6.3 Common Data Models (CDMs)

This step applies to a large project with a variety of different stakeholders. If there are, for instance, several organizations taking part in the study, you will need to list them and agree exactly what each will contribute.

Once you know this, you can discuss to determine if all parties already use a CDM or ontology, or whether there is one they can all adopt. If so, then the next step will be much simpler.

A data model is a behind-the-scenes organizational template of how the data entering the project is to be managed. The terms ‘common data model’, ‘ontology’ and ‘terminology’ are often used interchangeably during the early stages of project planning. We are talking here about a CDM that can be based on an ontology or terminology, and may indeed be a subset of an ontology. Though it is not the same thing, it can be thought of as a template for all the data elements that are going to be used in the project.

Without a CDM, different datasets will sit in different folders on the computer or the storage cloud, with no way of comparing or analyzing them. The project manager and team will decide in advance the unique ordering of the data to allow for an effective analysis.

In some cases, pre-existing agronomy data models can be used (see factsheet for a list).
The data model should naturally suggest itself according to the type of analysis required:

• A quantitative relation—e.g., the effect of fertilizer on output.
• A temporal basis of tracking trends—e.g., the climate variation over the long term.
• A geographical basis of a topological relation—e.g., the managing of crops in a complex environment.
• A semantic basis of identifying patterns—e.g., finding new patterns of crop-climate dynamics.

The data model is a crucial understanding before embarking on research, and must be suitable for the research to be undertaken. Here is an example of a geographical data model that is typical of Geographic Information System (GIS) map approaches:

To realize a good data model, one must focus on the question that one is bringing to the data.

Once the CDM is outlined, the next step is to see if data elements can be linked to relevant data standards. In this case, one would look at:

• Datasets can be layered based on different levels of abstraction and detail. Here is an example:

   

  Level 1: Raw sensor data

  This layer consists of the unprocessed data points directly captured by the sensors, including timestamps, sensor IDs, and raw sensor values (e.g., temperature readings in volts).

   

  Level 2: Processed sensor data

  This layer involves processing the raw data from level 1 to convert it into meaningful units (e.g., temperature in degrees Celsius) and potentially applying calibration or filtering techniques.

   

  Level 3: Derived features

  This layer utilizes the processed data from level 2 to calculate additional features or metrics relevant to agriculture. Examples include:

   

  • Vegetation indices: Derived from combinations of different sensor readings to assess plant health and growth.
  • Soil moisture index: Calculated using soil moisture sensor data to indicate irrigation needs.

  Level 4: Crop and field management data

  This layer encompasses data related to crop management practices, including planting dates, fertilizer application records, pest control activities, and yield data.

   

  Commonalities

  Identifying commonalities across different datasets can be crucial for integrating data and creating insights. Here are some examples:

   

  • Spatial data: Sensor data and field information both have a spatial component, often represented by geospatial coordinates. This allows for spatial analysis and visualization of data across different layers—for example, correlating sensor readings with specific locations within a field.
  • Time-based data: All data points will likely have timestamps, allowing for temporal analysis and comparisons, such as tracking changes in sensor readings over time or observing yield trends across different planting seasons.
  • Unique identifiers: Assigning unique identifiers to entities like sensors, fields and crops enables consistent referencing and data integration across different datasets.

   

  Benefits of identifying layering and commonalities

   

  Understanding these aspects helps you to:

   

  • Organize data effectively: Categorizing data into layers based on abstraction helps with data management, and facilitates analysis at the appropriate level of detail.
  • Facilitate data integration: Identifying commonalities like spatial and temporal references allows you to combine data from different sources for comprehensive analysis.
  • Extract valuable insights: By analyzing different layers and combining data points based on commonalities, you can gain deeper understanding of crop health, soil conditions, and factors influencing yield.

• The data model is used to ensure that the research is organized with respect to the key questions that need to be answered.

   

  • In some research, you already know in advance the key variables you want to measure and the relationships you want to test. In this case, you will already be able to set out a data model with the different elements—say, mineral inputs, soil moisture and crop yields—and the question of the relationship of dependence between them. The data model is self-evident, as the research is to test the effect of minerals and moisture on yields.
  • In other situations, the research is seeking to find patterns in time. The data model of climate research, for instance, must accurately show trends in weather patterns and how this affects farming practices. The data should be organized to evidence long-term patterns of change. The data model must be expanded to include the relation to time, as a central element in the research.

  The data model can also be based on the geographical context. This is the case in Waterways, where the different datasets are layered over the topological map. The layering includes:

   

  • The geographical spread of the satellite and field measurements.
  • The topographical features, such as the canal.
  • The farmer interviews, which are represented by virtual pins, to locate the information in context.

  The data model can be set up to investigate and analyze semantic relations.

   

  The data model is both the organizational template in which the data is gathered, and the frame through which the data product is communicated. The data model is a critical theme in the forming of an analysis.

   

  In Waterways, the data analysis relates the semantic interpretation of the interviews, the satellite data and the topographical features as telling a story together. It is important that one does not interpret this semantic web beforehand, but allows the research to discover the connections.

   

  In the case of Waterways, the surprising relation of the canal water to the surface temperature and crop health was the semantic key to the data model puzzle. The data analysis is only able to find the relation between these elements in the process of questioning the semantic relation between the different layers of data.