Polar-orbiting sensors are often preferred for wildfire monitoring due to their finer spatial resolution. However, next-generation geostationary wildfire observations can offer much more frequent information about wildfire activity. In this study, we use geostationary wildfire observations from Himawari-8 across the Australian continent (1 year) to derive additional information that can assist with wildfire characterization. We use the active fire information to derive categorize fires into groups and examine their spatial distribution. For each fire type, we explore whether the inclusion of active fire data can offer information that can complement commonly used burn severity indices. We argue that the two streams of data offer significantly diverse information that together, could further our understanding of wildfire impact. Future burn severity indices should incorporate fire intensity estimations in their derivation, as current techniques are often limited by pre-fire conditions, a dependency on field measurements, and are not straightforward to generalize.

Elevated areas on basalt lava flows, locally known as stony rises, are volcanic landforms located in the Victorian Volcanic Plain (VVP) in southeastern Australia. Stony rises are recognized as having high geological, ecological, and cultural values. Currently, the mapping of stony rises is carried out on a study-by-study basis, via traditional field surveys. Mapping of such features could be performed more rapidly and at a landscape level by utilizing remotely sensed datasets such as LiDAR, satellite and aerial imagery, and geophysical data, and applying detection techniques such as machine learning. This study incorporates an ensemble of remotely sensed data and predictor variables to characterize stony rise morphology through a machine learning approach that uses an object based Random Forest. We were able to successfully detect stony rises across various lava flows in the VVP. Future geomorphological studies should aim to incorporate the use of spectral and geophysical predictor variables for large area landform detection and analysis.

The 2-D Point Spread Function (PSF) of satellite imaging sensors is usually estimated from two perpendicular edge measurements. It has been shown that this method is only valid for a sensor with a low optical factor “$Q$” (defined as the wavelength times the F number divided by the pixel pitch), so a new method is required to estimate the PSF for sensors with moderate and high$Q$. In this work, a new three-edge method that estimates the PSF by quadratic interpolation in the spatial frequency domain is assessed and shown to complement the current two-edge method. The use of both methods allows the estimation of the PSF down to one order of magnitude below its peak response. The new method is assessed using a generic sensor system methodology that considers the optical design parameters of the sensor as independent variables. Results are graphically represented as constant Mean Absolute Percentage Errors (MAPEs) contours drawn on a plane of optical designs in which each point represents an imaging channel with a specific$Q$and optical aperture’s obstruction ratio.

Several types of analytic models are currently used to estimate the spatial response of satellite imagers, the accuracy of these models being critical for applications requiring precise knowledge about the spatial response of a given imager. The assessment of these models is complicated because the actual spatial response of an imager depends on its optical design, so evaluations based on a single kind of design are inherently biased. To reduce this bias, a new assessment methodology based on a generic imaging sensor is proposed; the key optical design parameters of this sensor are selected as independent variables, so the error of any spatial response model can be computed within a broad domain of possible optical designs. Assuming a generic sensor with an annular optical aperture and square detector elements, the optical factor$Q$and the aperture obstruction ratio$\varepsilon $are selected as key design parameters, allowing the error of spatial response models to be computed in the ($Q$,$\varepsilon$) plane. This approach is used to assess the separable point spread function (PSF) model, which assumes that the PSF is equal to the product of two perpendicular line spread functions (LFSs), concluding that it is only valid when$Q \le0.35$for PSF$\ge0.1$. This methodology can be used to assess other types of spatial response models for different shapes of optical apertures and detect elements. We contend that our approach provides a standard assessment procedure that will help end users select the correct model for their specific application.

This article introduces a new algorithm to detect active fires in geostationary remotely sensed data. The algorithm calculated dynamic statistical multispectral thresholds based on, and sensitive to, biogeographical region, subseason, and time-of-day. The spectral characteristics of nonfire and noncloud mid-infrared values were found to vary with biogeographical region, subseason, and time-of-day. These differences were exploited to define a new seasonal Biogeographical Region and Individual Geostationary HHMMSS Threshold (BRIGHT) multivariate adaptive threshold geostationary satellite fire anomaly algorithm. The algorithm was demonstrated on 12 months of daytime data acquired from the geostationary satellite system, the Advanced Himawari Imager (on Himawari-8) over Australia (7.69 million km2). The resulting hotspots were compared with those from the existing Moderate Resolution Imaging Spectrometer (MODIS) polar-orbiting fire-hotspot algorithm. The intercomparison showed that BRIGHT wildfire hotspots, detected using Himawari-8 and Interim Biogeographic Regionalisation of Australia (IBRA) data, were also detected by MODIS polar-orbiting fire hotspots 88% of the time. While MODIS hotspots were detected by BRIGHT hotspots only 39% of the time; the majority of the undetected MODIS hotspots had low radiative power. BRIGHT provides a new method for the remote sensing of active fires providing reliable observations at spatial and temporal scales useful for fire managers.

Estimating change in forest biomass is important for monitoring carbon dynamics and understanding the global carbon cycle. Multi-temporal airborne lidar data has been recently used to accurately predict change in forest attributes such as aboveground biomass (AGB). In this study, we assessed the ability of multi-temporal airborne lidar (2008 and 2016) and single-date inventory data to estimate forest biomass dynamics. To do so, we compared different imputation approaches to predict biomass, specifically direct (i.e., a model trained by the biomass variable or AGB) and indirect (i.e., a model trained by structure variables - basal area, tree volume and stem density) approaches. We also evaluated the ability of the selected model in temporally estimating biomass by relating biomass predictions with forest disturbance data. Our results demonstrated that AGB can be better predicted using an indirect imputation method in which lidar metrics were trained by a structure variable (basal area, RMSE = 95.09, R2= 0.89). While the model was developed for the date of inventory measurements (2016), the model was successfully applied to predict biomass for a historical date (2008). For both years, biomass predictions were highly consistent with disturbance history. This study further informs the benefits of multi-temporal lidar data to estimate forest biomass dynamics in instances when only single-date inventory data are available. The work thus can support forest researchers and managers in improving their scientific and practical tasks in forest management.

Estimation of aboveground forest biomass is critical for regional carbon policies and sustainable forest management. Both passive optical remote sensing and active microwave remote sensing can play an important role in the monitoring of forest biomass. In this study, the recently launched Sentinel-2 Multi Spectral Instrument satellite and Sentinel-1 SAR satellite systems were evaluated and integrated to investigate the relative strengths of each sensor for mapping aboveground forest biomass at a regional scale. The Australian state of Victoria, with its wide range of forest vegetation was chosen as the study area to demonstrate the scalability and transferability of the approach. In this study aboveground forest biomass (AGB) was defined as the tons of carbon per hectare for the aboveground components (stem, branches, leaves) of all live large trees greater than 10 cm in diameter at breast height (DBHOB). Sentinel-2 and Sentinel-1 data were fused within a machine learning framework using a boosted regression tree model and high-quality ground survey data. Multicriteria evaluations showed the use of the two independent and fundamentally different Sentinel satellite systems were able to provide robust estimates (R2of 0.62, RMSE of 32.2 t.C.ha-1) of aboveground forest biomass, with each sensor compensating for the weakness (cloud perturbations and spectral saturation for Sentinel 2, and sensitivity to ground moisture for Sentinel 1) of each other. As archives for Sentinel-2 and Sentinel-1 continue to grow, mapping aboveground forest biomass and dynamics at moderate resolution over large regions should become increasingly feasible.

The extensive and freely available archive of Landsat satellite data is used throughout the world to assess forest changes over large areas and long time periods (30-40 years). But analyzing Landsat data in time-series is not free of challenges (e.g. data processing and storage capabilities, dealing with cloud cover and other data gaps, and accounting for changes in illumination conditions due to atmospheric effects, sun angle and vegetation phenology). In this research, we present a method used to create annual seamless cloud-free mosaics for the entire state of Victoria, Australia (19 Landsat tiles), for a 30 year period. These mosaics were created by first constructing yearly Best Available Pixel (BAP) composites from over 3000 individual scenes. Then, forested areas were analyzed in time-series to determine breakpoints (e.g. a disturbance event such as fire). Following this, the breakpoints were used to fit a piece-wise linear regression model through each pixel's temporal trajectory. In this way, data gaps and other radiometric anomalies were removed. These gap-free composites can be used by a variety of stakeholders for land management, statutory reporting and decision making activities. This ensures state-wide consistency, and offers significant savings in processing and storage requirements.

The utility of Geostationary active fire detection and surveillance has recently been supplemented by two new algorithms developed by our group: the AHI-FSA (Advanced Himawari Imager - Fire Surveillance Algorithm) and the Broad Area Training (BAT) method [1], [2]. Here we present results from a large area validation of these products to support wildfire surveillance and mapping using the geostationary Himawari-8 satellite. The AHI-FSA/BAT algorithms were developed and tested on a number of case study areas in Western Australia. Initial results demonstrate a high potential as a wildfire surveillance algorithm providing high frequency (every 10 minutes) fire detections. However, the AHI-FSA and BAT products need to be validated over a large area to quantify the performance of the algorithms. This paper validates their performance in the Northern Territory of Australia (1.4 million km2) over a 10 day period by comparing AHI-FSA/BAT to well-established products from LEO satellites: MODIS (Moderate Resolution Imaging Spectroradiometer) and VIIRS (Visible Infrared Imaging Radiometer Suite). This paper also discusses difficulties in validating high temporal resolution products with existing low temporal resolution LEO satellite products. Results indicate that the multi-resolution approach developed for AHI-FSA/BAT significantly improves fire detection. When compared to the MODIS thermal anomaly products, BAT omission error was only 2%. High temporal frequency data results in AHI-FSA/BAT detecting fires, at times, three hours before the MODIS overpass with much-enhanced detail on fire movement.

This work introduces new ensemble margin criteria, to evaluate the performance of Random Forests (RF), in the context of large area land cover classification, using imbalanced and noisy training data. Experiments using binary and multiclass classification problems reveal insights into the behaviour of RF over big data, in which training data contains noise and may not be evenly distributed among classes. The margin-based RF performance evaluation is conducted using remote sensing and ancillary spatial data, across a 7.2 million hectare study area.

Prescribed burning, a common fire management practice is routinely carried out by government departments and land management agencies to reduce wildfire hazard. To understand and measure the impacts of these fires on the landscape, spectral data from Eucalypt trees of an Australian dry sclerophyll forest was captured pre- and post-burn at two spatial scales. Results of this research indicate that NDVI was able to detect significant epicormic growth in response to the burn event at the object scale. The timing of NDVI decrease and returns were comparable at the two spatial scales considered in this paper.

There is an urgent need within the international remote sensing community to establish a metadata standard for field spectroscopy that ensures high quality, interoperable metadata sets that can be archived and shared efficiently within Earth observation data sharing systems. Careful examination of all stages of metadata collection and analysis can inform a robust standard that is applicable to a range of field campaigns. This paper presents approaches towards a standard that encompasses in situ metadata collection and initiatives towards sharing metadata within intelligent archiving systems.

Collaborative ventures in research infrastructure can allow multiple stakeholders to benefit from outcomes that may otherwise be cost prohibitive. In this study, we discuss how the investment in research infrastructure by various sectors of the academic, scientific, and land management community is promoting high end forest ecosystem research in Australia. Three 25km2woodland and open canopy forests that are representative of Victoria's 8 million hectares of public forests were incorporated into the Terrestrial Ecosystem Research Network's calibration/validation campaign. The sites are being used to develop algorithms that will assist land management agencies across various states to characterize fundamental forest attributes at a landscape level. Wireless technology (VegNet) is also being trialed at these sites to investigate forest condition over time. This study provides an example of how the establishment and co-investment in research infrastructure amongst different sectors of the scientific community promote data sharing and ultimately expand our understanding of forest ecosystems, which can in turn be used for monitoring and to inform policy and land management decision making.

There is a need for accurate estimation of Australian woody vegetation parameters. State and Commonwealth land management agencies are mandated to report about forest condition every five years. The CRCSI 2.07 “Australian woody vegetation landscape feature generation from multi-source airborne and space-borne imaging and ranging data” aims at producing ready-to-use methods to report forest condition based on remote sensing data. The first efforts have focus on field data techniques and canopy structure characterization using LiDAR data. Results demonstrate canopy profile can be accurately estimated using Weibull probability density functions at 30×30m pixel size. Moreover different field techniques to measure vegetation fractional cover has been tested and compare finding differences up to 15%.

MAUP theory is applied to a LiDAR dataset acquired over a forested scene. The Weibull Probability Density Function (PDF) has been fit to LiDAR derived canopy height profiles for plots covering the complete 1 × 1 km scene. Ten plot sizes are tested from 10 – 300 m. Parameters describing the location and scale of the PDF are used as analogous of canopy height and canopy length respectively. Results suggest that, for a structurally homogenous forested scene, localised variance decreases for canopy height with increasing plot dimensions. The opposite is apparent for canopy length, it is suggested this is a result of a spatially heterogeneous understorey layer negatively skewing the distribution.

Calibration and validation of LAI products require accurate ground-based measurements. Many indirect ground-based sensors such as digital hemispherical photography (DHP), ceptometers, and terrestrial laser scanners (TLS) are used interchangeably to estimate reference values. However these sensors have biases in regards to the true LAI value, which can never be known in the field. Results from three representative woody ecosystems in Eastern Australia are presented from real field measurements. Significant differences were found between methods at the individual measurement and plot scale. Furthermore, one of the sites in South East Australia was measured and modeled in a 3D deterministic model. In this digital environment where the truth is known, sensors can be simulated to determine their bias.

Responding to the threat of wildfire on human and natural resources is important, particularly in countries such as Australia where large areas of land are required to be observed and managed. Remote sensing techniques are an obvious choice in providing fire related information over large areas for emergency services in responding timely to outbreaks of wildfires. The objective of this study was to compare the utility of the MODIS fire products, MOD14 and MCD45, against the experimental small satellite BIRD, using Landsat ETM+ as the validation dataset. The results show that MODIS has the highest accuracy for detecting active fire fronts, with 91.5% compared to BIRD with 83.1%, but that BIRD outperformed MODIS in burn scar detection with 81.5% and 76% compared to MODIS which had 49.5% and 65% for the respective tests.

This paper reviews the utility of polygon centroid algorithms for natural resource management applications. Following a review of ‘standard’ centroid derivation algorithms, and their limitations, alternative approaches are proposed. Building upon current polygon-reduction algorithms an iterative edge-erosion model is presented. Resilient centroids derived via this model are shown to: (1) maximise the distance to all internal and external edges; (2) be located within the polygon boundary; and (3) be insensitive to patch vertex outliers. The resilient centroid is demonstrated to have utility in a range of natural resource management applications including the matching of disparate data sources. Exemplar studies conclude that resilient centroids support an appropriate allocation of vegetation condition point samples both within and between native vegetation patches. Further, the utilisation of resilient centroids enables the (1) sub-division of vegetation patches into homogenous regions (core areas) and (2) interrogation of the relationship between vegetation condition and core area proximity.

With the establishment of the Australian Terrestrial Environmental Research Network (TERN) and the Integrated Marine Observing System (IMOS) Australia will significantly develop, and better coordinate, its activities in the calibration and validation (cal/val) of earth observation data in both the terrestrial and aquatic arenas. Consequently, there is the opportunity for Australia to make a systematic contribution to a range of international satellite missions. TERN will establish a cal/val program to provide for the calibration and validation of a range of land surface products. TERN AusCover is a new nationally consistent approach to collecting, validating and distributing timeseries of remote sensing-derived terrestrial land cover products to meet the requirements of ecosystem research within Australia. This paper will present a status report on calibration and validation activities in Australia. The latest developments in the TERN AusCover calibration and validation sites both proposed and established for autonomous monitoring will be outlined.