Surveying Indonesia’s Myriad Islands

With nearly 13,500 named islands, Indonesia is very much an archipelagic country. The challenge is surveying and mapping many of them. For Dr Catur Aries Rokmana, Gadjah Mada University the ideal platform is a UAV.

In total, Indonesia has 13,466 named islands. For an archipelagic country it is very important to document them because many of the baseline points which define maritime boundaries are located in the outermost islands. This article gives some practical experience of utilising aerial photography, taken using Unmanned Aerial Vehicles (UAV), for photogrammetic surveying of the outermost small islands of Indonesia.

Logistical Difficulties

There are unfortunately some challenges in surveying the islands. Some have an area of less than 500 hectares and are formed of rock and cliff coral morphology. Some are located more than 100km from the nearest airfield or are near the ocean with no regular transportation facilities. Some islands are now uninhabited, so it is difficult to stay there and the largest vessels available for transportation are 12-man boats.

This means that the surveying method must be portable enough to carry on a boat and that data acquisition must be fast enough to complete one island in a day. The method also has to be:

• suitable for operation by local staff,
• produce results of appropriate geometric accuracy and
• be cost effective.

UAV Solution

One of the best innovative surveying techniques available makes use of Unmanned Aerial Vehicle (UAV)-based photogrammetry. Many researchers have been using this technique due to its low cost and effectivness for surveying areas of less than 1000 ha.

To complete the task for each island, the portable UAV-based mapping system was carried by small boat from the nearest populated island. The maximum distance was less than 40km, or a maximum of two hours travel time by boat.

Planning and Logistics

The whole system can be carried by a few people in backpacks or by hand. To make the system as portable as possible a battery-powered remote-controlled plane was used. The ground station was run from a laptop powered by a small generator. A point-and-shoot digital camera with GPS photo tag capability was chosen as the optical sensor, which was modified with an external-extended Li-Ion battery to provide power for up to eight hours operation. The airframe was designed for take-off and landing in a limited area and was light weight (less than 3kg).

Flying by autopilot parallel to the wind direction results in a straighter path than using a cross wind flight direction. The wind speed over small islands can be more than 40km/h making it difficult for light-weight aerial platforms to handle the straight line flight. The flights were planned with large overlap (85%) and sidelap (25%) to ensure that there was no missing stereo coverage and furthermore, that all parts of the islands were covered by at least four images.

The UAV was hand-launched to fly for 20 minutes at altitude following the flight plan and covered up to 300ha in each flight, returning home autonomously. The ground control was surveyed at the same time as the flying.

Data Processing

The images and in-flight orientation data was downloaded and the imagery processed by  digital photogrammetry to produce an orthophoto mosaic and a digital surface model (DSM) a few hours after the downloading. Both of these datasets were the basic products for later visual interpretation for thematic or geo-information extraction.

The problem with point-and-shoot cameras is that their lenses are not calibrated and, if fitted with a zoom function, focal length can easily be varied accidentally. The best way to improve the camera geometry is to run the photogrammetric bundle adjustment with a self-calibrating camera or ’in-flight’ calibration. This can improve the precision to sub-pixel after adjustment. The computer used to process the Indonesian islands project was a Pentium core i7 with 20Gb RAM. This can handle up to 1,000 photo frames to produce an orthophoto mosaic and DSM with a processing time of less than ten hours. There is some noise in the production of the DSM in homogenous areas such as water/sea and sandy beaches.

Another difficulty is the filtering of the DSM to produce a DTM (digital terrain model) which can be varied dependently upon the landscape characteristics. Whilst the DSM elevation was generally accurate to 3 to 5 times ground sampled distance (GSD), the derived DTM was accurate to around 8 times GSD.

The orthophoto image and filtered DSM information were used to produce vector and contour mapping. This technique can produce an accuracy of better than two times GSD for horizontal position and four times GSD for elevation of DSM information. Aerial photography with 15cm GSD is suitable for mapping at 1:2500 scale and is of high enough resolution to identify natural resources and to assess the environment. Near the beach, in clear water it is possible to see underwater objects, which present another potential application for UAV aerial photogrammetry.

This article was published in Geomatics World May/June 2015