Notes on low cost drone surveying for community archaeology projects

Goals

• As cheap as possible
• As little equipment as possible
• Low technical knowledge (no compiling, no Arduino, no Rasberry Pi, no devboards)
• Works with existing free (as in price) off-the-shelf software (SW-Maps, QField, Emlid Studio)
• Practical in the field
• Produces results that meet academic standards (EAC & ADS)
• Produces results with accuracy that surpasses 'normal' GNSS (less than 0.5m?), if not much less

Prerequisite knowledge

• How to fly a drone legally, and more importantly, safely. Even if you don't need to, get an Operator ID, and flyer ID.
• How to read Ordnance Survey grid references. If you are comfortable using an Explorer or Landranger series map you should be just fine.
• What an Ordnance Survey trig point is and how they were used.
• Understanding of maths at an advanced school level. You don't need to do any calculations but understanding some of the concepts and operations involved in different coordinate systems will be helpful.
• Understanding of physics at an advanced school level. Some understanding of how gravity and radio works will give you a better understanding of how GNSS systems such as GPS work.
• Geography at a basic school level. Plate tectonics and the shape of the earth.

Equipment needed

• A camera drone
• An android mobile phone
• An RTK capable GNSS receiver
• An GNSS antenna
• A tripod or some other device to hold the antenna steady
• Ground control point markers
• A reasonably powerful computer

Software needed needed

• A GNSS data capture app such as QField or SW Maps
• QGIS
• Open Drone Map

Factors affecting accuracy

The antenna

Different GNSS antennae have different costs and properties. Some are designed to provide greater rejection of multi-path signals (where the signal has bounced off an object), some have built in amplifiers to increase signal levels. The more challenging the environment the better your antenna needs to be. If you are outside an urban environment with good horizons and few trees you should be able to made do with something cheap. Happily these are also the sorts of environments best suited to flying a drone.

The antenna mount

Your GNSS receiver is actually measuring the position of the Antenna Phase Centre (APC) of your antenna. This point shifts depending on where the satellite is and what frequency it broadcasts on but it's generally somewhere along the axis of the antenna.

You can hand hold the antenna. This will still be more accurate than a normal GNSS receiver but it's not ideal, particularly if you want to take repeat measurements. Instead, find a way of mounting the antenna on a support such as a tripod. The mount needs to allow the antenna to be positioned directly over the point being measured, and the antenna needs to be pointing at the centre of the earth. Use a spirit level for extra points.

Distance from the base station

Regardless of whether you are using RTK/PPK, NTRIP, or are setting up your own base station, your correction data will be less accurate the further away from the base station your RTK rover is.

Check the specifications of your receiver. It will usually be quoted as some fixed amount plus a ppm error. 1ppm is 1mm of additional error every 1km away from the base station. So if your receiver says 2cm + 1pmm and you are 5km away, your accuracy will be 2.5cm.

Your correction data

It can be difficult to get details on the properties of an NTRIP feed, particularly if they are free. Many commercial users of RTK are not interested in absolute accuracy. In the UK you will want to know how the data relates to OS Net and the OSGB36 TRF. One big issue is that RTCM messages do not indicate what CRS is in use. An unexpected mismatch between the base station and rover will result in the rover making measurements in the CRS of the base station without your knowledge (for example, OS Net is in ETRS89 and will transform measurements to ETRS89).

Random notes

RTK/PPK can produce very accurate results but you need to establish a baseline. You can use a base station to establish relative accuracy but this doubles the equipment needed, and you still need to fix the base station's location to provide absolute accuracy. You can use an NTRIP feed but these are expensive. There are free feeds but it's difficult to judge the provenance of the data (this isn't a criticism, they provide a great service). I'm also not sure if referencing against a free baseline will be reproducible in 5/10/100 years time. Even some of the paid feeds don't make this clear as a lot of the commercial use cases only require relative accuracy. Your nearest feed may also establish a long baseline and this decreases accuracy. You really want this to be less than 10 km away. My nearest free feed is 40km away. NTRIP requires a network connection. You can use PPK and free data from OS Net to get accuracy through post processing, but this requires a receiver that can provide you with raw data. On the other hand PPK does at least appear to be repeatable and reproducible as you can archive the correction data. Downside: doesn't give quick feedback, more complex.

Cheap gnss receivers are cheap. The cheapest are just development boards with no housing, no power, no antenna. Adding these things increases the cost. The receiver needs to be at least a little robust.

More expensive drones will of course produce better results. They will have better cameras, and some drones evan have rtk capabilities that include NTRIP and PPK. More featurful drones are not just more expensive but they are heavier and this places them into at least the C1/UK1 class, mandating that you must have an operator ID and a flyer ID and a remote ID system. C2 and the regulations further tighten. DJI Mini 2 is about the price point I'm aiming for. You can pick up a used one affordably, it sits in the least restricted class of drones. Downside is that the image quality can be poor, and I don't think it supports autonomous flight plans. Even with the poor camera quality you are likely creating better quality images.

Drones and GNSS equipment typically record position data according to WGS84. The North Atlantic and Erasian plates are spreading at an annual rate comparable to the accuracy of RTK. WGS84 coordinates will slowly drift across a fixed location in the UK, a four-dimensional reference is needed to fix a point in relation to the ground Anyone wanting to accurately use WGS84 has to provide the time (epoch) of when the data is captured. ETRS89 is an alternative frame of reference that fixes the Eurasian place at it's location in WGS84 epoch 1989. The systems already differ by about .92 meters. ETRS89 is recommend for mapping in Europe for this reason. The OS has a system for realising ETRS89 called OS Net. The OS has it's own system called OSGB36 which is the basis for the grid system we are familiar with. OSG36 was redefined in 2016 and is referenced against OS Net and consequently ETRS89. It is recommended to use ETRS89 for archaeology in Europe. Some use OSGB36.

GCPs need to be flat to the ground. If you use something like disc cones it could be difficult to precisely locate them in the images due to perspective. You could print small GCPs on paper but paper is fragile, gets damp, gets blown around and size might be limited. You can use existing points on the ground if there are some convinient ones. You can use spray markers or tape but these will be quite "wide". Floor tiles Bucket lids Corrugated plastic. Marker needs to big enough to be resolved with a centre, so at least 4x4 pixels to mark the centre, and the pixels need to represent an area comparable to the GNSS accuracy. When are 4x4 pixels what size? Depends on the properties of the drone camera. For a DJI Mini 2 - 82m -> 3cm, 41m -> 1.5cm, 20m -> 0.75cm. In practice it's going to be worse, the camera is not stable, the sensor won't be square with the marker, rolling shutter effects, lens imperfections. 20m sounds like a good height and is not far off what I've been doing so far. To place the centre accurately, in practice the target really needs to be closer to something 10pixels, so 30cm markers should be fine at 20m. Papers suggest that computer vision performance at identifying the centre of markers surpasses that of humans.

Experiment #1 Can I at least map out local, legacy, trig sites to even 1m of accuracy using the free NTRIP feed? Can I even find them?