“Predictive modelling” is used throughout these Guidelines to refer to work done prior to and during cultural heritage assessment by survey and excavation to map areas more likely to contain Aboriginal cultural heritage and to plan sampling strategies to guide the archaeological sampling of landscapes to test the predictive model.
Predictive models are expected to be developed and/or adopted before, and tested by, fieldwork during CHMPs. They are expected to form part of, and be documented within, plans for standard and complex assessment and salvage.
Why emphasise predictive modelling in CHMPs?
The object of predictive modelling is to identify high-potential areas for Aboriginal cultural heritage material, reduce development risks and guide archaeological field surveys and excavation strategies. The major benefits from its application include time and cost savings from better-targeted standard and complex assessments, and better efficiency in locating Aboriginal cultural heritage.
Predictive modelling v predictive place extent mapping
These are not the same thing, although may use similar logic and method.
DPC expects HAs to cease surveying and excavating Aboriginal places once their extent, nature and significance can be adequately determined, except when proper archaeological practice would be violated.
If a HA can determine the likely extent of an Aboriginal place from partial survey or excavation, and its significance and nature are also able to be established at that stage of investigation, neither the Regulations nor DPC require additional work on that place.
Mapping of the likely extent of that place should be possible at that stage, and the logic and reasons for that mapped predicted extent should be documented along with the place registration and documented in the CHMP.
Source data for predictive models
These include SAHA mapping, Aboriginal cultural heritage registry information system (ACHRIS) mapping, previous CHMPs and previous research. Environmental mapping data should also be used. Mathematical and machine predictive models should be enhanced by traditional knowledge obtained directly from Traditional Owners and RAPs, and from CVAs.
How should predictive modelling be done?
Predictive modelling ranges from simple judgmental weighting of selected variables, to more advanced methods like regression, classification, Bayesian statistics and other machine learning algorithms. Variables analysed may include soil type, terrain attributes, proximity to water and traditional knowledge to identify patterns associated with known places.
DPC expects HAs and RAPs to use their professional judgement when selecting suitable predictive modelling approaches, work together to adopt and adapt predictive models relevant to individual CHMPs, and document these in plans for standard and complex assessment and salvage work.
Testing and documenting predictive models
All archaeological fieldwork, including cultural heritage management fieldwork, should be testing predictive models and documenting results against those models.
Documenting field results against predictive modelling in CHMPs is critical for informing future CHMPs. This includes submitting spatial data in digital form. The more this work is done and documented, the more efficient future CHMPs will be. It will also help alleviate challenges with predictive modelling such as bias, changing landscapes and accuracy.
Examples can be seen at Academia website.
Updated