I Get Creative to Improve My Model

This week I have worked to make adjustments to my own model in preparation for an invited research talk and upcoming conference. While my model works well, I decided to revisit a few data points, make some improvements, and try and improve some aspects in the calibration process. As I mentioned last week, calibration is the process of using a time series of data and, often, a time series of environmental variables, to guide the model to replicate trends. When a model has a lot of data to work with, it often performs better if future conditions reflect a condition from the past. I am not training my model to predict microplastic pollution patterns of the past, but I need the model to appropriately reflect how the model groups change based on temperature and salinity conditions, so that a model group does not behave unexpectedly in forecast scenarios. 

For nearly all of the lab's modeling work, we turn to data collected by state and federal agencies as part of their independent monitoring programs. We also rely on stock assessments, which often evaluate species across the entire Gulf of Mexico (or Atlantic, Pacific) and are themselves an amalgamation of independent surveys, fishing vessel reported catches, and mathematical relationships and equations. Many of these stock assessments are freely available and published semi-frequently, so they are useful for many model groups, especially groups of key fishery species. Unfortunately, many of these stock assessments are really long and they sometimes refer the reader to another document to find the relevant information. Additionally, sometimes that data I need are presented in figures, rather than tables, so I have to get creative with extracting the data. I can upload a figure to an online tool that lets me drag and drop two x-coordinates and two y-coordinates and assign values to each. Then I can click anywhere on the figure, and the tool will tell me the x- and y-coordinates of the point based on its proximity to the coordinates I initially set. While this tool is perhaps not as exact as the scientists publishing the exact numbers, I can zoom in and work slowly to get usable information that I would otherwise miss in my model. 

I am hoping to finish up my model adjustments by early next week so that I can perform new calibrations, runs, and analyses and still have time to make any corrections before my talk in late-September. Also, we have a few exciting trips and events coming up in September, so stay tuned for blogs from our adventures on the water (and off)!