenologically discernible from native vegetation. Ground control points were recorded as close to the centroid of Bramble patches as possible. Collected Bramble patches ranged from 15 m × 15 m to 50 m × 50 m. Ground truth point data collected from Bramble patches were spatially independent from each other to compensate for the spatial resolution of the satellite imagery utilised. This ensured that each Bramble patch fell within a single image pixel and could be associated with the unique spectral reflectance of a specific pixel. Due to the area’s steep and mountainous terrain, hence restricted accessibility, only Bramble patches that could be accessed by foot were considered for this study. In addition, aerial photographs at a 0.5 m spatial resolution captured in 2016 were used to supplement and verify selected land cover ground truth points. In total, 15, 40, 45 and 60 ground truth points were used for Bare rock, Forest, Grassland and Bramble, respectively.
The Sen2Cor plugin ESA SNAP toolbox 3.0 (
Summer Synthetic Aperture Radar (SAR) data were downloaded from the Sentinel-1 data hub. Sentinel-1 level-1 Ground Range Detected (GRD) products were multi-looked and projected to ground range using an earth ellipsoid model. SAR Vertical-Horizontal (VH) polarised imagery was acquired using the Interferometric Wide Swath (IW) mode, with a spatial resolution of 20 metres and a 250 km2 swath width. Pre-processing of SAR imagery was conducted using the ESA SNAP toolbox following the methodology outlined in
Sixty-five Vegetation Indices (VIs), selected from the online Index Database (IDB) (www.indexdatabase.de), were calculated from summer Sentinel-2 surface reflectance optical imagery. The IDB is a tool developed to provide a simple overview of satellite specific vegetation indices that are usable from a specific sensor for a specific application (
Selected S2 derived VIP vegetation indices subsequently utilized for SAR fusion
VIP Vegetation Indices (VIs) | VI formula (S2 optical bands) | |
Datt2 (Simple Ratio 850/710) | Near Infrared (NIR)/Red Edge 1 |
|
PSSRc2 (Simple Ratio 800/470 Pigment specific simple ratio C2) | Near Infrared (NIR)/Blue | Blackburn 1998 |
RDVI (Renormalized Difference Vegetation Index) | Near Infrared - Red/(Near Infrared + Red)0.5 | Roujean and Breon 1995 |
SR520/670 (Simple Ratio 520/670) | Blue/Red |
|
SR672/550 (Simple Ratio 672/550) | Red/Green |
|
SR800/550 (Simple Ratio 800/550) | Near Infrared/Green |
|
SR833/1649 (Simple Ratio 833/1649 MSIhyper) | Near Infrared /Shortwave Infrared1 |
|
SR860/550 (Simple Ratio 860/550) | Narrow-Near Infrared/Green |
|
SRMIR/Red (Simple Ratio MIR/Red Eisenhydroxid-Index) | Shortwave Infrared2/Red Edge 1 |
|
TM5/TM7 (Simple Ratio 1650/2218) | Shortwave Infrared1/ Shortwave Infrared2 |
|
Pixel level image fusion, based on ground truth points, was used to merge the ten most influential VIP VIs and Sentinel-1 SAR imagery (a description of image fusion levels can be found in
Image classification was conducted post pixel level image fusion as outlined in
Support Vector Machine classification maps were generated for S2 optical imagery, Vegetation Indices and for the fused VIS and SAR imagery within a Python environment. Training data (70%) of all four considered land cover classes were used as the input for Bramble spatial cover maps. The respective test dataset (30%) was then used to assess classification accuracies across all imagery. A confusion matrix was generated from the SVM process and user and producer accuracies used to quantify the reliability of the resultant Bramble spatial cover maps. In a confusion matrix, the overall accuracy is determined by dividing correctly classified pixels by the total number of pixels checked (
Producer’s accuracy (%) = 100% – error of omission (%) (1)
User’s accuracy on the other hand is a measure of map reliability and provides information on how well a map represents ground features. It is expressed as:
User’s accuracy (%) = 100% – error of commission (%) (2)
The overall classification accuracy using S2 optical bands was 78% (Table
Support Vector Machine (SVM) confusion matrix using Vegetation Indices for Bramble mapping and discrimination. Where BR = Bare rock; BBL = Bramble; FR = Forest; and GR = Grassland, UA = Users accuracy; PA = Producers accuracy and OA = Overall accuracy.
S2 (Optical bands) | BR | BBL | FR | GR | UA (%) |
BR | 33 | 2 | 0 | 11 | 70 |
BBL | 0 | 24 | 0 | 30 | 46 |
FR | 1 | 1 | 51 | 3 | 92 |
GR | 2 | 3 | 7 | 94 | 89 |
PA (%) | 92 | 81 | 87 | 69 | |
OA (%) | 78 |
Support Vector Machine (SVM) confusion matrix using Sentinel-2 optical bands for Bramble mapping and discrimination. Where BR = Bae rock; BBL = Bramble; FR = Forest; and GR = Grassland, UA = Users accuracy; PA = Producers accuracy and OA = Overall accuracy.
Vegetation Indices (VIs) | BR | BBL | FR | GR | UA (%) |
BR | 51 | 11 | 0 | 0 | 83 |
BBL | 0 | 53 | 19 | 4 | 72 |
FR | 1 | 0 | 54 | 0 | 97 |
GR | 13 | 7 | 0 | 57 | 74 |
PA (%) | 83 | 78 | 76 | 91 | |
OA (%) | 84 |
A large overestimation of Bramble discrimination and spatial cover using S2 optical bands was evident (Figures
Support Vector Machine (SVM) classification maps produced utilising (a) Vegetation Indices; (b) S2 optical bands and (c) Fused VIs and SAR.
Discrimination and mapping of Bramble using vegetation indices produced the highest overall accuracy (82%) when compared to the benchmark of using only S2 optical image bands (Table
The classification map resulting from fused vegetation indices and SAR imagery showed the most accurate discrimination and spatial cover of all considered land cover classes. The Grassland and Bare rock classes were reliably discriminated (Figures
The ten most influential S2 VIs were selected for pixel level image fusion with S1 SAR imagery. Using VIP, the influence of VIs was identified by the importance on increasing Grassland and Bramble’s User’s and Producer’s accuracy, hence, the ten bands that generated the ten highest classification accuracies were selected. Five of the selected VIs incorporated the Near Infrared (NIR) optical band, while three selected VIs were derived using Shortwave Infrared 1 (SWIR1) and Shortwave Infrared 2 (SWIR2) optical bands (Table
The fusion of VIs and S1 SAR imagery produced the lowest overall accuracy (55%) when compared to the benchmark of S2 optical band results (Table
Support Vector Machine (SVM) confusion matrix using fused Vegetation Indices and SAR imagery for Bramble mapping and discrimination. Where BR = Bae rock; BBL = Bramble; FR = Forest; and GR = Grassland, UA = Users accuracy; PA = Producers accuracy and OA = Overall accuracy.
VIs and SAR | BR | BBL | FR | GR | UA (%) |
BR | 43 | 2 | 0 | 11 | 79 |
BBL | 0 | 15 | 0 | 38 | 29 |
FR | 1 | 0 | 45 | 17 | 73 |
GR | 0 | 53 | 0 | 39 | 42 |
PA (%) | 97 | 20 | 100 | 37 | |
OA (%) | 55 |
The SVM classification map, produced using fused vegetation indices and SAR, resulted in an underestimation of the Bramble class, while an overestimation of the Grassland class was observed (Figure
This study sought to determine the potential of derived Vegetation Indices (VIs) and fused VIs and Synthetic Aperture Radar (SAR) imagery to improve invasive alien species detection and mapping. The overall classification accuracy of optical imagery was used as the benchmark for comparison of the results achieved using S2 VIs and fused VIs and SAR. Opposing the expected outcome, fused VIs and SAR imagery produced the lowest classification accuracy (55%) compared to conventional S2 optical imagery (78%). Moreover, S2 derived VIs produced the highest classification accuracy (84%) when compared to conventional S2 optical imagery and fused VIs and SAR.
Poor performance of fused VIs and SAR imagery was unanticipated and opposes research done by
The use of S2 VIs outperformed the benchmark accuracy achieved by conventional S2 optical imagery. Similar results were achieved by
Eight of the ten VIP VIs selected for Bramble discrimination and mapping were derived from at least one of these three spectral bands. The strong relationship between NIR, SWIR and red edge bands to variable vegetation parameters could have resulted in the increased accuracy of Bramble discrimination and mapping. Moreover, reflectance within the visible region of the spectrum is largely determined by vegetation pigments and is commonly used to quantify vegetation physiological properties (
Several studies (e.g.
According to
This study utilised freely available advanced Sentinel-1 radar and Sentinel-2 optical imagery, with the aim of evaluating spectrally derived VIs and fusing Synthetic Aperture Radar (SAR) imagery for improving American Bramble (Rubus cuneifolius) detection and mapping. This study contributes to the evaluation of economically viable, efficient and large scale invasive alien species detection and mapping. Conventional S2 optical imagery was used as a benchmark for comparison with results achieved using S2 VIs and fused VIs and S1 SAR imagery. The use of S2 VIs increased overall classification accuracies when compared to traditional optical imagery results, while the fusion of S2 VIs and S1 SAR decreased the overall accuracies. Hence this study demonstrated that new generation S2 VIs have the potential to increase the detection and mapping of Bramble from surrounding native vegetation. Results further indicate that the fusion of VIs and SAR imagery for Bramble detection and mapping failed to increase overall classification accuracies, hence have limited utility when applied to Bramble detection and mapping. The new generation satellites, such as S1 and S2, possess unprecedented sensor characteristics like higher temporal and spatial resolution, as well as tandem acquisition of SAR data, hence valuable for improved landscape mapping. This study concludes that the recently launched Sentinel satellite, with optical and radar capabilities, holds great promise in landscape delineation and vegetation mapping.