Wildfires are environmental phenomena that contribute to global carbon emissions and can cause abrupt ecological changes in local environments. Within the context of a changing climate, wildfire frequency and impact are projected to worsen. The effective monitoring, characterisation and impact assessment of wildfires can be quite challenging due to their potentially large extent, which can include a variety of land cover types, fuel loads, moisture conditions and topography. Additional challenges arise also from rapid changes in intensity and direction caused by factors such as the wind velocity and humidity. To this end, collecting data that describes the different stages and aspects of wildfires is crucial for understanding and mitigating their effects.

Satellite remote sensing systems provide opportunities to monitor wildfires at a variety of spatiotemporal resolutions. Traditionally, Low Earth Orbiting (LEO) satellite sensors have been the main source of wildfire data, from hotspot detections to severity estimations. Due to their orbital limitations, however, LEO sensors have a temporal resolution (typically 12-24 hours) that is not adequate for capturing the rapidly changing course of an active fire. Meanwhile, GEOstationary (GEO) sensors are capable of capturing data multiple times in an hour but have coarser spatial resolution that reduces the ability to detect small and cool actively burning fires. Contemporary GEO sensors, such as the Advanced Baseline Imager (ABI) and the Advanced Himawari Imager (AHI), capture full-disk images of the earth every 10-minutes in a range of visible, near-infrared and thermal spectral channels, opening new pathways for high-frequency wildfire monitoring. This dissertation explores the utility of these satellite sensors for wildfire characterisation and investigates opportunities for new ways of fire impact classification, using AHI which unlike ABI has acquisition coverage over Australia. To address this aim, four research questions are posed.

The first research question examined the equivalency of Fire Radiative Power (FRP) estimates – expressed in SI units of megawatts (MW) – from LEO and GEO sensors, during the Black Summer Fires in Southeastern Australia (2019-2020). Specifically, the commonly used data products from the MODIS LEO sensor (MOD14/MYD14) and the AHI sensor (BRIGHT/AHI) were compared. The intercomparison was implemented across different geographical areas and scales, including regional segmentations, spatially and temporally continuous wildfire events and individual concurrent hotspots/pixels. Results show a high agreement between the products at the pixel level (r = 0.74), but with BRIGHT/AHI consistently underestimating FRP (by ~15%) due to its lower spatial resolution. However, BRIGHT/AHI’s temporal profile of fire activity was significantly more detailed at a regional level with up to 144 cloud free observation opportunities every 24 hours compared to MODIS, which has four observation opportunities per day. Therefore, the confidence in BRIGHT/AHI’s ability to capture equivalent wildfire detail to MODIS and reveal new insights, for an extreme event such as the Black Summer fires (2019-2020), was established.

The second research question progressed the intercomparison of the BRIGHT/AHI FRP estimations to the whole continent of Australia for an entire year, day and night, inclusive of a diverse range of land covers, burning conditions and durations. In addition to MODIS, LEO active fire detections from VIIRS (VNP14IMG) were included to further explore the effect of higher spatial resolution data in the intercomparison. The results suggest that LEO and GEO products captured similar wildfire dynamics, with a high agreement on a pixel level for concurrent detections (r = 0.74-0.77). The FRP estimations from BRIGHT/AHI, MODIS and VIIRS showed similar distributions across different land covers and regions, although with a clear positive bias for higher spatial resolution data upwards of 10 times the BRIGHT/AHI FRP estimations on average. Unsurprisingly, the regional diurnal fire intensity profiles captured by the LEO sensors demonstrated major temporal gaps between acquisitions compared to BRIGHT/AHI, especially around the times of peak and low fire activity. Examining individual localised events revealed that AHI captured a continuous stream of data that closely followed, although underestimated, all the temporal FRP peaks captured by either MODIS or VIIRS, with MODIS missing fire activity on some occasions. These findings indicate the ability of GEO data to capture active fire information accurately over large spatial scales, with an improved temporal detail over the LEO sensors.

With the capability of the BRIGHT/AHI product established, the third research question explored the association of this new stream of wildfire activity data to commonly used burn severity metrics. While burn severity has been extensively studied using bi-temporal spectral differencing indices, such as the Differenced Normalised Burn Ratio (dNBR), few studies have examined whether active fire observations capture the same aspects of fire activity as dNBR. Here, the BRIGHT/AHI FRP metrics were compared to Sentinel-2 dNBR metrics across Australia. Results reveal that the two groups of metrics were only weakly correlated for high maximum FPR fires (r = 0.33-0.39), while regional, land cover, and duration variations did not have a significant impact on the correlations. Higher correlations were only achieved after introducing different burned area classification thresholds to derive the fire fractional cover, or FFC, (percentage of an AHI pixel classified as burned by Sentinel-2 data) for each category of fire hotspots based on their fire intensity (FRP) and duration.

As the spectral differencing (Sentinel-2 dNBR) and FRP (BRIGHT/AHI) metrics capture independent aspects of wildfire activity, the fourth research question explored the combination of the two in a new wildfire impact classification method. Active fire metrics, such as the maximum FRP, the total energy emitted (Fire Radiative Energy – FRE) and the duration of the fire in a specific pixel location, were combined with the dNBR, the FFC and the pre-fire NBR (as a proxy of pre-fire vegetation health) in a dataset. Commonly used and state of the art dimensionality reduction techniques (e.g., PCA, t-SNE, UMAP, PaCMAP) were used to transform the dataset into two-dimensional projections that distributed wildfire pixels in terms of linear or non-linear associations across the six wildfire variables (Maximum FRP, FRE, Duration, dNBR, FFC, pre-fire NBR). Then, agglomerative hierarchical clustering was implemented to group the transformed data into clusters that were aggregated using ensemble clustering. The resulting ensemble clusters represented similar fires across a variety of land covers and were attributed different Composite Wildfire Impact (CWI) ratings based on their individual variable medians. The results reveal that expected fire regime patterns are effectively captured on a continental scale and over a variety of biogeographical settings. Furthermore, despite the spatial resolution differences, local burn severity assessments based on dNBR and in-situ data demonstrated a broad agreement with the proposed CWI rating. The proposed data-driven methodology can be adapted and applied to different environments globally, without the need for training data, and assist in the monitoring of fire regime patterns and trends over extensive spatiotemporal scales.

This thesis highlights the potential of contemporary geostationary satellite sensors in advancing wildfire monitoring and impact assessment. By providing continuous, high-frequency observations, GEO sensors can complement and, in some regards, surpass traditional LEO sensors in capturing the dynamic nature of wildfires. The combination of GEO fire intensity metrics with established LEO burn severity indices offers a new pathway to understanding and categorising wildfire impacts across landscapes. The methodology developed here enhances our ability to monitor wildfires on a continental scale, while it also provides a scalable and globally applicable framework for future research and operational monitoring. As climate change continues to increase the frequency and intensity of wildfires, the insights and tools provided here can inform mitigation strategies and improve our resilience to the environmental challenges of the future.