Robusta – An Application for Quantifying Uncertainties in Spatial Analysis in Archaeology

For the best experience on the web app, we recommend zooming out the browser window to 75% or 80%.

Introduction

Based on the paper by Herrera Malatesta and de Valeriola (2024), this web app allows interested researchers in computational archaeology to reproduce the paper’s results and quantify the uncertainty on their own spatial datasets.

In essence, Robusta aims to help colleagues who desire to better understand their models by quantifying their uncertainties. For an in-depth description of the framework, see the tab “Methods”.

Robusta exists both as an interactive web app and as an open-source tool that can be downloaded and run locally. The web app is primarily meant as a demo, and can become unstable when running bigger computations.

We therefore recommend that users download the app if they want to play around with their own datasets and larger numbers of simulations, and only use the test dataset and a low number of simulations when using the web app.

To download the app, go to this GitHub repo.

On Robusta you can either upload your own data for processing or use the Montecristo data used by Herrera Malatesta and de Valeriola (2024).
The Montecristi data is meant as a test dataset, where you can play a bit around with the parameters and see potential outputs.

(Optional) Upload your own data
When using your own data, the dataset should be in the shapefile format and consist of a series of georeferenced points, in addition to a georeferenced polygon of the research area.
Choose the tab “Upload” to upload the following shapefiles:

A vector file of the points to be analysed (e.g., sites, materials). You need to upload a .shp and a .shx file.
A spatial polygon file of the research area. This also needs to consist of a .shp and a .shx file.

Choose number of simulations and scenarios
After specifying data (test data or own data), you need to specify the number of Monte Carlo simulations and the number of Robustness Scenarios.
On the web app we recommend the following values:

For Monte Carlo simulations: values between 25 and 200.
For the Robustness Scenarios: values between 5 and 100.

When running locally we recommend the following values:

For Monte Carlo simulations: values between 1000 and 5000.
For the Robustness Scenarios: values between 100 and 3000.

Please note that these values are based on the standard computational power of a laptop, and increasing the values to even larger numbers can lead the app to behave in unpredicatable ways.
However, users should feel free to play around with the values to see how it changes the results, as well as modifying the source code to their needs.

Select Quantiles
As the final step, you need to select which quantiles to run with. In the Quantile dropdown menu, you will be presented with the option of choosing one or several quantiles. We recommend that you play with the options, but if you are looking for a comprehensive comparison of the results, choose all available quantile options.

IMPORTANT
Increasing the size of the free parameters can significantly increase processing time, meaning the calculations can take anything from a few minutes to several hours. Please be patient and do NOT refresh the page, as all data and processing will be lost. A progress bar will pop up to show how far the computations are. Due to the large processing time, we recommend downloading the results if you would like to return to it.

Results

The processing will produce 3 tabs:

PCF based on 100% of sites
Robustness PCF, with additional tabs for the different subgroups
Comparison Tools

Read more about these in the “Methods” tab.

NOTE: When downloading the results, only five robustness scenarios are downloaded and not all of the selected ones. This is because the web app is just for testing, and users are encouraged to use the local version of the app that can be downloaded from GitHub.

Creators and Contact

This web app was created by Niels Aalund Krogsgaard and Mia Jacobsen (miaj@cas.au.dk) from the Centre for Humanities Computing (Aarhus University) based on the research and codes by Eduardo Herrera Malatesta and Sébastien de Valeriola.

For more information about this web app or contact to its creators please contact:

Eduardo Herrera Malatesta, Centre for Urban Network Evolutions (Aarhus University) / Faculty of Archaeology (Leiden University) - ehmalatesta@yahoo.com
Sébastien de Valeriola, QuaDiHum Lab (Universite libre de Bruxelles) - sebastien.de.valeriola@ulb.be
Ross Deans Kristensen-McLachlan, Centre for Humanities Computing (Aarhus University) - rdkm@cas.au.dk

Funding Information

This research has received funding from Horizon Europe, HORIZON-MSCA-2021-PF-01-01, Marie Skłodowska Curie Action, Grant agreement n˚ 101062882. Granted to Eduardo Herrera Malatesta.

Framework and background

This app follows the framework proposed in Herrera Malatesta and de Valeriola (2024). This framework, called the Robustness Assessing Framework, focuses on assessing the robustness and uncertainties of conclusions drawn from applying point pattern analysis to, mostly, non-systematic regional data in archaeology. To achieve this, we have articulated the discussion on the reconstruction of past landscapes using computational methods around three key aspects:

the use of point pattern analysis (PPA) in archaeology
the quantification of uncertainties
the consideration of robustness.

The framework and subsequently this app are designed to aid archaeologists working with datasets that are known to contain sources of uncertainty in applying spatial statistical methods and achieve a higher understanding of the uncertainties of the resulting models.

Methodology

The framework itself consists of three simple steps:

Step 1 - The Observable

The first step is defined as “the observable” which is a point clustering metric whose changes are tracked and measured when considering deviations. In this case, this is the model resulting from a Pair Correlation Function (PCF) with a Monte Carlo simulation envelope based on 100% of the dataset used as input. The app will provide the result of “the observable “which will be used as a reference value for the second step.

Step 2 - The Experiment

The second step is “the experiment” where regular intervals of data will be sampled from the loaded dataset. More precisely, the app will deduct 10%, 20%, 30%, 40%, and 50% of the database’s sites and perform the PCF again with a Monte Carlo simulation envelope for each of the sampled groups. The resulting models are what we have called the “robustness scenarios”.

Note that in our paper we used two sampling methods, one using a uniform distribution and another one using an inhomogeneous distribution. To make the app work optimally, only the sampling via the uniform distribution is provided. If a more experienced researcher would like to access the code and consider the inhomogeneous distribution, please check out the original code on OSF.

Step 3 - Comparison Tools

The third step consist of the “comparison tools”, which are methods to assess the frequencies and interval midpoint densities. The comparison tools is the step that will allow the analyst to assess the robustness and quantify the uncertainty of the spatial models created based on their dataset.

Results

The figure of the first comparison tool will present the percentages of sites that are kept in each robustness scenarios against the percentage of robustness scenarios in which the conclusion is similar to “the observable”. This figure provides a direct percentage of the probability that, by extracting a particular percentage of the dataset, the results of the robustness scenario will be similar to “the observable”.

The second comparison tool goes deeper into understanding these patterns and provides insight into what can be further observed from the clustering patterns. At this point, it is important to clarify that the app is only analyzing the statistically significant cluster patterns on the PCF, and not the regular ones. Again, if a researcher would like to include this in their analysis, they need to modify the source code.

Original data

The Monticristi data used for the “Test” option is available at DANS and is produced by Herrera Malatesta (2017).

Original paper

As stated this web-app is build on the paper Ambiguous landscapes: A framework for assessing robustness and uncertainties in archaeological point pattern analysis by Herrera Malatesta and de Valeriola (2024). The paper can be found at PlosONE, and accompanying code and data can be found on OSF.

Code repository

The code for this web-app has been developed by Niels Aalund Krogsgaard, Centre for Humanities Computing and can be found at GitHub.

Data privacy

No data or results are saved permanently on the server. Everything is only kept momentarily, and will be deleted upon refreshing the browser. The web app is hosted on Aarhus University servers located in Denmark.

Settings

Simulations with Montecristi Dataset

Simulations with uploaded dataset