Quantifying the uncertainty of a model-reconstructed soilscape for archaeological land evaluation

Citation data:

Geoderma, ISSN: 0016-7061, Vol: 320, Page: 74-81

Publication Year:
2018
Usage 2
Abstract Views 2
Captures 9
Readers 9
Citations 1
Citation Indexes 1
Repository URL:
https://ro.uow.edu.au/eispapers1/1145
DOI:
10.1016/j.geoderma.2018.01.032
Author(s):
Finke, Peter; Jafari, A; Zwertvaegher, A; Thas, Olivier
Publisher(s):
Elsevier BV
Tags:
Agricultural and Biological Sciences
article description
Models can serve as temporal interpolators for landscape reconstruction, and geostatistical interpolation can assist in obtaining full-cover maps. Both activities are associated with uncertainty. We applied a hydrological model (MODFLOW), a soil development model (SoilGen) and regression kriging to reconstruct a Middle Bronze Age landscape. This reconstruction comprises maps of a set of land characteristics and is input for a land evaluation meant to explain low occupancy during this period. The land evaluation showed marginal suitability in 81% of the area for Bronze Age land utilization types, which explains the low occupancy. We evaluated the effects of model accuracy and interpolation accuracy on the uncertainty of the land evaluation, assuming no uncertainty about the land evaluation protocol. The effect of model (in)accuracy was assessed by generating 100 perturbations using a variance-covariance error matrix filled using independent measurements from 1953 CE. The effect of spatial interpolation errors was assessed by generating 100 realizations of the land characteristics by Sequential Gaussian Simulation. Results show that uncertainty contributions of the model errors and interpolation errors are similar in terms of entropy H′ between land evaluation maps (fairly low H′ of 0.47 and 0.38 for model and interpolation error respectively). Model-reconstructed land characteristics led to fairly reliable land evaluations. Priority improvements for the model were identified and ranked. Most improvements come at high computational cost.