During the final year of my Bachelor’s degree, I decided to pursue an ERASMUS+ placement, aiming to gain work experience in the field of Earth Observation (EO) and Geographic Information Systems (GIS) and also experience living abroad. I started reaching out to companies and Institutions mainly in Central Europe. After several interviews with different companies, I decided to pursue an internship at Spatial Services GmbH, located in Salzburg, Austria. The reason I chose the company was both for the projects I would be working on which focused on supporting humanitarian aid and the country it is located in. Since the beginning of my studies, it has been a goal of mine to do something meaningful with my work and have a positive impact, so when I learned that the project I would be working on was a collaboration with Doctors without Borders (Médecins sans frontières-MSF) I was excited to start. I worked at Spatial Services as a full-time ERASMUS+ intern from March 2022 until June 2022.

​

Project description

The collaboration between MSF and Spatial Services consisted of requests of different products and maps for areas mainly located in Central and East Africa and the Middle East. The request usually came from one of the MSF’s Operational Centers (OCs) requesting a specific product for a dedicated Area of Interest. The two main groups of orders were “Population” and “Environmental”. The “population” products included dwelling extractions (DWX) in refugee camps or towns, which were performed using very high resolution (VHR) satellite imagery. Derivative products of the dwelling extraction were the “Dwelling Density” (DWD) and “Dwelling Change” (DWC) estimation as well as the corresponding maps. The “environmental” orders usually included flooded area extractions, water body extractions and land cover classifications. These tasks were mostly performed using open-source high resolution satellite imagery, e.g. from the Sentinel-2 or Landsat satellites, or using data from commercial satellites like SPOT-6.

Depending on the request, if VHR data were necessary, a satellite would be tasked, or an archive satellite image was requested. After the request, the satellite image was received a few hours to a few days later.

 

Brief product description

 

• DWX: extraction of dwellings (tents, buildings made of natural material, tukuls) from VHR satellite imagery in the form of vector polygons using a combination of Deep Learning techniques with manual refinement.

• DWD: a hexagonal grid that aggregates the intersecting buildings with each hexagon. The final output was a number indicating the number of building “inside” each hexagon that was later translated into area density.

• DWC: a hexagonal grid that includes the result of a comparison of two time steps of DWX. The final output was a number indicating the increase or decrease of buildings inside each hexagon.

• Flood masks: raster or vector layers with the detected flooded area

• Water body masks: raster or vector layers with the detected

• Maps: created for all the products mentioned above, depending on the request.

 

My responsibilities and contributions

I mainly worked on the “population” orders, specifically manual refinement of the extracted buildings using Deep Learning, obtaining the DWD and DWC products and creating maps for all products. The products were created for different areas of interest, both refugee camps and towns or cities, meaning that the workflow needed to be adjusted for different building structures and environments.

 

Product retrieval methodology

 

• The Dwelling Extraction (DWX) product was mainly obtained by applying a Deep Learning (DL) model in ArcGIS Pro to a VHR image to extract the buildings. First, samples had to be created to feed the model. The number of required samples differed depending on the quality of the image, the complexity of the area and the number of different dwellings in the area. Then the model was trained using the extracted training samples. As a final step, the model was applied to the image and the final output was polygons corresponding to each building recognized by the model.  The DL output was then manually refined to deliver the final DWX product. Depending on the area complexity and since DL does not perform equally well in all areas, in some cases the DWX had to be performed manually.

• The Dwelling Density (DWD) product was usually created in ArcGIS Pro using the DWX product as input. First, a standardized hexagonal grid was created, where each hexagon had an area of 1ha. The two layers were then joined together, and a count attribute was created that represented the number of buildings that fell inside each hexagon. The number of buildings required to characterize an area as very dense or sparser was not standard and differed depending on the area characteristics.

• The Dwelling Change (DWC) product was also based on the DWX product, but in this case from two different time steps (T1, T2). The methodology followed to obtain the product is similar to the one for DWD. After a hexagonal grid of 1ha per hexagon was created, the layers were joined and the dwelling count from T2 was subtracted from T1 and the number of increased or decreased buildings was added in a new column. The change was visualized, with dark blue colors representing high decrease of buildings in the rea and dark red high increase of buildings.

• Flood mapping was performed mainly using the Google Earth Engine (GEE) software since it allows processing of large datasets on the cloud. The main data utilized for flood mapping were Sentinel-2 MSI, Sentinel-1 SAR and Landsat-8 OLI. The imagery from each sensor was used depending on the availability of the imagery. In general Sentinel-2 MSI data were preferred, because of their high spatial and temporal resolution. When high cloud coverage was observed over a scene, Sentinel-1 SAR data were used. For the optical data, the methodology used was based on water index calculation and thresholding. For the SAR data, the OTSU thresholding method was used to differentiate between the flooded and not flooded areas. Data postprocessing was also performed to exclude overestimated flooded areas (false positives). 

• A similar methodology was used for water body extraction, but additional optical VHR data were used depending on the size of the ponds requested to be extracted.

​

Workshops

During my internship, a few workshops were organized regarding Deep Learning in ArcGIS Pro for building detection and Flood Mapping in Google Earth Engine (GEE). The Deep Learning workshop focused on creating and applying a deep learning model in the ArcGIS Pro software to perform building detection. Some basic DL principles were explained first and then I got to familiarize with the different tools required to be applied step by step to perform the analysis. A detailed description of the tools and the parameters that need to be adapted was given and, in the end, I was able to create my own building detection model.

​

 

The second workshop that was provided focused on flood and water body mapping using optical and SAR data. The environment of the GEE software was first introduced and some simple scripts were created, for example, for accessing an image collection, calculating a vegetation index and plotting results. After that, different scripts were explained and shared with us regarding accessing high-resolution optical data in GEE, calculating different water indices, defining thresholds and exporting a water mask. Similarly, for the SAR data, two methodologies were presented, one based on OTSU thresholding and the second based on change detection. Finally, some post-processing steps were included to derive the end product.

 

Conclusions and recommendations

My experience of working as an ERASMUS+ intern was very positive. I experienced what is like working in a private company, delivering products, managing time, communicating with clients and partners, managing deadlines and most importantly learning new techniques in GIS and Earth Observation (EO). In addition, I had the opportunity to work in a very supportive and encouraging team that helped me broaden my GIS and EO knowledge and also gave me opportunities to evolve. For example, during the final month of my internship, I got to introduce the new interns to our projects and the workflows we were implementing and supervise them when working on different orders.

During this time, I also experienced working with an international humanitarian organization, how it operates, how to communicate with them, what their requirements are, delivering products to people with different expertise, where they are working and what services they offer to people in need. I believe this experience will be valuable for my future career.

During my internship, I also had the opportunity to experience a new city, with a different culture than I was used to, live on my own for the first time, travel within Austria, and witness beautiful landscapes and of course make new friends. All of the above made the internship an unforgettable time.

I would like to thank my team, and particularly Ahmad Alobaidi and Jonas Schultze-Naumburg, for their constant support during my internship and the technical skills they helped me develop, as well as Hubert Schöndorfer (CEO) for giving me the opportunity to perform my internship at SpaSe.

I would recommend Spatial Services to students who are looking for companies for their internship, as it provides a variety of projects the interns could work on based on their interests, as well as for the supportive and positive work environment.

​

Background image credits: Planet Labs, PBC

​