Skip to main content
U.S. flag

An official website of the United States government

Scientists at the Earth Resources Observation and Science (EROS) Center have developed an algorithm that fuses Landsat 8 and Sentinel-2 data in a way that generates more accurate daily evapotranspiration (ET) maps with a much higher spatial resolution.

color figure from paper on Landsat and Sentinel harmony
NDVI maps of the California side of the Palo Verde Irrigation District for three selected image dates (first column), and zoomed NDVI maps from Landsat (second column) and Sentinel (third column)

That’s big news for those working in such fields as hydrology, climatology, ecology, and agriculture, where the quantification of actual water consumption is essential and critical. The fused data will mean additional support and benefits to policy makers and water managers trying to make informed decisions on sustaining limited water resources.

In what is basically a proof of concept at this point, Research Physical Scientist Gabriel Senay’s team at EROS has developed a way to integrate actual Landsat-like Sentinel-2 data with the acquisition of cloud-free Landsat 8 data to reduce the amount of interpolation—or estimation—of daily crop water consumption.

What it offers is the potential of greater accuracy compared to the current approach, in which scientists take two cloud-free Landsat images and attempt to fill in the gap between those images through interpolation or estimations involving reference ET from weather datasets.

Senay’s team evaluated its algorithm over an irrigated area on the California side of the Palo Verde Irrigation District for three image dates. What they found in integrating the Sentinel-2 data with the Landsat images was that at an area-wide scale, the fusion approach reduced the relative error in daily ET estimations from 48 percent to 10 percent on average. At a pixel-wide comparison, the relative error dropped from 49 percent to 17 percent.

That said, they did find that the fused data had relatively larger biases across low-vegetation conditions than high-vegetation conditions.

Senay said the fusion algorithm is really an extension of the application of the Operational Simplified Surface Energy Balance (SSEBop) model that he and his team developed to look at ET. SSEBop provides increasingly accurate and repeatable estimates of actual ET—the amount of combined water that either evaporates from the soil and vegetation surfaces or is transpired through plants. Now with the fusion data, they can look at “how we can improve the performance of the SSEBop model when we don’t have thermal data,” Senay said.

Fusing the data for ET is just one example of how scientists are finding new ways to use Landsat and Sentinel-2 data together. “Our biggest problem is not having enough, more frequent, Landsat data,” Senay said. “The more we learn about how these two satellite systems work together, and how they complement each other, the more opportunities. As long as we know what we’re doing, the use of these two datasets will improve our science and decision-making by stakeholders in agriculture and water resources in particular.”

Besides the additional imagery, Sentinel-2 provides its 10-meter resolution to the ET work as well. That’s a higher resolution than Landsat’s 30-meter resolution, or even the 1,000-meter coarser resolution within the thermal data that NASA’s MODerate Resolution Imaging Spectroradiometer (MODIS) brings to ET work.

With Sentinel-2’s 10-meter resolution, “we can blend it in the same fashion as how we blended the image temporally,” Senay said. “By improving our information at the sub-100-meter level, which improves our spatial resolution, farmers will have more information in more parts of their fields.”

color figure from paper on Landsat and Sentinel harmony
The daily evapotranspiration maps of the California side of the Palo Verde Irrigation District for the three selected image dates from Landsat-only (a, d, g-first column), Landsat-interpolation (b, e, h-second column), and Landsat-Sentinel data fusion (c, f, i-third column).

What Sentinel-2 doesn’t have are thermal data required to compute land surface temperature. Under the concept of energy balance principle, farm fields and plants off of which water is evaporating or transpiring are cooler than those where there is less ET occurring. That’s important knowledge in determining water consumption.

However, Sentinel-2 does have a band necessary to help derive a Normalized Difference Vegetation Index, which is valuable for gauging vegetation greenness. Studies have shown that NDVI is closely related to fractionalized ET. Senay said this study focused on evaluating how integrating the ratios of fractionalized ET to NDVI from two satellites could potentially improve the ET estimation during the data gaps from Landsat 8.

Senay and his team have been working with Brazil’s National Water Agency on the use of SSEBop to study irrigation water consumption in that South American country. The Brazilians like what they see with the fused Landsat-Sentinel-2 data, Senay said.

“To monitor field-scale agricultural production and actual ET performance over a large country with known cloud prevalence, they need Landsat or equal-value data,” he said. “That’s why they appreciated how we fused Sentinel-2 with Landsat so that more temporal coverage would be available at field-scale resolution.”

U.S. farmers trying to decide how much water to apply and when to apply it to croplands would like to get the kind of accurate daily ET information that his fusion project could offer as well, Senay said.

“We can convert the information we’re gathering into a daily evapotranspiration maps so farmers know at what rate the crops are transpiring and then can adjust their irrigation scheduling,” he said. “I can see farmers using it for such purposes.”

Growers working with irrigation scheduling companies can use the fused ET maps to water at different rates for different crops, he said. State water managers or state engineers looking at water volumes for different irrigation districts can use the more accurate maps to estimate water demands downstream.

Having tested the fusion model originally in a relatively cloud-free environment, Senay said the next step will involve expanding their study to larger areas where clouds may become more of a factor. In time, as they expand to more areas of the U.S. and globally, he said they will likely move to a cloud computing environment to process larger and larger amounts of data as well.

“We have to expand this study to where clouds are problems, which means our interpolation is going to get harder and harder,” Senay said. “It will be challenging, but if it works the way we expect it to, the benefits will be even higher in those cloud-affected regions.”

Related Content