I was recently asked, “Where is imagery in digital farming?” Imagery, especially from satellites and increasingly from drones, is often proposed or inferred as the core functionality of digital farming. Why else would almost every digital farming discussion center on it and every platform lead with it as its centerpiece functionality?
Digital farming is about two key objectives. The first objective is getting inside the decision loops of the cultivation and making decisions before they become a risk. The second objective is enabling digital farming, the digital twin, to use its data to anticipate, assess, and explore alternative courses of action.
The user must first understand the agricultural decision cycle, and then the user must separate the variable data from the invariable data. The agricultural process will contain continuous decision loops throughout the cultivation, and the invariable data will provide the foundation for decision analysis. On the other hand, the variable data will enable the development of decision information.
The agricultural decision process looks like a decision loop developed in the 1960s for the military. It was called the OODA loop. From Wikipedia: “The OODA loop is the decision cycle of observe, orient, decide, and act, developed by military strategist and USAF Colonel John Boyd.“
For agriculture, digital farming might employ a modified OODA Loop for its decision loop. The modified OODA Loop is: Indication, Detection, Identification, Localization, Mitigation, and Validation. The Observe step is Indication. The Orient steps are Detection, Identification and Localization. The Decide step is deciding upon mitigation based upon analysis from the digital twin. The Act Step is Mitigation and Validation.
Imaging is used at the Detection, Identification, Localization, and Validation steps. Imaging exists inside of the decision loops for each of these steps and only exists to develop the decision information required for each loop. Each imaging technology has its optimum OODA use case.
Imaging provides both variable and invariable data. The best known of the imaging technologies measures the amount of the incident energy that is reflected by vegetation, soil, and other artifacts in the cultivation. These imaging sensors have varying spatial, spectral, radiometric, and temporal resolutions, and each has an optimum utility in the decision process. Other sensors that that might be used include Light Detection and Ranging (LIDAR), Synthetic Aperture Radar (SAR), and earth penetrating radars. These sensors transmit a pulse of incident energy and then measure the reflected energy, including the time and intensity of the return. This data tends to be invariable.
File sizes for imaging environments are massive and are often measured in multiple gigabytes. Generating, collecting, managing, and analyzing this data is expensive. The size of the files demands that they are optimally used and managed before sensing starts. It also demands application automation and integration. Automation and integration can only be accomplished when Imaging is aligned with its specific decision loop.
To date much of the imaging has generated information about the cultivation that does not support a decision, is not automated, and has already happened. Imaging today often has limited contribution to the overall decision process. This might be because it was the wrong sensor, the wrong resolution, or analyzed too late to do any good. Imaging’s contribution to the decision process is where it establishes its value.
When does imaging technology get used, how often, and what is its objective? The following is a sample of where imaging is in a digital farming sequence. There are potentially many different decision cycles that could be synthesized; each decision cycle will be crop and objective dependent.
As a functional starting point there are some key takeaways that were stated in the previous article about a digital farming system. Many of these factors represent invariant data. First, soil is the productive asset. There needs to be a digital soil model in this ecosystem, i.e., a digital twin that models the soil and its productive capability. Soil is the common denominator; one cultivation to the next. Next, seed production generates a wealth of information about that crop and the metrics for that production are available to the next cultivation. This includes crop and variety information like time to first emergence, timing of vegetative states, root growth profiles, nutrient and moisture utilization, and fruit size and maturation. Finally, the stresses that impact the cultivation are persistent. The geo-location and persistence of these stresses applies to weeds, fungi, insects, and biologicals. Early indications of these stresses are locked in the memory of the soil, and the digital twin will remember exactly where.
These data and their metrics can be statistically described, and these results can be used to digitally predict, specify, and schedule when imaging events in the next cultivation should be happening. This “memory” can be used to develop the indication, and like seed production their occurrence can be predicted and used to develop the various OODA loops.
Some examples of imagery used in digital farming include:
- Using LIDAR to establish and determine the topography of the field. Topography is the height of each soil asset relative to the soil assets around it. LIDAR systems on airborne systems can develop topographic or elevation maps of a field in a very short time, and LIDAR can provide resolutions in centimeters/ inches. This map with Electrical Conductivity maps will assist in determining where to take soil samples. Its value will be reinforced over and over as it gives insight to the soil horizons and soil contents at depth.
- The next area might be crop emergence. We know what the planting plan was and when it was implemented, and we can geo locate every seed. The computer can anticipate emergence, and with imagery it can count each seed with unitary precision. With emergence the digital twin can predict time to vegetative states, crop moisture and nutrient usage and requirements, potential fruit sizes, and production yields. It can execute this process for every seed if required. Each of these predicted steps gives rise to a new OODA loop.
- The digital twin will provide indications from memory or from environmental inputs that indicate the presence of these stresses, and properly used imaging will note the possible first detection of these stresses. The digital twin will predict when to start the next OODA Loop. Each stress will have its own signature, its sensible profile, and its temporal offset. This data in a digital twin defines the imaging technology and when and where to look beyond indication to achieve detection, identification, and localization.
Indeed, imagery is a part of digital farming, and, potentially, it is useful at many steps in the decision process. The image requires a level of precision that enables the required decision. It should be appropriate to the action required and no more, but its value is as a complement to the decision process to provide the valuable inputs that support detection, identification, localization and validation. It should seek to inform the process before an event so that it is inside the decision loop and there is time to take the appropriate action.