General data investigation conducted for each parameter of interest for processing

[soutobias]

General data investigation conducted for each parameter of interest for processing, to determine (i) any images to be removed or subsetted, (ii) criteria or threshold values for processing.

Think about the desired aim of this processing step with reference to the overall biodiversity aims of the project/deployment, and any ancillary aims (e.g., retaining the maximum number of images). Establish what would be considered successful processing and any criteria for testing that success.

• Think about important conditions or occurrences identified from the cruise report or deployment notes that might have affected the parameter of interest that might particular processing or special care (e.g. image removal), or suggest boundaries for the parameter
• Plot parameter of interest along deployment(s), and also with camera position (latitude, longitude, water depth, altitude above seabed), with camera performance metrics, and as a histogram. Note patterns/shapes of curves and parameter values and inflection points, extreme values that may be important for setting criteria or thresholds.
• Consider any statistics that might be helpful in determining thresholds (e.g., identifying extremes or outliers, identifying a range of acceptable values).
• Examine images from both outside and within any threshold values to confirm visually that the threshold is correct.
• Test the processing step using the criteria established above.
• Repeated the data investigation steps (3-5) to the retained/processed images to determine success, employing any success tests established in step 1

[soutobias]

What:

General data investigation involves a thorough analysis of each parameter of interest related to image processing. This step is crucial for determining whether any images should be removed or subsetted and for establishing criteria or threshold values for processing. It helps in understanding the data distribution, identifying important patterns, and ensuring that the processing aligns with the overall objectives of the project.

Why:

Conducting a general data investigation ensures that the processing criteria are well-founded and tailored to the specific characteristics of the dataset. This is essential for achieving meaningful and accurate results, particularly in biodiversity studies where precise image analysis can impact the assessment of species and habitats. It also helps in setting appropriate thresholds and detecting any anomalies that could affect the quality and reliability of the processed data.

How:

1. Define Aims and Success Criteria:

   • Determine the primary objective of the processing step in relation to the overall biodiversity goals of the project. For example, the aim could be to retain the maximum number of usable images while ensuring high quality.
   • Establish what constitutes successful processing, including specific criteria or metrics for evaluating success. For instance, successful processing might involve retaining images within a defined quality range or achieving certain statistical thresholds.

2. Consider Conditions from Deployment Notes:

   • Review cruise reports or deployment notes for any conditions or occurrences that might have impacted the parameter of interest. For example, specific environmental factors, equipment malfunctions, or operational constraints should be considered, as they may affect image quality and influence decisions about image removal or special processing needs.

3. Visualize Data:

   • Plot the parameter of interest across the deployment or image capture event, including camera position metrics (latitude, longitude, water depth, altitude above seabed) and camera performance metrics. Create histograms to visualize the distribution of parameter values.
   • Analyze patterns, shapes of curves, inflection points, and extreme values. Identify any trends or anomalies that might help in setting appropriate thresholds or criteria.

4. Determine Thresholds and Criteria:

   • Use statistical methods to identify extremes, outliers, and acceptable ranges for the parameter of interest. For example, calculate mean, median, standard deviation, and percentiles to establish thresholds.
   • Plot images within and outside these thresholds to visually verify that the chosen criteria correctly distinguish between usable and non-usable images.

5. Test and Refine Criteria:

   • Apply the established criteria to the image dataset and evaluate the processing step based on the defined success metrics. Assess whether the criteria effectively identify and handle images according to the project's goals.
   • Repeat the data investigation steps as necessary to refine the criteria. Reevaluate the processed images to ensure that the processing meets the success criteria and aligns with the project's aims.

What to expect:

The outcome of this investigation should be a well-defined set of criteria or thresholds for processing images, informed by statistical analysis and visual verification. The processed dataset should reflect the quality and characteristics desired for achieving the project's biodiversity objectives.

What makes it difficult:

• Complexity of Data: Large and diverse datasets may present complex patterns, making it challenging to establish clear and effective criteria or thresholds.

• Subjectivity: Decisions about which images to remove or retain can be subjective, depending on the criteria set and the goals of the processing.

• Dynamic Conditions: Changes in environmental conditions or camera performance during the deployment can introduce variability that complicates the determination of thresholds.

Success Metrics:

• Accuracy of Thresholds: Successful identification of appropriate thresholds and criteria for processing, ensuring that the processed images meet the project's quality standards.

• Retention of Usable Images: The percentage of images retained after processing should align with the project's goals, such as maximizing usable data while minimizing errors.

• Consistency with Objectives: The processed dataset should effectively support the biodiversity assessment objectives, with clear and reliable images for analysis.

[LoicVA]

I would add here: identify and verify relationships with other variables. Is this Jen's document? Otherwise, I would rather correct that one directly.