GIS 09_2019

Training in advanced GIS 22nd – 24th September 2019 Mahres (Tunisia).

The training that took place in September 2019 focussed on advance techniques of data processing in GIS for the monitoring and protection of cultural heritage. This training session focussed on two main topics: processing of data collected with the Unmanned Aerial Vehicle (UAV) within a GIS platform and the development of a methodology to create predictive risk maps for the region of Sfax. The training duration was three days, but all the topics have been covered successfully.

Day 1

This first day was dedicated to the processing of data collected with the drone within QGIS in order to enhance the visibility of archaeological features and to map features of interest. The preceding week participants familiarised with the drone data collection and discussed all the steps from flying to data outputs. The two main outputs of drone data are the orthomosaic and the digital elevation model (DEM). To produce these outputs from the images collected with the UAV, participants worked with Agisoft Photoscan. But in order to use this data for archaeological applications they needed to learn how to process it in a GIS platform.

Because of the high resolution of the drone imagery we used only a small section of the data collected that refers to the area around the fort at the site of Iunca, which everyone is familiar with. After importing the data into QGIS we discussed different ways of displaying the orthomosaic to make the features of interest more visible. In particular we stressed the importance of using such high-resoluation imagery in combination with all the other methods of archaeological documentation because these allow to map ephemeral remains that could easily be overlooked by satellite imagery or field walking. In the particular case of Iunca we saw how it is possible to distinguish and better visualise the remains of big structures (Fig. 1-2) as well as more ephemeral remains of walled structures across the landscape. It was clear to the participants how drone data can help even the simple mapping of a structure that is hard to reach from the ground and is not visible from high-resolution satellite imagery to the same detail as from the UAV (Fig. 3).



Fig. 1. UAV orthomosaic covering the area around the fort of Iunca, Tunisia.                         Fig. 2. Detail of an ‘unknown’ structure to the southwest of the fort at Iunca, Tunisia.



Fig. 3. Digital Elevetion Model (DEM) derived from UAV data at the site of Iunca.                    Fig. 4. Hillshading effect calculated for the DEM of the fort at the site of Iunca, Tunisia.

Subsequently, we discussed how to process the digital elevation model in order to visualise and semi-automatically extract archaeological features of interest. The utility of digital elevation models derived from UAV data relies on the fact that they can expose micro-topographic details which can reveal even ephemeral archaeological remains. The processing of DEMs within Photoscan is limited so we discussed a number of ways to enhance the visibility of archaeological features within QGIS.

At first, we covered the basic processing of DEMs such as hillshade and slope topographical analyses that provide with a clearer image to observe. Already with hillshading (Fig. 4) it possible to identify features of interest. The main problem with hillshading though is that it is dependent on the direction of the artificial light that it is generated in order to produced shadows; namely if a linear feature (e.i. a wall) sits aligned with the direction of the light, it won’t generate a shadow, hence it won’t be visible. Therefore, slope is a much better way of displaying DEMs as it is not direction-dependent – features with all orientations are clearly visible (Fig. 5). Moreover, with a simple reclassification of the slope values displayed it is possible to highlight features of particular interest such as upstanding wall, as they are characterised by steep slope values (Fig. 6)


Fig. 5. Slope topographical analysis of the DEM of the area around the fort in Iunca.            Fig. 6. Reclassification of the slope values in order to enhance the visibility of areas with slope                                                                                                                                                               higher than 8 degrees (in red). Which allows for the detection of high wall like the ones at the fort.                                                                                                                         

Further in the analysis of the DEM we also considered to enhance the local micro-morphological detail of the model in order to display more clearly the features of interest.

The Local Relief Model (LRM) usually applies to Lidar data, but it can be used for analysing UAV high-res DEMs too. It allows the enhancement of local micro-morphological details that often represent archaeological remains or in general features of interest. This procedure is based on smoothing the original DEM through low pass resampling filter using different kernel size (e.i different smoothing effects). The difference between various smoothing effects enhance the visibility of those features removed during the loss pass filter such as walls and other sharp linear features (e.i. roads, structures). It is a technique applied in archaeological research for automatic extraction of feature of interest from digital terrain models DTMs.

We applied a simplified version of LRM to the UAV data collected at Iunca. A Low Pass filter using a Kernel size of 5×5 was first performed on the original DEM, followed by a Low Pass filter of 9×9. The difference between the two was then calculated and result can be seen in Fig. 7.

Fig. 7. Local Relief Model (LRM) applied to the DEM derived from UAV data collected at the site of Iunca, Tunisia.

Overall, the first day covered a number of different techniques for the visualization of the two main outputs of drone data, orthomosaic and DEM. Participants seems appreciative of the benefits that these techniques can bring to the mapping of archaeological features across a large site such as Iunca.


Day 2

It  was dedicated to the processing of satellite imagery for the classification of different land uses. In particular, we were interested in mapping urban areas and agricultural lands, and how they changed over the last 40 years. In order to do that we used Landsat imagery form two different missions 5 and 8. The Landsat mission 5 provides imagery from the 1980s – early 1990s, whereas Landsat 8 is the current mission collecting data everyday. The choice of these two datasets was dictated by that fact that they cover a long time lapse and therefore significant changes in the land use can be appreciated. We focussed on the region of Sfax as this is the project’s sample area, which also contains the site of Iunca which has been used for all the field activities in the last 2 years of the project.

At first we discussed how to import and manipulate satellite imagery in QGIS and how to handle multi-band datasets. Subsequently, we familiarised with the Semi-automatic Classification Plugin (SCP) of QGIS that allows to managing data download, pre-processing, processing and post-processing – Dr. Moftah Ahlddad gave a brief introductory lecture on the main functionalities of the plugin (Fig. 8).

At first we looked at how to classify vegetation in both sets of imagery using standard vegetation indexes such as Normalised Difference Vegetation Index (NDVI) which can be calculated in every GIS or Remote Sensing software. The NDVI algorithm uses the red and the near-infrared bands to enhance the visibility of vegetation, and specifically young vegetation with fresh chlorophyl (Fig. 9). This is useful to detect areas with new vegetation or crops as they will have plants with high content of chlorophyl. Vegetation and agriculture are one of the main threats to the archaeology across the world including arid countries like Tunisia and Libya. Therefore, large-scale mapping of vegetation growth is very useful for archaeologists and heritage professionals working in regions like Sfax, as it provides a simple tool to monitor the level threat that vegetation is causing to archaeological sites and landscapes. Being NDVI a commonly used standard procedure makes it a sustainable way of monitoring vegetation growth through time. Analysing how vegetation and agriculture expanded in the last 40 years allows to understand the future pattern of growth in the near future. We tested this by calculating a NDVI for the Landsat 8 image that dated to two weeks prior the training course and compare the two. By simply calculating the difference of 1987 NDVI and 2019 NDVI it was possible to clearly distinguish areas that have been now occupied by vegetation and the expansion of cultivated fields. Of course, the change detection exercise has be dealt with care as it is easy to misinterpret seasonal changes for long-term vegetation changes. This topic haas been discussed during the class and it was actually raised by few of the participants before I addressed it. It was stressed how it is important to compare image acquired in the same time of the year in order to overcome the issue of seasonal changes.

With further skills, in the future, participants will be able to quantify vegetation growth or loss in different parts of the region.

Fig 8. Screenshot of the Semi-Automatic Classification Plugin of QGIS. 


Fig 9. Example from the hinterland of Sfax of how to map vegetation growth. Landsat 8 false colour 5-4-3 band combination (top). Normalised Difference Vegetation Index (NDVI) calculated using Landsat 8 (bottom).

Following the vegetation mapping exercise, we focus our attention on urban development a major threat to the archaeology of the region of Sfax and in general in North Africa. Many studies now proved that unconsidered urban sprawl is the major cause of destruction of archaeological sites and landscapes in this part of the world.

Using the same datasets, Landsat 5 and 8, we used the Semi-Automatic Classification plugin of QGIS to run some image processing analyses to highlight the visibility of built-up areas (eg. urban) in the region of Sfax (Fig. 10).

In particular, we have calculated the Principal Component Analysis (PCA) on three bands of Landsat 8 and 5. We used bands 7-6-4 which is the best for highlighting built-up areas from the rest of land uses. We performed the PCA for both Landsat 5 (1987) and 8 (2019) so that it was possible to appreciate the difference in urban growth of the city of Sfax, for example.

Figure 11 shows the classification process and allows to appreciate the usefulness of such tools when used on a large scale.

Participants were very impressed with the procedure and deemed it very useful for a regular monitoring of large region, as well as the understanding of how and how much urban centre developed in the last 50 years. This will definitely support decision making and future policies with regards of development control. It is important and members on national heritage institutions are aware of these tools as it will facilitate the dialogue with local government in the hope of new legislation that protects Tunisian and Libyan cultural heritage.

Fig. 10. Landsat 8 (2019) displayed in false colour bands (5-4-3) of the area of Sfax city.

Fig. 11. Principal Components Analysis (PCA) of Landsat 5 (top) and 8 (bottom) images that cover the city of Sfax. It is clear the historical rate of urban expansion of the city over a 30 years time-span

Day 3

The participants used these techniques to process data by themselves in order to practice. The aim of the advance training is also for participants to be independent in performing exercises and procedures. We left half of the morning for independent practice and use the second half for Q&A session on topics covered in the whole course of the project.

There were quite a few questions and discussions showing the critical appreciation of new technologies from the participants.

Overall, it was a successful training session that concluded the Training in Action training programme in GIS and Remote Sensing.