Unearthing Cultural Volcanic Landscapes Using Remote Sensing

<p dir="ltr">Lava rise and tumuli features known as stony rises are volcanic landforms that can be found on younger basaltic lava flows (<1 million years old) within the Victorian Volcanic Plain (VVP), in southeastern Australia. Stony rises are important landscape features as they have cultural importance for both Indigenous Australians and early European settlers, and ecological significance, hosting an array of native flora and fauna species. In recent years, many stony rises in metropolitan areas have been destroyed and others are under threat from various types of large-area staged land development. Currently, a limited number of stony rises are protected by heritage regulations and there are growing calls from various stakeholders including Aboriginal parties and heritage specialists for their protection.  At present, no large-scale mapping of stony rises has occurred, as mapping these landforms is limited to case study scales performed via traditional field reconnaissance. </p><p dir="ltr">Remote sensing datasets such as Light Detection and Ranging (LiDAR), aerial and satellite imagery, and airborne geophysical data sets such as radiometrics and geomagnetics provide an opportunity to detect and map stony rises over large areas from an aerial perspective. This thesis seeks to map stony rise landforms at a landscape scale using multi-source remote sensing data and machine learning.  Additionally, the land use of stony rises by Indigenous Australians is explored. To address this aim, four research questions (RQ) have been developed.</p><p dir="ltr">The first research question, presented at Victorian Archaeology Colloquium and published in Excavations, Surveys and Heritage Management in Victoria, defines metrics for characterising stony rises at a landscape scale. An extensive literature review on stony rises was conducted by analysing academic papers and archaeological reports to understand how stony rises and their physical characteristics have previously been described, often limited in detail. Then, the utility of the characteristics was measured in remotely sensed and geospatial data, where a typology of data categories was created encompassing: geology, topography, vegetation, thermal, heritage, airborne radiometric, and spectral signature and indices. This research question successfully identified metrics for characterising stony rises at a landscape scale in open-sourced remotely sensed datasets and additional spatial data. </p><p dir="ltr">The second research question, published in Geomorphology, aimed to use remotely sensed data and machine learning to map older (weathered) stony rises (~800,000 years old) from a subsection of the Mt Fraser lava flow that is experiencing rapid land development (Victoria, Australia). The metrics identified in RQ1 were used as predictor variables in a machine learning classification. An Object-based Random Forest model was implemented. First Object Based Image Analysis (OBIA) was used for image segmentation on the aerial imagery to create training data segments. Then, a Random Forest classifier was used to detect stony rise landforms in the landscape. This approach detected stony rises to an accuracy of 89.97% and found an additional 314 potential new stony rise sites. The results achieved highlight the success of using an ensemble of predictor variables to map volcanic landforms such as stony rises at a case study level using a remote sensing and machine learning approach. Additionally, it has resulted in a dataset that captures a volcanic landscape and its landforms before it is lost to land development. </p><p dir="ltr">The third research question, published in Remote Sensing, expanded on the previous RQ by mapping stony rises across a much larger region (an entire basalt lava flow). The stony rises from the Warrion Hill and Red Rock Volcanic Complex were mapped to an accuracy of 78.9%, with an additional 2,716 new stony rise sites being identified. This previously unmapped region is much younger in geological age (~42,000 years old) and highlights that the age of the lava flow needs to be considered when mapping stony rises with remote sensing and machine learning, as younger volcanic landscapes rely more on topographic predictor variables compared to older ones. </p><p dir="ltr">The final research question, currently under peer review in Australian Archaeology, explored the land use of stony rises by Aboriginal Australians.  Here, the spatial association between stony rises and Registered Aboriginal Places was examined through clustering and spatial analysis. The results generated from RQ2 were used to explore the cultural values of stony rises from the Mt Fraser lava flow in Wurundjeri Woi-wurrung Country. Here, the distance from stony rises to Registered Aboriginal Places was calculated using buffers of 50 metre increments up to 1 kilometre. Additionally, the spatial distribution of Registered Aboriginal Places within the study area was examined through a Kernel Density map. The results found that 41.4% of Registered Aboriginal Places were found on a stony rise, and an additional 17.7% of Registered Aboriginal Places were found within 50m of a stony rise. The strong spatial association found between stony rises and Registered Aboriginal Places contributes to knowledge regarding the importance of these natural landforms to Indigenous Australians, particularly the Wurundjeri Woi-wurrung people. </p><p dir="ltr">This thesis investigated mapping volcanic landforms (i.e., stony rises) with multi-sourced remote sensing data and machine learning. Defining stony rise landform metrics and where they are found using remotely sensed data, allowed a definition of stony rises to be developed. Their subsequent examination with regards to places of cultural significance highlights the importance of these landscape features for Indigenous Australians who advocate for their protection. As land (particularly in metropolitan areas) continues to be developed, there is a need for more efficient landform mapping techniques. The methodology implemented in this thesis can be applied at case study and landscape scales to identify volcanic landforms (such as lava rises and tumuli) and enable targeted mapping and landscape analysis. While land development is inevitable, it is important that the geological, ecological, and cultural values of a landscape, such as those studied in this thesis, are recognised, studied, and preserved (where possible) for future generations before they are lost forever.</p>