Automatic Image Processing to Determine the Community Size Structure of Riverine Macroinvertebrates

Body size has been prioritized as a key trait for monitoring. Yet measuring individual size is time consuming. Our method allows the routine use of macroinvertebrate size spectrum to assess impacts on freshwater ecosystems.

Our protocol provides a standardized way to automatically determine individual size distribution of river macroinvertebrates in a sample in about an hour. Demonstrating the procedure will be Rosa Guri from the University of Vic. To begin, turn on the scanner.

And switch on the light in the dual position. To project white light from top and bottom. Next, clean and rinse the scan tray with tap water.

To create the blanks, pour 110 milliliters of tap water at room temperature into the scan tray until the glass is covered. Place the large frame on the scan tray with the corner at the top left part. And fill it with tap water until it covers the step of the frame to avoid a meniscus effect which would alter scanned images.

Next, open the image processing software, select the working project, and click on scan convert background image. Next, open the scanning software and click on preview. Check the image for no lines or spots and wait for at least 30 seconds before starting the scan.

Then press okay in the instructions window before the second scan to send the data from the scanning software to the image processing software. Pour 110 milliliters of 70%ethanol into the scan tray until the glass is covered. And place the large frame.

Then pour the macroinvertebrates sample into the scan tray edged by the frame and cover it with more ethanol if needed. Then, using a wooden needle, homogenize the sample throughout the frame area. Placing the largest individuals in the center of the tray for proper image processing.

And sink the floating organisms. Also, separate apart the clustered organisms. And pull in the organisms touching the frame edges to the center.

Next, proceed to scan by clicking on scan sample with zoo scan for archive no process. Selecting the sample and following the instructions. Preview the image in the scanning software for no lines or spots.

After 30 seconds, click on the scan button in the scanning software. And verify that the raw scanned image is correct. Remove the frame and wash it above the scan tray with a 70%ethanol filled squeeze bottle to recover any attached macroinvertebrates.

Lift the upper part of the scanner to retrieve all of the organisms and ethanol through the scan retrieval funnel into a beaker. With the upper part of the scanner still lifted, clean the tray with the squeeze bottle to sweep along any remaining organisms. After recovering all of the specimens, clean the tray with tap water.

To predict the identity, click on data analysis in the automatic identification software. In select learning file, locate PID_Process, then select learning_set to choose the learning set file to be used. Next, in select sample files, choose the sample to be predicted from the PID_results folder.

In select a method choose the random forest method and then tick the save detailed results for each sample button. In original variables, untick the position variables. Lastly, in customized variables, tick only ESD.

Click on start analysis and save the results as analysis_name. txt in the PID_process prediction folder. To manually validate, copy the analysis sample_dat_1 txt files from the PID_process prediction folder to the PID_process, PID_results folder.

Select extract vignettes in folders according to prediction or validation in the image processing software. Then select used predicted files from PID_results folder. And with the default settings, pressing okay creates a new folder.

Now go to the PID_process sorted vignettes folder and copy the newly created folder named with the sample name, date, and time to validate. Rename the folder to validate with validated. To manually validate the automatic classification, open the renamed folder sample name, date, and time validated.

Review all the vignettes from each subfolder to identify misclassified objects if any. When an object is misclassified, drag the vignette to the correct folder. Next, select load identifications from sorted vignettes.

Keeping the default settings, select the file named date, time, name validated to be processed. Then go to PID_process, PID_results, and then dat1_validated. And open the file named ID_from_sorted_vignettes date and time txt to verify that the last column, prediction, validated ID, date and time, which specifies the expert classification of each object, has been created.

While testing the system for macroinvertebrates, some scans were of poor quality in the processed images. Despite this, a fine sub-sample with good scan quality of the raw and the processed image looks as depicted. In the set of analyzed vignettes, 86.1%corresponded to debris.

Including detritus, fibers, body parts, or scanning artifacts. And the remaining 13.9%of the detected objects corresponded to the invertebrate organisms. Automatic recognition followed by manual validation of objects exhibited high recall for all the categories.

While the contamination one precision was rather low except for the other invertebrates. The comparison of automatic versus validated macroinvertebrate abundance showed a high correlation with a slight overestimation by the automatic performance due to contamination from debris. The probability density functions of the individual size distributions of the automatic prediction strongly occurred with the validated predictions for the fine and coarse subsamples.

The subsamples validated after the separation of touching objects from selected natural samples showed increased abundance. But the mean ellipsoidal volume was close to the validated samples. The size distributions of corrected samples differed slightly from the validated ones.

But a strong correlation was observed. The normalized bio volume size spectrum was similar between the treatments. Except for a few size classes in a couple spectra.

Take the required time to separate touching organs to obtain a fine scan and clean with water this scan tray after every scan with ethanol to avoid precipitation. We have adopted this procedure created for plankton to river macroinvertebrates. Thus, it could potentially be adapted to obtain individual body sizes of other finer groups.

for example, terrestrial invertebrates. This method will allow a systematic estimation of macroinvertebrate community size structure over large, special, and temporal gradients. And investigate the role of body size in river ecosystem functioning and by assessment.