GIS applications to cultural heritage protection

The training that took place in June 2019 part of the Training in Action project focussed on advance techniques of data processing in GIS for the monitoring and protection of cultural heritage. The project is a partnership between the Institut National du Patrimoine de Tunisie (INP) and Durham University, in collaboration with the Department of Antiquities of Libya and other UK universities, University College London and King’s College London.

Training objectives

This training session focussed on the application of GIS to create risk maps for the site of Iunca that help the monitoring of the areas of the site which more under threat from modern looting. In addition, the training introduced the use of the mobile GIS application Qfield, and the database format Geopackage within the software QGIS.

Training

The first day we introduced the concept of predictive and risk maps for research purposes in archaeology and cultural heritage protection. The introduction focussed on how we can get from predictive maps that show the potential for archaeology and for threats to cultural heritage, to risk maps that show areas where there is high potential of finding archaeological remains and that are more under threat from a variety of factors. It was soon made clear how creating predictive and risk maps needs a degree of agency and decision making from the practitioners, as nothing can replace the knowledge of the territory and not everything can be quantified and represented in a map.

To practice we focussed on producing a simple risk map for the site of Iunca, in order to predict which areas are at higher risk of looting. Evidence of looting at Iunca have been recorded in during the field season September 2018, and we used this data to assess the accuracy of the models we created.

Creating predictive maps means trying to estimate human behaviour in a medium to large area, based only on quantifiable factors. The archaeological scientific community invested a lot of energy and time into exploring the full potential of predictive modelling and still there are big gaps to fill. In general, the more complex the model is the more accurate, however, sometimes human behaviour is simpler than we think and basic modelling can be quite truthful. Therefore, we decided to try with basic variables and then see how adding more factors to the model would have changed the final result.

After the introduction we discussed what sort of variables we had as GIS layers that we could use to create a basic risk map for looting at Iunca.

The two main variable used were the viability which represents the level of access to the different parts of the site; and archaeological monuments which show where the most visible archaeological material is located (Fig.1). These two factors can be used in different ways as they can be attractors or inhibitors for looters. In some cases looters might be attracted by the more accessible areas of an archaeological site, as they can reach out with vehicles which means that they can carry more and heavier material. On the contrary, some other times looter are discouraged by easy accessible areas as the likelihood of being seen is higher. Similarly, areas around monuments are likely to have more material to be looted, but in other cases, if the monument is easily accessible and open to the public that would inhibit a looter to go there.

Therefore, the two factors have to be converted into variables, thus showing how these can be attractors and inhibitors for looters. The standard way to do this is to create two maps, one that shows the distance from viability and one that shows the distance from monuments. The value of distance then can be reclassified in order to give this factors a value of attraction or inhibition.

Fig. 1. Two of the main attractors or inhibitors of looters on the site of Iunca: viability and archaeological monuments.

A third variable to consider is the presence of archaeological material on the surface as this will definitely be an attractor for looters. The data collected during the two field survey seasons in 2017 and 2018 will serve this purpose brilliantly (Fig. 2).

Fig. 2. Distribution of surface material quantities across the survey area at the site of Iunca.

Preparing the independent variables for the predictive modelling

The next step after identifying the factors to include in the model is to prepare the independent variables that will be used to create the model. In our case the variables are: 1) distance from viability, 2) distance from archaeological monuments, 3) density of surface artefacts. In order to produce the first two we performed a Multi Ring Buffer around the lines representing roads and paths across the site and around points representing the location of archaeological monuments (Fig. 3a-b).

 

 

 

 

 

 

 

 

 

 

 

Fig. 3. Multi Ring buffer calculated on a) viability and; b) archaeological monuments at the site of Iunca, Tunisia.

After rasterizing the buffers we reclassified the distance values giving them a value of risk for that specific factors. For instance we first gave high risk to areas closer to roads and paths, as they could attract more looters than other less accessible zones. Therefore, we use the Reclassify tool and assign values from 1 to 10 representing different degree of risk. Thus meaning areas closer to roads and paths with high risk (values closer to 10) and areas further from roads lower risk (values closer to 1). The same has been performed for distance from archaeological monuments, but in this case higher risk values were assigned to areas further from monuments (Fig. 4).

The first two variables are now ready and the third one needs to be calculated and reclassified so that the risk levels have the same meaning for the three rasters.

In order to calculate the density of surface material across the survey area we used a Kernel Density Estimation (KDE) which is a common tool used for these purposes in archaeological research. The KDE shows the density distribution of a specific value across the landscape under consideration – in our case the quantity of archaeological material found on the ground surface during the field survey. The density shown in Fig. 5a has been reclassified to values from 1 to 10 like the other two variables, thus all three are now comparable and can be combined into a final model.

 

 

 

 

 

 

 

 

 

 

 

Fig. 4. Risk level for a) distance from viability and; b) distance from monuments at the site of Iunca.

Fig. 5. a) Kernel Density Estimation of the survey surface material (green higher density – read lower density); b) reclassification of the surface material density with higher risk values for more dense surface material.

Predictive model and risk map for looting at Iunca, Tunisia.

 In order to produce a predictive model the three raster maps, containing the risk values (from 1 to 10) for the three threats, needs to be combined into one. In the resulting predictive map each pixel will have a value which is the combination of the three risk maps. In order to do that in GIS we used the tool Raster Calculator which allows to map algebra, so in our case to sum the three rasters. When the sum is made it is possible to choose how each variable contribute to the final model, namely giving a weight to each raster, so that one variable can be taken into consideration more than others in the sum (Fig. 6).

Fig. 6. Dialogue window of the Raster Calculator where the three variables are combined, but each is contributing differently to the final model.

The capability of choosing the contribution of each variable to the model is crucial in that this is where the knowledge of the territory by the local professional come into play. Only local knowledge of the archaeology and the territory can enhance the accuracy of the predictive model that is trying to estimate human behaviour using maps and numbers. As mentioned in the beginning this is a second point where an informed decision is taken by the practitioner who is going to benefit from the final result.

In the first test we tried with an equal contribution of distance from roads and paths (40%) and density of survey material (40%) and a lower contribution of distance from monument (20%). The first risk map is shown in Fig. 7, where the areas in red are more at risk of looting based on the three variable considered, with different weighting, as mentioned. During the third day of training after producing the first risk map the participants tried different weighting for the same three variables and different reclassification of the factors. These two are the parameters that a practitioner can play around with in order to improve the model. To test the accuracy and effectiveness of the risk map, we cross-checked it with the location of some looting pits that have been recorded on the ground during the field season in September 2018.

The discussion continued and we all agreed that the more data someone has, both in terms of potential threats to the archaeology and in terms of archaeological remains, the more accurate the model/risk map will be. This was just an initial practice to demonstrate how to produce risk maps, but of course the results were interesting, but maybe not the most accurate, considering that we only used three variables. Nevertheless, all participants were interested and deemed the topic of great interest and will continue practising on their own territories.

Fig. 7. Risk maps for the site of Iunca, indicating the areas (in red) more at risk of looting based on three variables: 1) distance from roads (40% contribution to the model), 2) density of surface material across the site (40% contribution to the model), 3) distance from standing monuments (20% contribution to the model).

QField: a mobile application for field recording

 In the second part of the training we discussed two relatively new features of QGIS. The first one is an Android mobile application called QField, which allows for “bringing to the field” the GIS project for field recording. This facilitates the synchronisation of field data with the GIS database. QField allows to load a saved QGIS project on a tablet or smartphone and to edit the attribute tables created in QGIS. Using a plugin QFieldSync it is possible to sync the data between the mobile and the desktop so to facilitate the import/export process and avoid copying mistakes.

Dr. Muftah Ahlddad gave an hour lecture and demonstrated the use of QField so that the participants could try it from the classroom. He gave examples from the site of Iunca as well and showed how they can use QField for field survey, which was one of the training topic of the project (Fig. 8a-b).

Fig. 8. a) screenshot of the mobile app QField showing the data as is was saved in the QGIS project file; b) a screenshot of the editable attribute table of the shapefiles created in QGIS.

The participants appreciated the practicality and the benefit of using such application and although there was not possibility to test it on the ground, they will test it in their own daily workflow. QField was well received also because it shows how the different strand of training are connected, in particular GIS and field survey. It shows how GIS can be used for fieldwork planning aa well as for processing field collected data.

The GeoPackage: a better alternative to shapefile for creating databases

The last topic of this training was a recent file format that has been implemented particularly in the latest 3.0 (and later) version of QGIS. GeoPackage is a data format implemented as SQLite database container within QGIS. It allows to store vector and raster layer and has much less limitations than the widely used ESRI shapefile format. For this reason it felt natural to introduce the use of it to the now advanced Tunisian and Libyan GIS users. The advantages of using GeoPackage include no limitation on field names, text field length, file size, and a better use of form and widgets for data entry. We discussed in particular the latter and we saw how to create an attribute table and create a number of widgets that facilitate the population of the form with data (Fig. 9). Figure 9 shows the form that we created in class, reproducing the condition assessment form that the KCL team prepared for the exercise at the site of Thyna the following week.

Fig. 9. Screenshot of the form that participants created for the Condition Assessment of the site of Thyna.

At the very end participants had the chance to practice all has been discussed during the week, in their own mini-project data. This gave them the opportunity to test how to adjust and customize the methods and skilled acquired in order to adapt them to different case studies. This has always been stressed during every training, namely that what we discuss in class is just an example of one possible application, but everything can (and should) be adapted from case to case.

Conclusions

 Overall, we had a very successful training session where we explored new methods and skills, fact that was highly appreciated by the trainees. The atmosphere was collaborative as usual and more advanced trainees helped the ones with struggling. At the end of the training it is clear that both Libyans and Tunisians have mastered the use of GIS in its various applications and functionalities. They are now ready to make a wide use of it in their day-to-day workflow. The mini-projects served at practising and learning not just to repeat what was done in the classroom, but to actively apply methodology and skills to the ground.