High resolution 3D mapping in flood-exposed communities using Unmanned Aerial Vehicles (UAVs)

Authors: Wahaj Habib, Giuseppe Molinario (WB), Sasanka Madawalagama (AIT)
Resource Persons and Technical lead: Sasanka Madawalagama (AIT), Chatumal Madhuranga (AIT)

Project Overview
One of the first work groups to get started in the first week of the un-conference was the drone or unmanned aerial vehicle (UAV) working group. This was not a coincidence, as the flow for the un-conference was planned to follow the data: from data acquisition, to mapping and analysis and finally risk data communication.This working group was led by a team from the
Geoinformatics Center of the Asian Institute of Technology (GIC-AIT) and the World Bank. Work started with an introductory training to the elements of drone mapping, then after selection of the study area, the group went to the field to practice setting up ground control points (GCPs), basic maneuvers of drone flight and automatic flight plan execution. Subsequently, nine flight missions were flown in the Sanpatong area situated about 45 minutes drive south of Chiang Mai city. The second mission was conducted in the Mae Chan district of Chiang Rai province in the third week. The results were 3cm resolution orthophotos, digital surface models (DSM) and digital elevation models (DTM) of flood-affected areas. These datasets were processed and uploaded to GeoNode and OpenAerialMap to be shared freely. Moreover, this exercise illustrated the use of drone technology to provide a cheap and effective way to map out areas for disaster vulnerability mapping, and to support relief efforts in post disaster phases.

Figure 2. Hands-on session on processing drone images to obtain map products

3) Flight planning

Designing and executing a good image acquisition plan is the most important part of any drone mapping project as the ultimate success of any photogrammetric work depends upon the quality of the images. A flight plan generally consists of two items: a flight map, which shows where the photographs are to be taken, and specifications, which outlines how to take them.
The image acquisition area in Sanpatong was 2 km2 and the one in Mae Chan was 1 km2 . DJI Phantom 4 and DJI Mavic pro drones were used to capture the high resolution aerial images for mapping. As the drones have limited flight time of 30 mins, the study areas were divided into subsections and flight paths were designed accordingly.

Drone Deploy, an app that has cross-device support as well as a web-interface and the DJI GS Pro app were used for flight path generation and execution. The images are acquired along a flight path, much like a scanner, where the software finds the most efficient path up and down a specific area of interest (AOI), given certain parameters, such as the overlap between images that is wanted, the flight altitude, etc. The overlap was set to 75 percent for both sideways and forward direction, with a flight altitude of 100m.

Figure 4. DJI Mavic pro and Dji phantom 4 getting set up for flight

4) Data collection/field work

After flight planning, the next phase was data collection (i.e. having the drone collect the images). As mentioned before, both in Sanpatong and Mae Chan there were several separate missions that were planned and flown to collect the images. With the help of Google maps, suitable takeoff and landing sites were located for each mission. The drone deploy and DJI GS Pro apps are built with the DJI application programming interface (API) which means that the app is able to connect directly to the drones, and provide an alternative flight control to their native control software (DJI GO). Once the drone is connected to the app, the flight plan is loaded to the app. After the pre flight check, the mission is executed autonomously with the push of a single button on the connected android or iOS smartphone or tablet. The drone takes off and flies along the flight path to collect images according to the given overlap and then automatically come to its home location and land after mission is completed. There are also a few failsafe security measures built into the drone software: if the battery falls below a certain threshold set by the user (usually 10%) or if the controller loses radio connection with the drone, then the drone flies back by itself and lands given the GPS coordinates of a “go home” point that it stores on-board at the start of each flight. Furthermore, radar obstacle-avoidance sensors allow the drone to avoid any obstacles found on its “way home”. The images are then downloaded from the drone and subsequently processed to map products.

5) Ground Control Establishment – Sanpatong

Figure 7: Ground control network established in Sanpatong

The requirement for the Sanpatong study area was to generate high resolution map products with high geometric accuracy. It is essential to establish a good ground control of the mapping area to generate accurate map products. Ground control points (GCP) are points of known coordinates in the area of interest which are clearly visible in the acquired images. A well distributed ground control network was established and surveyed with a high end Real Time Kinematic (RTK) Global Navigation Satellite System (GNSS) unit. GNSS is the correct terminology for what is frequently just called “GPS” but in fact, the Global Positioning System (GPS) constellation of satellites which allows navigation on the ground, is just one, of many, GNSS’s. Others are for example GALILEO or GLONASS (from the European Union and Russia, respectively). Most of our smartphones are not equipped with GPS, but GNSS, as they are capable of using many different constellations to track the users spatial position. Thailand’s CORS (Continuously Operating Reference Station) network was used to obtain RTK corrections through the internet. A total of 9 ground control points were established to an accuracy of 2 cm. Ground control establishment and surveying was done by GIC-AIT, and elevated considerably the rigor of the drone mapping activities in the un-conference. In Mae Chan, the missions were flown solely with the onboard GPS unit providing the location of each image.

6) Post-processing

Figure 9. Individual drone images

Even with high processing speed computers it can take some time to process all the images that are collected, as they are usually thousands if not tens of thousands. Essentially, what the software does is stitching the images together after removing their inherited errors to produce a map. In addition to the generated 2D orthomap, it is possible to map the elevation by adopting photogrammetric methods. The processing was done by the GIC staff using Pix4D mapper software at GIC-AIT in Bangkok.

Three different products were obtained from the drone surveys: Orthophotos: which are geometrically corrected aerial photographs such that the scale of the photograph is uniform and can be utilized as a map. Digital surface models (DSM) and digital terrain models (DTM) represent elevation data in the images. Together both DTM and DSM are sometimes referred to as digital elevation models (DEMs). A DSM shows the surface elevation at the “top” of the surface, like tree canopies and house roofs, for example, whereas a DTM shows the elevation at the terrain base, or in other words the elevation of the ground.

7) Sharing

Many of the products were shared on the GeoNode web GIS application for the UR field lab and also on the OpenAerialMap platform

Using the maps developed:
The products developed are used in the context of flooding to map in higher spatial and temporal resolution the vulnerability of given communities to flooding. Drones can be deployed quickly, and can image under the cloud cover, where satellites cannot. Drones provide very high resolution imagery, where minute details can be picked out of the imagery. Such high resolution images from satellites cost a lot of money, and even if money were no object, it is not guaranteed that the area of interest (AOI) has been imaged recently in good quality (for example, with low cloud cover). Additionally, drone imagery can be deployed even more quickly, forgoing post-processing steps, if imagery is needed promptly in rescue and post-disaster phases. Our work highlighted the need for clear regulations and communications that identify who the operators are and what they are capable of in terms of acquiring drone imagery and maps in pre-disaster and post-disaster phases. Currently there are conflicting resources on drone flight regulations (licensing and registration) and no-fly zones, restricting the use of this leap-frog technology.

Acknowledgements

Thanks to the GIC-AIT team, the authors of this blog post as well as Dr. Kavinda Gunasekara (Senior Program Specialist- Geoinformatics Center, AIT) and Dr. Manzul Kumar Hazarika (Director- Geoinformatics Center. AIT). Special thanks to Dr. Phonpat Hemwan and his tea from Faculty Of Social Sciences, ChiangMai University for providing historical flood data and for the support during the field work.