Photo Credit: David Clode
Scientific Frontline: Extended "At a Glance" Summary: Optical Earth Observation for Surface Water Detection
The Core Concept: The application of optical satellite imagery and machine-learning algorithms to detect and map seasonal surface water extents across varying geographic landscapes.
Key Distinction/Mechanism: High-resolution commercial satellite datasets (such as PlanetBasemap at 4-meter resolution) excel at detecting small, unobstructed bodies of water using visible and near-infrared bands. Conversely, moderate-resolution public datasets (such as the USGS Landsat Dynamic Surface Water Extent at 30-meter resolution) incorporate shortwave infrared bands, making them vastly superior at detecting surface water obscured by dense vegetation and forest canopy.
Major Frameworks/Components:
- PlanetScope Basemap: A high-resolution (4.77 m) commercial dataset limited to red, blue, green, and near-infrared spectral bands, optimizing precise pixel-level detail for small-scale geographic features.
- Dynamic Surface Water Extent (DSWE): A publicly available, moderate-resolution (30 m) dataset derived from the Landsat program that utilizes shortwave infrared bands to penetrate vegetative cover.
- Machine-Learning Classification: Algorithmic sorting of spectral band data to categorize pixels into "confidence classes," quantifying the probability of surface water presence.
Branch of Science: Remote Sensing, Hydrology, Environmental Science, and Geospatial Analytics.
Future Application: Advanced modeling of flood dynamics, precise monitoring of localized methane emissions from small ponds, and more accurate predictions regarding downstream water volume and ecological quality.
Why It Matters: Precise surface water detection is foundational to global environmental monitoring. Identifying the specific operational thresholds of spatial resolution versus spectral breadth enables researchers to optimize data selection based on study area size, seasonal vegetation density, and specific hydrological targets.
A new study finds that commercial satellite imagery data often outperforms public data sets when identifying surface water, but that public data sets may be better at detecting water hidden by forest cover.
Satellite imagery is a powerful tool for mapping surface water, from the movement of rivers and streams to water levels and even water temperatures. The effectiveness of those satellites depends on their ability to identify water in the images they capture. To do this, satellites use machine-learning algorithms to analyze color data across spectral bands, many of which are not visible to the human eye. This information comes from data sets that are either purchased commercially or available to the public, with commercial data typically having higher-resolution images with far more detail at the pixel level.
To understand the impact of higher-resolution imagery in detecting surface water, researchers compared the commercial PlanetBasemap data set to the Dynamic Surface Water Extent, a public data set built from the United States Geological Survey Landsat program. Lead author Mollie Gaines, who led the study as Ph.D. candidate at North Carolina State University, said that Planet Basemap’s higher resolution made it more capable of detecting small bodies of water.
“The Planet data is approximately four-meter resolution, which means that each pixel is approximately a four-by-four-meter square. That leads to a much more detailed image compared to the DSWE’s 30-meter resolution,” she said. “We’re seeing that the commercial data set often identifies more of the smaller water bodies, as well as river extents.”
However, Gaines said, that changes during seasons when high levels of vegetation obscure the water. The public DSWE data captures a wider portion of the electromagnetic spectrum resolution than PlanetBasemap, which makes it particularly good at detecting water hidden underneath vegetation.
“The Planet Scope data, which is what PlanetBasemap is built on, is limited to red, blue and green, or what the human eye can see, and near infrared,” she said. “DSWE includes the shortwave infrared band, which is the best option for this kind of water detection.”
This benefit was most pronounced when researchers included all three of DSWE’s “confidence classes,” categories that the classified satellite imagery data is sorted into based on how likely it is to contain water. With all three classes included, DSWE data captured more water in places like streams and rivers, where their winding paths can sometimes throw off imagery classifications.
These results show that both data sets have legitimate use cases, and that publicly available data is a strong option when used in the right circumstances.
“When studying very small bodies of water like ponds, the commercial data is the more reliable product,” she said. “But if you’re looking at a larger study area, the publicly available product is a really good option.”
Funding: This research was supported by NASA FINESST Grant 80NSSC21K1606, NASA CSDA Grant 80NSSC24K0053, and MGT’s funding through NC State. This work utilized data made available through the NASA Commercial Satellite Data Acquisition (CSDA) program.
Published in journal: Geophysical Research Letters
Title: Impact of Spatial Scale on Optical Earth Observation-Derived Seasonal Surface Water Extents
Authors: Mollie D. Gaines, Mirela G. Tulbure, Vinicius Perin, Darcy Boast, Henry Castellanos Quiroz, Rebecca Composto, Varun Tiwari, and Júlio Caineta
Source/Credit: North Carolina State University | Joey Pitchford
Reference Number: es032726_01
.jpg)
.jpg)