Project topics and materials
Get Complete Project Material File(s) Now! »
Remote sensing data
Five TerraSAR-X X-band dual-polarized (HH and HV), four RADARSAT-2 C-band quad-polarized (HH, VV, VH, and HV) and two ALOS PALSAR L-band dual-polarized (HH and HV) SAR intensity datasets (summarized in Table 3.1) were acquired to cover the study transect shown in Figure 3.1. Only dual polarized SAR data (HH and HV) was used because the HV polarization parameter is known to better model the structure of woody vegetation through volumetric backscatter interactions, while HH is also reported as been sensitive to structure although to a lesser extent than the cross-polarized band (Collins et al., 2009; Mathieu et al., 2013; Mitchard et al., 2009). Further, HH/HV was the common polarization configuration available for all three sensors. Winter seasonal SAR acquisitions were chosen because winter in the Lowveld is the dry season and exhibits the lowest level of moisture in the landscape. The tree leaves are off along with dry soil and dry grasses. This reduced the chance of interference of the SAR signal with variable moisture content while allowing a greater penetration of microwaves into the canopies. In the same region (Mathieu et al., 2013) reported the best retrieval of woody structural parameters with RADARSAT-2 data acquired in winter. An extensive airborne LiDAR dataset (total coverage of c.a. 63000 ha) were acquired for this study (Figure 3.1) by the Carnegie Airborne Observatory-2 AToMS sensor during April-May 2012. For our datasets, the LiDAR was operated at a pulse repetition frequency of 50 kHz with a 0.56m laser spot spacing and an average point density of 6.4 points per m2 from a flying altitude of 1000m above ground level (Asner et al., 2012). In comparison with the LiDAR dataset, the SAR images were acquired during the winter 2009 (RADARSAT-2), 2010 (ALOS PALSAR), and 2012 (TerraSAR-X).
Unfortunately, the last ALOS PALSAR winter scenes were acquired during 2010 in the study area and no RADARSAT imagery were available closer to 2012.
Field data
Field data were collected in April – May, and November – December 2012 across 38 sampling sites (in Figure 3.1). These sites provided ground truth data to model and validate the LiDAR derived woody structural metric products to be used to model the SAR-based woody structural metrics. Ground sampling sites were located to represent the diversity in woody structure of the different vegetation types, management regimes, and geological substrates mentioned above. Each site covered a 100m X 100m area and vegetation measurements were taken from four clustered 25m X 25m sampling plots (with minimum distance > 50m, identified from geostatistic range assessments, (Wessels et al., 2011)), located at each of the four corners of the site (Figure 3.2). The 100m X 100m sites were positioned using high resolution imagery from Google Earth as well as earlier LiDAR datasets acquired in 2008 – 2010 to ensure that they are representative of the surrounding landscape.
Field AGB estimates were derived from height and stem diameter measurements using an allometric biomass estimation equation ((Colgan et al., 2013) – Equation 3.1 in Appendix 3A). The allometric equation was developed following destructive harvesting of 17 savannah tree species present in the study area (Number of trees sampled =707; R2 = 0.98; relative Root Square Error = 52%; ranging from 0.2 – 4531 kg per tree, (Colgan et al., 2013)). Tree height was measured using a height pole and Laser vertex/rangefinder, while stem diameter was measured using callipers and Diameter above Breast Height (DBH) tape. Stem diameter was measured at 10cm above the ground and for multi-stemmed plants every individual stem was measured as separate individuals (e.g. species such as Dichrostachys cinerea).
Due to logistical and time constrains associated with measuring every tree within the sample plot two main stem diameter ‘zones’ were identified inside the site to increase sampling efficiency while still yielding representative quantities of biomass estimates (Figure 3.2). The first diameter zone was the 25m X 25m plot where all trees with a stem diameter of 5cm and greater were recorded, provided that they had a height of 1.5m or greater, and the second diameter zone was a 10m X 10m area positioned at the inner corner of the 25m X 25m plot where all trees with a stem diameter
between 3 and 5cm and greater than 1.5m were also recorded. This allowed catering for a few sites, mostly in the communal lands, where most of the AGB consisted of dense stands of multi-stemmed plants (coppicing) with low DBH (Matsika et al., 2012). A total of 152 25m X 25m biomass plots were sampled. Individual tree level AGB was derived using Colgan’s allometric equation (Colgan et al., 2013). AGB was then calculated for each diameter zone by summing the relevant tree level AGB values which was then subjected to particular AGB up-scaling factor (Equation 3.2 in Appendix 3B). The complete plot level AGB was calculated by summing all the corrected AGB subtotals for the stem diameter zones.
One or two sampling plots were chosen for most sites for CC data collection – the north east 25m X 25m plot and/or the south west 25m X 25m plot (DBH zone 2 – Figure 3.2). CC values were estimated following the vertical densitometer protocol (Ko et al., 2009; Stumpf, 1993), conceptually a point intercept sampling approach, and one of the most time-efficient techniques to implement. The point intercept method is a small angle approach well suited to measure the vertical canopy cover – i.e. vertical projection of canopy foliage onto a horizontal surface –, and as such is the most directly comparable with cover derived from remote sensing imagery such as LiDAR (Fiala et al., 2006). The sampling procedure involved laying down transects along a fixed 25m measuring tape orientated from north to south and moving from west to east within the subplot at 2m increments (Figure 3.2). Along these transects, the presence of canopy cover was determined using a 5m pole placed vertically above each sampled points every 2m along the transects. At each sampled point the presence of cover was coded as Y. For plot level canopy cover, in terms of percentage at the 25m X 25m scale, the CC presence and absence data were subjected to the formula below (Equation 3.3):
Plot level CC (%) = (ƩY/169) X 100 Equation 3.3
Where Y represents the presence of cover data. The value 169 represents the total number of sampling points in a 25m X 25m plot conducted at 2m sampling increments. A total of 37 (25m X 25m) plots of CC were recorded during the field campaign.
Chapter 1: Introduction and literature review
1.1 Savannahs and the importance of its woody component in ecosystem and monitoring processes.
1.2 The current status of monitoring the woody component in South Africa
1.3 Woody structural metrics
1.3.1 Woody biomass
1.3.2 Woody canopy volume
1.3.3 Woody canopy cover .
1.4 Remote sensing of woody structure
1.4.1 Passive remote sensing of woody structure
1.4.2 Active remote sensing of woody structu
1.5 Multi-sensor and multi-temporal remote sensing data integration
1.6 Study aim
1.7 Study objectives and chapter breakdown
1.8 Primary and secondary research questions
Chapter 2
Chapter 3
Chapter 4
Chapter 5
Chapter 2: Assessment of the Performance of Global Forest Cover Products in South Africa – Establishing the benchmark
2.1 Abstract
2.2 Introduction .
2.3 Study Area
2.4 Materials and Methodology
2.4.1 Global forest products
2.4.2 LiDAR validation datasets.
2.4.3 Global Forest Product Pre-processing .
2.4.4 Airborne LiDAR Data Pre-processing
2.4.5 Data extraction process
2.4.6 Global Product Accuracy Assessment .
2.5 Results
2.5.1 ALOS PALSAR FNF validation results
2.5.2 Landsat VCF validation results
2.6 Discussion
2.7 Conclusions
Chapter 3: Savannah woody structure modelling and mapping using multi-frequency (X-, C- and L-band) Synthetic Aperture Radar (SAR) data
3.1 Abstract
3.2 Introduction – Background, Aims and Objectives
3.3 Study Area .
3.4 Materials and Methodology
3.4.1 Remote sensing data
3.4.2 Field data
3.4.3 LiDAR data processing, woody structural metrics and validation
3.4.4 SAR data and processing .
3.4.5 Data integration, modelling protocols and mapping
3.4.6 Error assessment
3.5 Results
3.5.1 Modelling Accuracy Assessment .
3.5.2 Tree Structure Metric and Error Maps
3.6 Discussion
3.7 Concluding Remarks
Chapter 4: Integration of Optical and L-band Synthetic Aperture Radar (SAR) datasets for the assessment of woody fractional cover in the Greater Kruger National Park region
4.1 Abstract
4.2 Introduction
4.3 Study Area
4.4 Materials and Methodology
4.4.1 Remote Sensing Data .
4.4.2 LiDAR Data Processing
4.4.3 SAR Data Processing
4.4.4 Landsat-5 Optical Data Processing and Derived Products
4.4.5 Data Analysis Grid
4.4.6 Modelling Algorithms, Modelling Scenarios, Model Validation and CC Mapping
4.5 Results
4.5.1 Individual and multi-seasonal Landsat-5 reflectance compared to SAR
4.6 Discussion
4.7 Conclusions
Chapter 5: Scaling-up methods for national woody fractional cover mapping: Experiments and guideline on the amount of field plots and airborne LiDAR data required for training and validation
Chapter 6: Study conclusions, recommendations and ways forward
References
