Landscape modelling of Aboriginal site features and their application in regional heritage planning'

1
Landscape modelling of Aboriginal site features
and their application in regional heritage
planning.

• Mal Ridges

2
Need for regional scale models

• Many planning decisions are made at the landscape
  scale
• Demand for rapid responses in planning contexts
• Requirement to consider inter-generational equity
  (ie accumulated impacts)
• Interpreting the interplay between relative
  likelihood, detectability, and historical impacts

3
Working on a suite of modelling products

• Application of presence only GAMS
• New variable types for soils vegetation
• Pre-1750 models of feature distribution
• SGA analysis for model reliability
• Land-use parameters for current likelihood
• Accumulated impacts
• Survey priority
• Classification into archaeological regions

4
1. GAMS- with GRASP
OUTPUTS
Variable contribution plots
Non-linear logistic regression technique
(Generalised Additive Modelling) Implemented in
S-Plus with GRASP
FULLY RANDOM
Non-site sampling method when faced with heavy
survey bias)
STRATIFIED RANDOM
Fitted model curves standard error margins
Sites
ABSENCE IN PREVIOUS SURVEY
Surveyed areas
5
2. New variables
Explores landscape aesthetics
Visibility
Cost-distance from streamline
Better measures of water availability
Helps address survey bias transforms
categorical variables in ratio variables
Turnover in soil physical properties
Cost-distance from streamline weighted by
stream-order
Turnover in Pre1750 vegetation
6
3. pre-1750 models validation

• Validation involves
• Ground truthing
• Application in survey design
• Research (land-use history)

Bongil Bongil
Molong TSR assessment
7
4. SGA analysis for model reliability
The SGA algorithm measures improvement gained in
representing model input variables by iteratively
adding each non-sample point calculating the
change in distance between P1 P2
P-median for entire multivariate space P1
P-median for artefact sample P2
P-median is calculated as the weighted median in
multivariate space, where the weight on each
demand point is model likelihood. The
contribution each variable makes to the P-median
is also weighted by the contribution it makes
within the model
8
5. Land-use parameters for current likelihood
Original likelihood
Parameter matrix for land-use


Land-use layer
Current likelihood
9
6. Accumulated impacts
Low impacts
Impacts on each feature (Original Current)
SUMMED
Summing the impacts on individual features
produces a map of accumulated impacts on sites
over the whole landscape. This can provide
context for inter-generational equity.
High impacts
10
Land-use parameters
7. Survey priority
SGA result for each feature model
X
SUMMED

• The SGA result highlights those areas in the
  landscape that are under represented for each
  feature.
• Combining it with land-use parameters identifies
  priority survey areas allowing for impacts

High priority
Low priority
11
8. Classification
In the same way satellite images can be
classified into classes, the stack of feature
models can be classified into regions based on
similar suites of feature likelihood
12
Summary

• Generating one model to cover everything over-
  simplifies a complex problem
• The suite of spatial products presented here
  increases applicability interpretability of
  archaeological predictive modelling
• Complexities with model validation in the context
  of land-use history is a key research objective
• Such research is critical for estimating
  appropriate parameters to use with these
  techniques