We are all aware that the planet is orbited by a wide array of Earth observation satellites, each transmitting back specialized data used in Earth science research. As valuable as these individual data streams are, when data scientists effectively combine two or more of these streams into a single data set, the value can increase even more. This is the purpose of the Harmonized Landsat Sentinel-2 (HLS) project led by IMPACT.
Current publicly available land surface observation data requires the science user community to pick between high resolution data available every couple of weeks or coarser resolution data available every day. By harmonizing observations from platforms with similar instrument characteristics, the HLS project provides a unique opportunity for the science community to get plot-scale (30-meter resolution) observations every 2–4 days. HLS data products should greatly improve upon current capabilities for applied science initiatives related to natural disasters, land cover land use change, and seasonal vegetation health.
At a high level, the HLS project exploits the similarity between the United States Geological Survey Landsat-8 sensor and the European Space Agency Sentinel-2 satellite sensor. The HLS project takes input data from Landsat-8 and Sentinel-2 to generate a harmonized analysis ready data product that improves the temporal resolution of these sensors when used independently. It reduces the subtle variation within each data stream and between the two data streams to provide high quality and high temporal frequency Earth surface measurement at moderate spatial resolution.
The HLS project produces surface reflectance observations from 0.4–2.3 micron spectral range for both the L30 and S30 data products. In addition, the L30 data product includes the thermal bands available on Landsat-8 reprojected onto the 30 meter HLS grid. What this means is that the HLS project provides high radiometric quality, high temporal frequency and coregistered HLS data in an easy-to-use format. In other words, it provides Analysis Ready Data (ARD), which in general eliminates the need for users to perform demanding data preprocessing and thus allows more people to use the data and to focus on their scientific investigations. Its high quality, time series and global scale combine to make the HLS data valuable for the monitoring of crops, forests, pasture, and wetlands.
The value of the HLS data set is that it is not only forward-looking as new data flows into it on a daily basis, but the project team has worked to extend the data set backwards in time. IMPACT data scientists will also generate data reaching back to 2013 and 2015 for the L30 and S30 data products respectively.
Team member Dr. Junchang Ju describes the appeal of the HLS project:
I got my PhD degree in vegetation remote sensing by using only a few Landsat images over a relatively very small area. Free access to the satellite data and the availability of supercomputing power really changes the game. I’m very excited to command a huge cluster of computers to do remote sensing work that couldn’t be imaged two decades ago.
In developing the data products for HLS, we tried to be cognizant of the fact that cloud-friendly data formats are becoming more widely used by the Earth science community. In addition, analysis from cloud-friendly data formats has been much more efficient than traditional access methods. In line with that trend, we decided to produce HLS data products as cloud-optimized-GeoTIFFs (COGs) to increase the capability of the analysis that can be done with it. One of the main applications of HLS is time-series analysis of the land surface for intra- and interseasonal land surface observations. By providing the data products in a cloud-friendly data format, the user community should be able to perform really exciting research that may not have been possible with more traditional data formats.
Currently the HLS data is in a beta release phase where data has been distributed to 9–10 frequent users of the data. In the coming weeks, there will be a second internal release to a larger group of land users who will be asked to provide the HLS team feedback. In the next month, there will be a public provisional release of forward-processed HLS data. While forward processing of the provisional data is expected to begin in October, historical processing of the remaining data archive from 2013–2020 will take place over the next year and is expected to be completed in late 2021.
The value of the HLS project extends well beyond the science community as HLS project lead Dr. Brian Freitag explains:
I can see HLS data being particularly useful for the insurance community after natural disasters. While there are limitations with using optical imagery (i.e. clouds) to analyze the land surface, after a natural disaster, insurance companies could use HLS data to develop an initial area affected by the disaster to expedite the claims process. For example, the derecho in Iowa on August 10 had hurricane-force winds embedded in a line of severe thunderstorms that caused significant crop damage to cornfields. Claims submitted by farmers in regions where crop damage can be observed in the HLS imagery can be quickly reviewed by the insurance company and approved. This reduces the burden on the insurance company, but also gets the affected person the financial help they need quicker — a true win-win.
More information on the HLS project can be found here.