The observation has a shape
A proximal sensor never observes a target in the abstract. It observes from somewhere, at a particular distance and angle, through a particular field of view, at a particular time. These relationships form what we call the sensor-target geometry. Think about holding a spectrometer over a tomato plant. You can control the angle at which the sensor points towards the plant. You can hold it directly above a leaf, point it at an angle across several leaves, or move closer to measure a smaller area. You might think that you are measuring “the tomato plant” in all three cases, but the sensor is actually seeing three different things.
From directly above, the measurement might mostly contain one leaf. From an oblique angle, it might include several leaves, stems, shadows, and perhaps some soil in the background. Move the sensor a little further away and the footprint becomes larger again. If the sun is behind you in one measurement and to the side in another, the illumination of the leaves can change as well. The tomato plant has not changed between these measurements. The observation has.
Geometry is often treated as a technical detail because it can be difficult to summarize in a methods section. Yet it can determine what enters the measurement, which surfaces dominate the signal, how illumination is distributed, how comparable repeated observations are, and whether a result can be transferred to another site or date.
Distance changes the target
Distance influences spatial detail, footprint, perspective, and the amount of material included in a measurement. A camera moved one metre backwards may suddenly include more of a crop row, more soil between plants, more sky, or a different balance of sunlit and shaded leaves. The image may still look very similar to a human observer, but the pixels going into an index have changed. The same applies to non-imaging sensors. A handheld spectrometer held close to a leaf may measure a relatively small and well-defined surface area. Hold the same sensor further away and its footprint may include several leaves, shadows, soil, and neighbouring vegetation. What started as a measurement of a leaf can quietly become a measurement of a small piece of canopy. This is particularly important when measurements are repeated over time. If the sensor is held at slightly different distances on different days, some of the apparent change in the signal may come from the changing footprint rather than from the target itself.
Distance also changes the importance of the measurement path. At short range, atmospheric effects may be small compared with other sources of uncertainty, although this depends on the wavelength, conditions, and required accuracy. As the path between target and sensor becomes longer, haze, humidity, temperature gradients, and other atmospheric effects can become increasingly relevant. At that point, we start entering the logic of remote sensing, where the atmosphere between the target and the sensor can have a substantial influence on the signal.
Angles, surfaces, and backgrounds
Viewing angle matters because most natural surfaces are not optically uniform. Leaves are tilted, curved, layered, and partly shaded. They are not usually flat surfaces pointing directly towards the sensor. A measurement from above can therefore be quite different from one taken from the side.
Water provides another good example. A calm water surface can reflect the sky or the sun depending on the viewing and illumination angles. Once the surface becomes rough or waves are present, the geometry becomes even more complicated. Soil roughness creates shadows and directional effects. Snow and many built materials can also behave differently depending on the direction from which they are observed.
The background matters too. An oblique vegetation camera may include sky, soil, tree trunks, buildings, or distant landscape. A small change in camera position can change the proportion of these different surfaces in the image. For an RGB index or spectral measurement, that change can affect the resulting signal even when the biological target itself has not changed. In controlled experiments, it is sometimes possible to remove some of these problems by controlling the background. For example, the non-relevant parts of a scene can be covered with a black, non-reflective cloth. This can work well in a greenhouse or laboratory experiment. Outdoors, however, it becomes much more difficult. Wind moves the cloth, vegetation grows around it, and keeping a clean and stable background in a field or forest can quickly become impractical. A good design therefore considers both the intended target and the unwanted parts of the scene.
The best viewpoint is not necessarily the one that produces the most attractive photograph. It is the position that produces a stable, representative, and interpretable observation.
Geometry through time
Repeat observations are most useful when their geometry remains stable. Fixed cameras, towers, and carefully marked sampling positions can reduce a large amount of avoidable variation. But fixed does not mean unchanged. Wind can move a camera mount or deform the canopy in front of it. A pole can settle into the soil. A curious cow, a hiker, or field machinery can move equipment. A tree can slowly grow into the field of view. A crop canopy can increase in height and change its structure. And, as I have mentioned before, spiders can decide that the space directly in front of the lens is an excellent location for a web.
Even replacing a sensor can introduce a change. A new camera may have a slightly different lens, focal length, field of view, or mounting position. To a person standing at the site, the difference may appear negligible. In a long time series, it can create a noticeable break.
Long-term systems should therefore be checked against reference images, site photographs, and documented orientation. A simple comparison of images before and after maintenance can sometimes reveal a small shift that would otherwise become an unexplained change in a multi-year dataset. This is particularly important for automated systems. Once a sensor has been running for months, we tend to trust the continuity of the time series. But continuity of data files does not necessarily mean continuity of observations.
Illumination is part of the geometry
For passive optical sensors, geometry also includes the position of the illumination source. For most outdoor measurements, that source is the sun. Its position changes throughout the day and across the year. Clouds alter the balance between direct and diffuse light. Shadows move across the scene. The same leaf can therefore be illuminated very differently at 08:00, 12:00, and 17:00. This becomes particularly obvious in a forest. A small change in sun angle can move a patch of sunlight across the forest floor or change which leaves are illuminated within a camera’s field of view. An RGB index may respond to that change even though the vegetation itself has not changed.
Artificial illumination introduces its own geometric considerations. Lamp position, intensity, spectrum, distance, and stability all become part of the measurement design. This is why repeat RGB photography often uses a consistent daily time window or daily summary approach. It is also why spectral and thermal measurements need clear acquisition protocols.
The observation is not simply the sensor and the target. It includes the conditions under which energy reaches the target and the way that energy travels back to the detector.
Designing for comparability
A field protocol should record the basic geometry of the observation. Depending on the sensor, this may include sensor height, distance to the target, viewing direction, orientation, field of view, target footprint, camera lens, focal length, sampling location, time window, and known obstructions.
For imaging systems, it is useful to retain a reference photograph with the ROI and important landmarks marked. A tree, fence post, building, or other stable feature can provide a useful reference when checking whether a camera has moved. For mobile systems, record the trajectory, speed, sensor orientation, and positioning information where relevant.
It is also worth looking for metadata standards that are already commonly used in your field or supported by your devices. Following an established metadata framework can make it much easier to compare your observations with those from another research group or combine datasets later. Comparability does not require identical conditions in every campaign. That is often impossible in field research. What matters is knowing which conditions changed and assessing whether those changes matter for the intended interpretation.
Geometry is a scientific variable
Sensor-target geometry deserves more attention because it often explains why seemingly similar measurements disagree. Two sensors can point at the same crop and produce different signals because they see different parts of the crop. The same sensor can produce different measurements on two days because its position has changed by a few degrees. A camera can remain in exactly the same position while the growing canopy changes the geometry of the scene around it.
These effects occur before the researcher has applied any algorithm, model, or index. Treating geometry as a scientific variable leads to better deployment design, stronger metadata, clearer quality control, and more credible comparisons across time and space.
One of the advantages of proximal sensing is that we are often close enough to the target to control much of this geometry. We can choose where to place the sensor, how far away it should be, which direction it should face, and when it should collect data. Good proximal sensing makes use of that opportunity. The closer we are to the target, the more control we have over the observation. We should use that control deliberately.
Explore further:
- Schaepman-Strub, G., Schaepman, M. E., Painter, T. H., Dangel, S., and Martonchik, J. V. (2006). Reflectance quantities in optical remote sensing: Definitions and case studies. Remote Sensing of Environment, 103(1), 27–42. https://doi.org/10.1016/j.rse.2006.03.002
- Sonnentag, O., Hufkens, K., Teshera-Sterne, C., Young, A. M., Friedl, M., Braswell, B. H., Milliman, T., O’Keefe, J., and Richardson, A. D. (2012). Digital repeat photography for phenological research in forest ecosystems. Agricultural and Forest Meteorology, 152, 159–177. https://doi.org/10.1016/j.agrformet.2011.09.009
- Rançon, F., et al. (2023). Designing a proximal sensing camera acquisition system for geometric characterisation of vines. Sensors, 23, 886. https://doi.org/10.3390/s23020847
Leave a Reply