interstitialRedirectModalTitle

interstitialRedirectModalMessage

Stories in Arizona

Using Satellite Images & Machine Learning to Map Surface Water

Protecting nature starts with science.

Ariel shot of a river and a green landscape with mountains in the distance.
Lower San Pedro River One of the last large, undammed rivers in the Southwest, the San Pedro's flow is not continuous, and shrinking, raising concern for the critical habitat the river supports. © Adriel Heisey

Voices From the Field

"Nothing is more critical to Arizona right now than understanding and tracking our water systems. Technology offers us the chance to generate surface water maps in unprecedented detail." -Jessie Pearl, TNC Arizona's Freshwater Scientist

Project Objective

In Arizona, water can be fickle. Major rivers may be vibrant, aquatic systems may be resilient for miles, then run completely dry for a stretch, and then return to a full flow further downstream. Understanding where and when water is flowing is crucial to helping us assess a river system’s health and informing our water management decisions. The challenge is how to collect the most accurate and useful information about surface water flows.

Since 1999, The Nature Conservancy (TNC) has worked with volunteers and partners to ‘wet/dry’ map the San Pedro River, one of the last major undammed rivers in the southwestern United States. The goal of the mapping is to assess aquatic connectivity and to evaluate the effectiveness of conservation actions on surface water presence. Volunteers conduct the survey every June, the hottest and driest time of year, to capture an assessment of baseflow. 

While this 25+ year data set is rich, it has limits. Each year, the length and location of the survey changes, making consistent analysis challenging, and the collection of one data point per year may not capture other seasons or phenomena of interest (for example, a large monsoon season that occurs after the survey).

A person wearing a backpack walks toward a small waterfall with sunlight streaming in from above.
Ramsey Canyon Preserve survey Appropriately training the model requires wet/dry surveys outside of the summer months. Here, Dr. Pearl surveys water presence in late winter. © Dana Lapides

TNC Arizona freshwater scientist Dr. Jessie Pearl is now working with Dr. Dana Lipides from the USDA Agricultural Research Service (ARS) and Dr. Shang Gao from the University of Arizona to determine if they can use remotely sensed imagery and machine learning to more accurately map surface water presence in the San Pedro River basin across the year. 

Two people stand in a narrow, rocky area while looking above.
Box Canyon Certain environments, like a box canyon, can make detecting surface water presence from space tricky! © Jessie Pearl

Progress & Opportunity 

Today, commercial satellites are revolutionizing the data available to scientists, and they can offer unprecedented access to information about the presence of water in a stream channel. The challenge is processing the sheer quantity of remotely sensed imagery data for analysis. That’s where Dr. Pearl, Dr. Lipides and Dr. Gao hope a machine learning model can help. What is machine learning? It’s a specific type of artificial intelligence (AI). Machine learning can identify patterns and extract valuable insights efficiently from the thousands of images taken by satellites. In this case, the science team is building a machine learning model to detect water presence in the San Pedro almost daily with high-resolution satellite images. This new approach wouldn’t be possible, though, without the valuable data collected through the annual volunteer wet/dry mapping. That’s because scientists are using the ground-based observation data to train their machine learning model. By harnessing the latest technology, the scientists are greatly increasing the usability of the field-collected wet/dry data.

The team already had initial success piloting this model in the regions of Cienega Creek Basin and Ramsey Canyon, and they’re hoping to expand its use this year—providing nearly real-time data to inform the way we protect and sustain these dynamic and vulnerable water resources.

Animated image showing the results of a preliminary machine learning model of surface water presence/absense.
Surface Water of Ramsey Creek Preliminary machine learning model results of surface water presence/absence in Ramsey Canyon from 2016-2024. We've gone from 1 data point a year to wet/dry records for almost every day over the seven years. Red is dry, blue is wet on stream line. Total wetted length (y axis) is how many km of Ramsey Canyon Creek are wet over time. © Dana Lapides
View of a narrow river reflecting the sunlight with lush trees and vegetation surrounding it.
Flowing stream San Pedro Riparian National Conservation Area © Laura Fawcett