Photogrammetry Insights

By Leila Lemos, PhD Candidate, Fisheries and Wildlife Department, Oregon State University

After three years of fieldwork and analyzing a large dataset, it is time to finally start compiling the results, create plots and see what the trends are. The first dataset I am analyzing is the photogrammetry data (more on our photogrammetry method here), which so far has been full of unexpected results.

Our first big expectation was to find a noticeable intra-year variation. Gray whales spend their winter in the warm waters of Baja California, Mexico, period while they are fasting. In the spring, they perform a big migration to higher latitudes. Only when they reach their summer feeding grounds, that extends from Northern California to the Bering and Chukchi seas, Alaska, do they start feeding and gaining enough calories to support their migration back to Mexico and subsequent fasting period.

 

Northeastern gray whale migration route along the NE Pacific Ocean.
Source: https://journeynorth.org/tm/gwhale/annual/map.html

 

Thus, we expected to see whales arriving along the Oregon coast with a skinny body condition that would gradually improve over the months, during the feeding season. Some exceptions are reasonable, such as a lactating mother or a debilitated individual. However, datasets can be more complex than we expect most of the times, and many variables can influence the results. Our photogrammetry dataset is no different!

In addition, I need to decide what are the best plots to display the results and how to make them. For years now I’ve been hearing about the wonders of R, but I’ve been skeptical about learning a whole new programming/coding language “just to make plots”, as I first thought. I have always used statistical programs such as SPSS or Prism to do my plots and they were so easy to work with. However, there is a lot more we can do in R than “just plots”. Also, it is not just because something seems hard that you won’t even try. We need to expose ourselves sometimes. So, I decided to give it a try (and I am proud of myself I did), and here are some of the results:

 

Plot 1: Body Area Index (BAI) vs Day of the Year (DOY)

 

In this plot, we wanted to assess the annual Body Area Index (BAI) trends that describe how skinny (low number) or fat (higher number) a whale is. BAI is a simplified version of the BMI (Body Mass Index) used for humans. If you are interested about this method we have developed at our lab in collaboration with the Aerial Information Systems Laboratory/OSU, you can read more about it in our publication.

The plots above are three versions of the same data displayed in different ways. The first plot on the left shows all the data points by year, with polynomial best fit lines, and the confidence intervals (in gray). There are many overlapping observation points, so for the middle plot I tried to “clean up the plot” by reducing the size of the points and taking out the gray confidence interval range around the lines. In the last plot on the right, I used a linear regression best fit line, instead of polynomial.

We can see a general trend that the BAI was considerably higher in 2016 (red line), when compared to the following years, which makes us question the accuracy of the dataset for that year. In 2016, we also didn’t sample in the month of July, which is causing the 2016 polynomial line to show a sharp decrease in this month (DOY: ~200-230). But it is also interesting to note that the increasing slope of the linear regression line in all three years is very similar, indicating that the whales gained weight at about the same rate in all years.

 

Plot 2: Body Area Index (BAI) vs Body Condition Score (BCS)

 

In addition to the photogrammetry method of assessing whale body condition, we have also performed a body condition scoring method for all the photos we have taken in the field (based on the method described by Bradford et al. 2012). Thus, with this second set of plots, we wanted to compare both methods of assessing whale body condition in order to evaluate when the methods agree or not, and which method would be best and in which situation. Our hypothesis was that whales with a ‘fair’ body condition would have a lower BAI than whales with a ‘good’ body condition.

The plots above illustrate two versions of the same data, with data in the left plot grouped by year, and the data in the right plot grouped by month. In general, we see that no whales were observed with a poor body condition in the last analysis months (August to October), with both methods agreeing to this fact. Additionally, there were many whales that still had a fair body condition in August and September, but less whales in the month of October, indicating that most whales gained weight over the foraging seasons and were ready to start their Southbound migration and another fasting period. This result is important information regarding monitoring and conservation issues.

However, the 2016 dataset is still a concern, since the whales appear to have considerable higher body condition (BAI) when compared to other years.

 

Plot 3:Temporal Body Area Index (BAI) for individual whales

 

In this last group of plots, we wanted to visualize BAI trends over the season (using day of year – DOY) on the x-axis) for individuals we measured more than once. Here we can see the temporal patterns for the whales “Bit”, “Clouds”, “Pearl”, “Scarback, “Pointy”, and “White Hole”.

We expected to see an overall gradual increase in body condition (BAI) over the seasons, such as what we can observe for Pointy in 2018. However, some whales decreased their condition, such as Bit in 2018. Could this trend be accurate? Furthermore, what about BAI measurements that are different from the trend, such as Scarback in 2017, where the last observation point shows a lower BAI than past observation points? In addition, we still observe a high BAI in 2016 at this individual level, when compared to the other years.

My next step will be to check the whole dataset again and search for inconsistencies. There is something causing these 2016 values to possibly be wrong and I need to find out what it is. The overall quality of the measured photogrammetry images was good and in focus, but other variables could be influencing the quality and accuracy of the measurements.

For instance, when measuring images, I often struggled with glare, water splash, water turbidity, ocean swell, and shadows, as you can see in the photos below. All of these variables caused the borders of the whale body to not be clearly visible/identifiable, which may have caused measurements to be wrong.

 

Examples of bad conditions for performing photogrammetry: (1) glare and water splash, (2) water turbidity, (3) ocean swell, and (4) a shadow created in one of the sides of the whale body.
Source: GEMM Lab. Taken under NMFS permit 16111 issued to John Calambokidis.

 

Thus, I will need to check all of these variables to identify the causes for bad measurements and “clean the dataset”. Only after this process will I be able to make these plots again to look at the trends (which will be easy since I already have my R code written!). Then I’ll move on to my next hypothesis that the BAI of individual whales varied by demographics including sex, age and reproductive state.

To carry out robust science that produces results we can trust, we can’t simply collect data, perform a basic analysis, create plots and believe everything we see. Data is often messy, especially when developing new methods like we have done here with drone based photogrammetry and the BAI. So, I need to spend some important time checking my data for accuracy and examining confounding variables that might affect the dataset. Science can be challenging, both when interpreting data or learning a new command language, but it is all worth it in the end when we produce results we know we can trust.

 

 

 

Print Friendly, PDF & Email

Leave a Reply

Your email address will not be published. Required fields are marked *