US Geological Survey - science for a changing world

USGS Flow Photo Explorer Newsletter

U.S. Geological Survey sent this bulletin at 04/16/2025 07:00 AM EDT

Thanks for your interest in the U.S. Geological Survey Flow Photo Explorer (FPE) tool! This ad hoc newsletter is to inform you of updates to this platform. You have been added to this list due to your prior involvement in the project. If you do not want to receive future FPE newsletters, please use the unsubscribe button at the bottom of this email.

What’s New?

A publication pre-print describing model performance

We recently submitted a paper to the journal HESS describing the performance of the modelling system at 11 camera sites co-located with USGS reference gages in western Massachusetts. We are excited to share these results, which help guide how we communicate the use of this tool. A pre-print of the paper is available here: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-1186/

A growing network

Thanks to federal, state, local, university, tribal, and NGO partners, we now have more than 8 million photos at more than 400 sites and have trained over 60 ranking models. Our USGS camera network is expanding rapidly, including deployments this year focused on studying stream baseflow resilience in the Delaware River basin.

Big “Thank You!” to annotators

Without annotations we wouldn’t be able to train models! So thanks to all who have completed annotations for their sites. We have also had a huge amount of help from high school students from across the country through the USGS STEP-UP program. We are proud of them and very much appreciate all their hard work!! You can read more about the program here: STEP-UP: Secondary Transition to Employment Program – USGS Partnership | U.S. Geological Survey

Testing Methods Developments

We have been experimenting with better ways to do the modeling, including by adding in weather information, testing different types of models, and building models using the images from multiple sites. In general, we find that our original approach is hard to improve upon, giving us confidence to continue with the same methodology. Our web developer and data scientist Jeff Walker produced a summary report that is visible here: https://doi.org/10.5281/zenodo.15133341

What’s Planned

New frontiers in monitoring

We have pilot ranking models developed for algae monitoring, snow depth monitoring, wildlife monitoring, wetland monitoring, spring monitoring, and more! In the not-to-distant future we hope to upgrade the platform to more directly support these new non-streamflow applications. It might involve a name change for the platform – stay tuned!

Improved imagery screening

One of the limitations on model performance is the presence of some images that are obscured by glare, vegetation, fog, or are simply too dark. We’ll be working on automated screening of these images to boost model performance and we’ll skip showing these images to annotators to save them a few clicks.

Annotation training module

We are finalizing an annotation training module of 250 annotations to better track how annotation performance varies among people. This will support our desire to build the best models with the fewest annotations. Plus, sometimes people want to practice annotations before moving on to working on their sites. We may request some account holders to (voluntarily) complete this module.

Frequently Asked Questions

How can I be involved?

Uploading imagery and performing annotations is the best way to contribute! If you have a specific non-streamflow monitoring application you are interested in, please contact us at ecosheds@usgs.gov.

Does the model predict volumetric streamflow?

The model predicts a relative measure of streamflow that we call the model score, which is correlated with streamflow. Using these scores, we can sort the images from lowest to highest score, which also sorts them from lowest to highest predicted flow. To compare the predicted scores against observed flows, we can compute the rank percentiles of each, which is the rank of each image based on its score (or flow) and then divide by the number of images to scale from 0 to 100%. We are actively working on a methodology for converting predicted model scores directly to volumetric streamflow, but for now, the rank percentiles are the best way to compare predicted model scores directly to observed flows.

Can I have a model built for my site to estimate streamflow?

Yes! We are currently running models for sites that have 1,500 or greater annotations and 1 year or more of imagery. Our paper shows that performance tends to level off after 1,000 annotations used train a model; we are asking for a few extra in case those annotations contain bad photos or the user selection is “unknown”.

If you think your site meets these criteria but it doesn’t have a model yet, you can email us at ecosheds@usgs.gov to check status.

Are the modeling results reliable and citable?

Currently all modeling results should be considered provisional and is subject to revision. It is being provided to meet the need for timely best science. The information has not received final approval by the U.S. Geological Survey (USGS) and is provided on the condition that neither the USGS nor the U.S. Government shall be held liable for any damages resulting from the authorized or unauthorized use of the information.

Best,

The Flow Photo Explorer Team

Update your subscriptions, modify your password or email address, or stop subscriptions at any time on your Subscriber Preferences Page. You will need to use your email address to log in. If you have questions or problems with the subscription service, please visit subscriberhelp.govdelivery.com.

This service is provided to you at no charge by U.S. Geological Survey.