The Evolution of GCTs
A compromise between science, cost, and logistics at Precision Agriculture trials sites.
James Caudery
Spatial/Data Analyst at GHD
James is a Spatial Analyst at GHD in Melbourne.
James worked for 2Excel Geo as a Geospatial Analyst from 2015-2018. James was a functional lead for Field Operations and was responsible for the planning and execution of multiple successful field trials.
The exploitation of hyperspectral imagery is significantly enhanced if the apparent surface reflectance can be accurately recovered from the measured radiance at the sensor. This means correcting for ambient illumination and the effects of the atmosphere. There are a number of techniques to achieve this automatically, however an alternative is to deploy materials with known reflectance in the area to be imaged, called Ground Calibration Targets (GCTs). This allows the calculation of the relationship between surface reflectance and at-sensor radiance, which is then applied to the remainder of the imagery. This technique is known as the Empirical Line Method (ELM). Although this method requires the manufacture and deployment of GCTs, it provides the most accurate correction and so is the preferred approach when conducting early research campaigns.
The most important characteristics of GCTs are their spectral reflectance and physical size. A minimum of two GCTs are needed with a spectral reflectance that straddles the range of the spectral reflectance of the image, and has a spectral reflectance which is sufficiently different over the entire spectral range of interest. A minimum of one bright and one dark target are needed to apply a two-point correction, however further intermediate shades would improve the correction. Each target should be sufficiently large to present a ‘pure pixel’ to the imaging sensor. This is dependent upon a number of sensor characteristics (which we can model) and collection height, but ideally it should be many times the ground sampling distance. Physical size however is limited by practicality, and so we have decided to deploy GCTs with a minimum size of 3mx3m. Each side of the GCT is equivalent to ~10 times the Ground Sampling Distance (GSD) of our hyperspectral sensors, operating at 1000m above ground level.
The first iteration of GCTs were constructed in 2015 and deployed during that growing season. These consisted of two 8x4m tarpaulins, each of which had a pair of 4x4m targets. The result was four targets with a reflectance in the visible range of approximately 5%, 10%, 40% and 80%. This solution required a large, flat area and was difficult to keep clean whilst deployed. This restricted deployment to only those days when airborne collection occurred, creating a constant deployment and recovery burden.
Figure 1a
First iteration of a Ground Control Target using a costly four point correction.
Two white GCTs.
Figure 1b
Two dark GCTs
Figure 1a
First iteration of a Ground Control Target using a costly four point correction.
Two white GCTs.
Figure 1b
Two dark GCTs.
At the close of the season, the disadvantages of the current solution were addressed. We desired a large, flat structure with inexpensive material coatings or coverings that had suitable reflective properties and were resilient to the environment.
Various materials were trialled and measured until two candidates were found that met the requirements. The materials selected were white Tyvek, a roofing membrane available at most building supplies stores, and black weed-control material, available from most garden centres. The weed control fabric had a further advantage in its porosity, which enabled it to self-drain water from its surface.
Figure 2a
Tyvek reflectance.
Figure 2b
Weed control reflectance.
Figure 2a
Tyvek reflectance
Figure 2b
Weed control reflectance.
Prior to the 2016 growing season, new GCTs were constructed from 4×4 arrays of wooden pallets, giving a single target size of 4.8×4.0m. The result was effective but not ideal. The pallets were found to be susceptible to rain splash, resulting in surface contamination. Storage and transport were challenging, due to the bulky size and weight of the pallets. However, overall these targets were fit for purpose.
At the close of the season, a further review was conducted in anticipation of increased numbers of trial sites the following season. This was an opportunity to develop new, innovative GCTs incorporating the lessons identified in previous seasons. The result of this process was a design based upon a metal frame, upon which the calibration material would be affixed. A prototype of the design was manufactured and small changes fed back prior to the production of a large number of frames.
Figure 3a
The bespoke frames are now available as a solution for semi-permanent GCT deployment.
Bright GCT.
Figure 3b
Dark GCT.
Figure 3a
The bespoke frames are now available as a solution for semi-permanent GCT deployment.
Bright GCT.
Figure 3b
Dark GCT
Each frame is 1.5x3m with a central hinge allowing it to fold in half and be stored and transported in a 1.5×1.5m configuration. The frames are constructed from coated aluminium (about one third of the weight of steel) which is resilient to the environment. Each frame has removable legs and stands 0.3m above ground level, and can be pegged into the ground for added stability. The centre of each frame is spanned by aluminium mesh, which keeps the material surface flat and allows natural drainage. Material is placed on top of the frames and clamped into place to prevent movement. We utilised frames in pairs to construct GCTs each 3x3m in size, although other configurations are possible.
These new GCTs have proven to be very effective. They remained clean over the course of the season and significantly reduced the maintenance burden. Monitoring of the materials exposed to the environment showed that the Tyvek spectral reflectance drifted marginally over time. This was accounted for in our application of the ELM correction.
Overall, the frames performed outstandingly. We strongly recommend that clients consider use of our GCTs for projects at field scales.
