Study and summarize methods and key findings from hydrometeorological data integration paper

[DaniJonesOcean]

Issue Description:

To enrich our understanding of Great Lakes surface temeprature and hydrological data analysis, and to broaden our perspective on sensor data integration, let's review existing research in the field. Please read and summarize this highly relevant paper linked to below. I'm especially interested in how this work compares with our DeepSensor project.

The paper in question presents a method for incorporating nontraditional measurements (like those made by ship captains) to enhance hydrometeorological estimates in the Great Lakes. Here’s the citation:

Fries, K., and B. Kerkez (2017), Big Ship Data: Using vessel measurements to improve estimates of temperature and wind speed on the Great Lakes, Water Resour. Res., 53, 3662–3679, https://doi.org/10.1002/2016WR020084

Issue activities:

• Read the paper thoroughly, focusing on the methodologies proposed for integrating diverse measurement sets.
• Understand the probabilistic framework introduced and how it was applied to create a new spatial data product.
• Summarize the paper's primary contributions, methodologies, and analysis of the data product.
• Identify how this framework compares to our GreatLakes-TempSensor project, specifically the DeepSensor tool, and what insights we can draw from it.

Please also review any key references within the paper that you believe will provide further context or be generally useful.

Please share your findings and thoughts in a format that you're comfortable with (such as a written document, a series of slides, or a short presentation) and let me know if you run into any problems!

[eredding02]

@DaniJonesOcean I have provided a link with my notes for this paper

https://docs.google.com/document/d/1E8YLyscAf48ptJjeuP4sltQjmOHp0JP2bGJ_KAEYqys/edit?usp=sharing

[DaniJonesOcean]

@eredding02 Great summary, thanks! I love your idea of pre-training with a physical model and then refining the pre-trained weights using observations. I've seen that used in a convolutional neural network sea ice forecasting model, where the model was pre-trained with climate model data and then refined using satellite data. It seemed to work well (although surprisingly, if I recall correctly, the pre-training didn't help as much as one might think). I think pre-training would be a good area to explore - I think the first step would be identifying a good source of model data that is validated and trusted, such as the FVCOM model data that David Cannon provided us.

Happy for you to move this to the "done" column and close the issue when you're ready!

[DaniJonesOcean]

Oh yes, and for some reason the link above didn't work. This one seems to:

https://docs.google.com/document/d/1E8YLyscAf48ptJjeuP4sltQjmOHp0JP2bGJ_KAEYqys/edit?usp=sharing