Jim Vogelmann discovered something interesting about Analysis Ready Data (ARD) these last few months as he and his colleagues worked on finding a better way to quantify fire risk in the Great Basin of the western United States.
ARD, Vogelmann learned, works really well.
After some initial uncertainty about merging bits and pieces of data from an ARD system that uses tiles instead of path/rows, “all of a sudden, it sort of clicked,” said Vogelmann, a scientist at the Earth Resources Observation and Science (EROS) Center near Sioux Falls, SD. “It was sort of like, these are really easy to use. And not only were they easy to use, but they were ready to go as well. I really ended up liking the ARD.”
Vogelmann’s testimonial is exactly what EROS was hoping for when it reconditioned the 46-year-old Landsat archive to produce the best imagery available processed to a common tiling scheme. With Top of Atmosphere and surface reflectance already figured in, ARD promises to take much of the pre-processing out of time-series work.
Using ARD to Measure Vegetation Health in the Great Basin
Here’s how it helped in measuring vegetation health in the Great Basin. Vogelmann and his colleagues received funding from NASA’s Wildland Fires Application Program to first improve on the mapping of shrub and grasslands in the Great Basin, and then to transfer that into operational fire-potential assessments. They started out by looking at Monitoring Trends in Burn Severity (MTBS) data to identify where and when blazes had occurred in the Great Basin. They then used historical remote-sensing images to not only assess vegetation conditions leading up to those fires, but also to come up with monthly averages for those conditions during the growing season based on 18 years of data.
The latter will enable the prediction of potential wildfires based in part on where a current growing season falls in relation to the average.
In fusing Landsat’s higher spatial resolution with the temporal frequency of NASA’s Moderate Resolution Imaging Spectrometer (MODIS) sensor, Vogelmann and his colleagues created datasets of average vegetation greenness each May going back to 2000. Then they did the same for June, July, and August. If there were gaps in available imagery in the historical record, they used an interpolation algorithm called STARFM to fill in the holes.
STARFM takes the Landsat and MODIS images for each place and time period, and analyzes the relationships between the images and the time periods, Vogelmann said. As part of that interpolation process, the algorithm is able to create an image product for a particular time period if, for example, a Landsat image for that time is unavailable.
Vogelmann said the STARFM product works better than not having any image. “You look at STARFM images closely, and it’s sort of like, ‘Holy smokes, this is pretty good,’ ” he said. “But most users would rather have the real data, even if it’s imperfect real data.”
ARD is Real Data That Goes Back Much Further in Time
ARD provides the real data. And it can go much farther back than MODIS, which only became available starting in 2000, Vogelmann said.
Being able to compare average May vegetation health through the years to current conditions certainly is an important factor in predicting potential fires later in the season. But so is having data that show how previous years played out during June, July, and August, and what happened to the condition of the vegetation before the wildfires broke out.
“Of course, you need the context of what’s going on during the season as well, otherwise you’re just sort of guessing,” Vogelmann said. “We’re really looking at greenness followed by brownness. But if your greenness is very high early in the season, then I think you can make the case that this could potentially be more of a fire year, assuming that you’ve got the triggers for the fires later in the season.”
While ARD ended up being a good option for their work, Vogelmann said he can see how it could be even more user friendly. For starters, he was pulling bits and pieces of Landsat imagery from different ARD tiles. It would have been nice, he said, if there was a simple program that allowed for merging those bits and pieces into one mosaic or composite with the best information available for the time periods needed.
“Otherwise, you spend a lot of time downloading these individual scenes, looking at them and saying, ‘Oh, that’s got too many clouds; I’ll try another. Or, oh, maybe that will be OK. Well it’s not,’ ” he said. “I think if you just sort of had an automatic process that allowed you to put them all together, I think you could get a lot of what people would want fairly quickly.”
ARD Has Potential to Be Even Better
Software that would have allowed his team to access the greenest pixels during a particular time of year without downloading multiple time stacks of pixels and partial tiles would have been very helpful, Vogelmann said.
“I’m not a coding person, but wouldn’t it have been nice to say, ‘OK, my window is May 1 to May 31, and I want the greenest pixel, and I want to exclude any pixel that is a cloud, shadow, or ice pixel?’ ” he said. “Give me the best you’ve got … boom, and you get the dataset. I don’t think it’s that difficult for those in the know to be able to develop that sort of stuff.”
Having worked with it now, Vogelmann said he can see how ARD would be valuable in a wide variety of research efforts, from monitoring drought or insect defoliation events to the kinds of work he did in assessing fire potential.
Of course, part of the challenge for ARD is convincing people to give up their pet approaches for working with Landsat data, he said. They get accustomed to a particular way of doing things, and old patterns die hard.
“People would rather work with something that they understand intuitively. I get that,” Vogelmann said. “So, I think what needs to happen is simply getting people used to working with ARD. And maybe having some of the tools to make it easier so they don’t have to struggle initially.
“The fact is, once I figured out what I was doing with it, I ended up liking it. I really did. I liked it a lot. I think others will, too.”