Water, Water Everywhere - But How to Find It?
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.”
The paper, “Impact of spatial scale on optical Earth observation-derived seasonal surface water extents,” is published in Geophysical Research Letters. Co-authors include Mirela G. Tulbure, Darcy Boast, Rebecca Composto, Varun Tiwari and Júlio Caineta, NC State University; Vinicius Perin, Planet Labs Inc.; and Henry Castellanos Quiroz, Colombia Institute of Hydrology, Meteorology and Environmental Studies.
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.
-pitchford-
Note to editors: The abstract of the paper follows.
Impact of Spatial Scale on Optical Earth Observation-Derived Seasonal Surface Water Extents
Authors: Mollie Gaines, Mirela G. Tulbure, Darcy Boast, Rebecca Composto, Varun Tiwari and Júlio Caineta, NC State University; Vinicius Perin, Planet Labs Inc.; and Henry Castellanos Quiroz, Colombia Institute of Hydrology, Meteorology and Environmental Studies.
Published: Feb. 5, 2026 in Geophysical Research Letters
DOI: 10.1029/2025GL119880
Abstract: Landsat-derived products are the most prominent, publicly available sources of large-scale surface water extent data. However, few studies have assessed the limitations of spatial scale on such products. Here, we mapped seasonal surface water extents utilizing high-resolution (4.77 m) PlanetScope Basemap imagery and machine learning. We conducted a pixel-wise comparison of these high resolution classifications with a set of classifications from a moderate resolution (30 m) Landsat product. The vast majority (< 93%) of areas classified as water by the Landsat product were similarly classified by PlanetBasemap; however, only 65%–75% of the PlanetBasemap water area was also classified by the Landsat classes. Of the Landsat classes, only the partial surface water class comparably detects smaller water bodies (widths > 50–70 m) with PlanetBasemaps. Our results indicate that higher resolution imagery detects more small water bodies, which are instrumental to better understanding flood dynamics, methane emissions, and downstream water volume and quality.
Geophysical Research Letters
Imaging analysis
Not applicable
5-Feb-2026
The authors report no conflicts of interest