The project will involve the following key tasks: a) formally define the habitat metrics to be estimated in this study, b) select the study area, c) acquire radar and high resolution multispectral imagery, and other thematic maps (e.g. bedrock geology, landform, hydrology, soil, etc.), d) conduct intensive field data collection, e) process, analyse, and develop the model, and f) assess the image classification accuracy and validate the model.
The habitat metrics initially identified in this study are a) ground cover attributes (amount of leaf litter, fallen timber, fallen bark, etc.), b) presence of rock outcrops, c) presence of low-lying shrubs, d) presence of native trees, e) vegetation community or association, f) aboveground biomass, and g) selected vegetation structural metrics (e.g. canopy height, canopy cover, etc.). Attempts will be made to "combine" datasets to explore potential synergies in generating information. For instance, the use of radar imagery, bedrock geology and elevation may give information on rock outcrops and surface roughness.
The potential study sites in the southern Brigalow Belt have already been identified: a) near Rockhampton, b) south to St George, and c) west to Chesterton Range National Park. These areas were reported to have a high number of yakka skink (Richardson, 2006). Populations of yakka skink have recently been discovered at Thurshton National Park and Culgoa Floodplain National Park. These candidate sites will be assessed based on the presence of microhabitat attributes.
ALOS imagery, particularly the fine beam mode Phased Array type L-band Synthetic Aperture Radar (PALSAR) data, will be acquired through Geoscience Australia. Radar images acquired on single (HH or VV) or dual (HH+HV or VV+VH) polarisations will be purchased. The 2.5m panchromatic image (PRISM) and the10m visible and near-infrared range image (AVNIR2) will be also acquired. The PRISM imagery presents an opportunity to generate high resolution DEM to derive topographic information, such as slope, aspect, terrain ruggedness, etc. To complement the AVNIR2 data which can provide information on vegetation and non-photosynthetic vegetation (NPV) cover, we will also acquire high resolution IKONOS, QuickBird or 2.5m SPOT-5 imagery.
We will selectively employ the field methods in remote sensing as recommended by McCoy (2005). In particular, stratified random sampling strategy will be used to locate field sample plots. Initially, the ground cover theme will serve as the basis in determining the number of categories or strata (e.g. 3-4 classes) and their significance. Along with accessibility factors, available resources, and ground resolution of the image data, the strata information will help determine the total number of samples to be collected. Approximately 70% of the total number of samples will be used in training and calibration, while the remaining 30% will be set aside for accuracy assessment and validation.
Some of the measurements will involve the use of "standard" measurement techniques, e.g. determining tree height and crown diameter, measuring slope, biomass, measuring leaf litter, counting hollow logs, etc. For variables that may be difficult to quantify (e.g. rock outcrop), the literature and experts will be consulted.
Various pre-processing techniques for spatial datasets will be implemented, e.g. projection and coordinate transformation, clipping, image transformation, masking, etc. These are essential prior to data analysis, since many of the datasets will come from disparate sources, formats, and coverages. There are few statistical and modelling techniques that were developed for wildlife habitat mapping. We will examine the suitability of different techniques, such as logistic and multiple regression analyses (e.g. Sergio et al., 2004), evidential reasoning algorithm (e.g. Franklin et al., 2002), and maximum likelihood classification (e.g. Danks and Klein, 2002). For this project, we will assess various options, and possibly develop our own methods.Image accuracy assessment will be done to quantify the success of image classification. Using field data, we will construct a confusion/error matrix to generate several statistics (e.g. kappa index of agreement) indicative of classification accuracy. In validating the output map from predictive modelling, this project will use the test samples collected during the field work
The habitat metrics initially identified in this study are a) ground cover attributes (amount of leaf litter, fallen timber, fallen bark, etc.), b) presence of rock outcrops, c) presence of low-lying shrubs, d) presence of native trees, e) vegetation community or association, f) aboveground biomass, and g) selected vegetation structural metrics (e.g. canopy height, canopy cover, etc.). Attempts will be made to "combine" datasets to explore potential synergies in generating information. For instance, the use of radar imagery, bedrock geology and elevation may give information on rock outcrops and surface roughness.
The potential study sites in the southern Brigalow Belt have already been identified: a) near Rockhampton, b) south to St George, and c) west to Chesterton Range National Park. These areas were reported to have a high number of yakka skink (Richardson, 2006). Populations of yakka skink have recently been discovered at Thurshton National Park and Culgoa Floodplain National Park. These candidate sites will be assessed based on the presence of microhabitat attributes.
ALOS imagery, particularly the fine beam mode Phased Array type L-band Synthetic Aperture Radar (PALSAR) data, will be acquired through Geoscience Australia. Radar images acquired on single (HH or VV) or dual (HH+HV or VV+VH) polarisations will be purchased. The 2.5m panchromatic image (PRISM) and the10m visible and near-infrared range image (AVNIR2) will be also acquired. The PRISM imagery presents an opportunity to generate high resolution DEM to derive topographic information, such as slope, aspect, terrain ruggedness, etc. To complement the AVNIR2 data which can provide information on vegetation and non-photosynthetic vegetation (NPV) cover, we will also acquire high resolution IKONOS, QuickBird or 2.5m SPOT-5 imagery.
We will selectively employ the field methods in remote sensing as recommended by McCoy (2005). In particular, stratified random sampling strategy will be used to locate field sample plots. Initially, the ground cover theme will serve as the basis in determining the number of categories or strata (e.g. 3-4 classes) and their significance. Along with accessibility factors, available resources, and ground resolution of the image data, the strata information will help determine the total number of samples to be collected. Approximately 70% of the total number of samples will be used in training and calibration, while the remaining 30% will be set aside for accuracy assessment and validation.
Some of the measurements will involve the use of "standard" measurement techniques, e.g. determining tree height and crown diameter, measuring slope, biomass, measuring leaf litter, counting hollow logs, etc. For variables that may be difficult to quantify (e.g. rock outcrop), the literature and experts will be consulted.
Various pre-processing techniques for spatial datasets will be implemented, e.g. projection and coordinate transformation, clipping, image transformation, masking, etc. These are essential prior to data analysis, since many of the datasets will come from disparate sources, formats, and coverages. There are few statistical and modelling techniques that were developed for wildlife habitat mapping. We will examine the suitability of different techniques, such as logistic and multiple regression analyses (e.g. Sergio et al., 2004), evidential reasoning algorithm (e.g. Franklin et al., 2002), and maximum likelihood classification (e.g. Danks and Klein, 2002). For this project, we will assess various options, and possibly develop our own methods.Image accuracy assessment will be done to quantify the success of image classification. Using field data, we will construct a confusion/error matrix to generate several statistics (e.g. kappa index of agreement) indicative of classification accuracy. In validating the output map from predictive modelling, this project will use the test samples collected during the field work
