Coming from Reston, VA, to the rural landscape surrounding the Earth Resources Observation and Science (EROS) Center near Sioux Falls, SD, has brought about some challenging transitions for Peter Doucette, the new Integrated Science and Applications Branch Chief at EROS.
Geography isn’t one of them, says Doucette, who grew up in small-town Maine. But in moving on from the work he enjoyed for five years as deputy program coordinator for the National Land Imaging (NLI) Program, well, “the decision to leave Reston,” he said, “was not easy.”
That said, Doucette had hoped someday to immerse himself even more in the world of Earth-observing science. Now EROS is giving him that chance.
Perhaps the most challenging part of the move, he said, came with getting buy-in from his wife and children. While that took some doing, “I think everyone has made the mental adjustment for now,” he said. “This was our collective ‘carpe diem’ moment.”
Doucette talked about his aspirations and visions as the EROS Integrated Science and Applications Branch (ISAB) chief in this conversation.
As the first EROS Science branch leader to come from outside the EROS ranks since the 1970s, do you consider that an advantage or a challenge?
“Since the ‘70s? Wow, now I feel special. Coming from the outside comes with pros and cons, and as with many things in life, it comes down to tradeoffs. But, of course, it’s not as though I’m a complete unknown at EROS. Since the bulk of NLI funding goes to EROS, I’ve been getting to know folks across the Center over the past five years from my time with NLI. Not to mention the four-month detail I did at EROS in the summer of 2018. But to your question, I think bringing in perspectives from the outside can be a good thing from time to time, especially in times of change. So, I’ll certainly try to leverage that angle of it and bring my experiences as a fed and a contractor scientist across a few different agencies over the last 30 years. But at the end of the day, the main thing that will matter to people is if changes, or new directions taken, are to the betterment of the organization. To put it another way, there’s a big difference between intentions and outcomes, and only time will tell that. I always look forward to taking on that kind of challenge.”
You certainly must have residual duties that will follow you from your position as NLI deputy program coordinator. What are your plans for transitioning from Reston to Sioux Falls?
“Of course, I plan to maintain connections with NLI people, and hopefully can help to mediate and translate needs in both directions. I’ll plan to stay involved in activities such as the LAG (Landsat Advisory Group), NCAC (National Civil Applications Center), and discussions on SLI (Sustainable Land Imaging), USGEO (U.S. Group on Earth Observations), data science, AI/ML (artificial intelligence/machine learning) and cloud services, and such. Needless to say, I’ll be in a good position to collaborate on Science vision and objectives with NLI and CSS (Core Science Systems). Again, I hope to make the most of five years of building relationships in Reston and across the Bureau.”
From your past experience in an acting capacity at EROS, and your longer role in the USGS and EROS through your NLI Program responsibilities, can you describe your thoughts on how EROS Science can be better used to support the USGS and EROS missions?
“I’m not sure just how ‘clear’ my understanding is at this point, but I do believe that the EarthMAP vision is a good platform to promote the value of EROS Science. For instance, we’ve already demonstrated foundational elements of EarthMAP concepts, such as landscape monitoring and change science through the MLRC (Multi-Resolution Land Characteristics) consortium, LCMAP (Land Change Monitoring, Assessment, and Projection), and other science projects. And we’ve been showing nice progress toward developing a landscape ‘projection’ piece through the FORE-SCE (FOREcasting SCEnarios of Land-use Change) work. I don’t think it’s a stretch to say that EROS has some of the most mature science ready for incorporation into EarthMAP development activity. But, of course, tying the EROS Science strategy to EarthMAP means hedging our bet on its ability to survive changing administrations over time, and I believe it can. Perhaps a bigger challenge is the cultural change needed across the Bureau. The EarthMAP team is aware of these dynamics and is planning accordingly. Throwing in the big changes afoot in computing technology, and the fact that EarthMAP thinking is in line with the natural progression of EROS Science in any event, makes me believe that it’s a good vision for us to embrace. And let’s not forget that the EarthMAP concept originated from the COSSA (Council of Senior Science Advisors) report on Grand challenges for integrated USGS science, which puts us in pretty good company.”
Based on your understanding of the state of the EROS Science program, what do you see as the key opportunities for expanding our science impacts in the coming year or two?
“One of those key opportunities is certainly with the release of LCMAP version 1 products, and how that supports the larger vision for land change science going forward. For that, we’ll need to shift more attention to the “A” part of LCMAP—assessment—so we can better explain the cause of change after detecting it. Also, EarthMAP discussions have shed new light on the value of our fire science and surface water monitoring research and products. Another key opportunity is taking advantage of new computing capabilities from HPC and cloud. That’s not a trivial shift because it comes with a steep learning curve, but we’ve got to embrace it, and move forward in synch with EROS’ DCS. As we’re hearing more about another buzz topic, AI/ML, which in reality is a coded way to refer to the application of neural network-based learning methods, we need to better understand how we could make use of it. Funny thing about neural nets is that the basic math model hasn’t changed all that much since I was first introduced to them in the late ‘90s. What has changed is the dramatic increase in training data available, and of course the increase in computing power needed to train the much larger neural net models, which has come to be known as ‘deep learning.’ Together, those advances are what pushed neural nets past traditional machine learning methods in terms of performance accuracy for many data classification problems. The challenge will be cutting through a lot of the hype in getting to the bottom of where the real value is, and for which science projects.”
So, having already discussed the USGS EarthMAP concept, what role would you like to see EROS play as EarthMAP is developed?
“First of all, I’d like to recognize Terry Sohl as our PMT rep to EarthMAP, and to say that we have a high degree of confidence with Terry’s ability to represent EROS Science toward EarthMAP objectives. When it comes to understanding landscape change monitoring on a national scale, we need to make sure it’s understood why EROS is the go-to organization, and that it’s due in large part to the Landsat program, end-to-end, which makes that the case. Many of our science projects can support key issues that EarthMAP is endeavoring to address, such as ecological impacts from land fire, flooding, and drought. We have many operational monitoring products for use today, such as NLCD (National Land Cover Database), shrubland, impervious surface, tree canopy, LANDFIRE, dynamic surface water extent, burned area extent, and fractional snow cover. Plus, we have LCMAP version 1 land cover and change products coming online in the spring of 2020. As many have recognized, we need to monitor past change in order to understand it before we can build models from which to forecast future change, which is the ‘actionable intelligence’ the Director (Jim Reilly) speaks of. To that end, Terry’s own FORE-SCE research project has shown tremendous potential for a scenario-based approach to forecasting landscape change. So, given our maturity level of landscape monitoring science and products, together with our R&D (research and development) efforts in forecasting change, we have a very credible voice at the EarthMAP development table. We should use that voice.”
Have you established any immediate goals or actions that you intend to pursue as you take over the reins of science?
“Sure, a fundamental issue is finding ways to better integrate activity across the branch. We need to leverage commonalities in how we do analysis across varied application areas, such as land cover, fire, coastal, water, vegetation, climate, and food security. One way to motivate that is through a coordinated analysis and development environment. For example, one of the more common analysis activities among ISAB projects is tuning computer algorithms to derive thematic information from remotely sensed images. That is, these algorithms label individual pixels into different categories of land cover or condition, based upon the spectral signal measured by observing platforms such as Landsat. A big part of the analysis process is investigating ways to improve the accuracy and speed of these pixel-labeling methods. Now, with the advent of HPC and cloud computing, there may be new ways to approach these methods that can dramatically improve performance. It’s the ways that we’ve yet to even think of that gets me energized about the possibilities.”
Give us some ideas on how to do that integration.
“Establishing some form of integrated analysis and development environment, with cutting edge computing tools, can help us to optimize common activities. As a result, we’ll have more time to spend on more complex science problems, like understanding causes of land change. We need an environment that provides us flexibility within rapidly evolving technology cycles, and agility to continue flying the plane while building it out. We need an analysis environment that can stimulate imaginative thinking. And finally, we’ll need to pursue this strategy all while striking a balance between science and engineering needs across the branch. That is, science often looks to answer ‘why,’ and engineering ‘how.’ While these are highly interconnected in practice, they involve different ways of thinking. Science tends to be more exploratory and iterative by nature, with more uncertainties, compared to engineering. To sum up, none of these lofty sounding ideas will go anywhere unless we put people first. So, I do hope that the Science staff is excited about our prospects.”
When you were here as acting deputy director, you said you thought there was a cultural divide between the rapidly evolving IT community and traditional domain scientists, where each side has a unique philosophy and uses different jargon. Do you intend to enable a dialogue between the two on that?
“From what I’ve seen, this perception applies across USGS and well beyond. I think a good start is to set up a strong working relationship between ISAB and DCS people. I also expect many more training opportunities to come online, but folks will also need to step up to the challenge of embracing the new tech. And perhaps most critically, we need to be recruiting a good mix of ‘data science’ and domain science skills through any means available to us, be it Federal, post-doc, contracting—on and off-site. The field of data science has been tremendously hyped over the past several years, seeing different sectors grapple with defining it. Some have argued that it boils down to doing high-powered statistical analysis, and in some ways, I think that’s a fair characterization. To me, the most defining quality of data science is having the ability to effectively apply the latest tech from HPC, database, visualization, cloud and AI/ML, to data-driven solutions for science problems.”
Again, give us an example of what you’re talking about.
“A good example of this is the numerical simulation methods used to better understand the effects of data uncertainty that propagate through non-linear physical models. By the way, numerical simulation is a fair analogy to understand how deep learning works. And it’s finding ways to creatively communicate these kinds of analogies across the sciences that can go a long way to bridging the jargon and culture gap.”
How do you hope to see the use of machine learning and artificial intelligence algorithms ingrained in the science we do at EROS?
“AI/ML is one of the fastest growing services in the cloud. I believe this is being driven in part by the connection between big training data sets and compute in getting the highest performance out of deep learning methods. As the pace of innovation of AI/ML technology in the cloud continues at a phenomenal clip, it would make a good deal of sense to explore how to take advantage of the Bureau’s CHS (Cloud Hosting Solutions) vehicle, looking at costs versus capability. The key activity for us is to understand where the performance gains are to be had. At the same time, we need to start making better use of the AI/ML ‘Tallgrass’ machine at EROS. Tallgrass is optimized to support deep learning software packages like TensorFlow and PyTorch, which we’ll need to get smarter with using in any environment. I expect there’ll be good justifications for cloud and HPC paths. One of the advantages of doing science in general in the cloud is the potential to facilitate interagency collaboration in terms of people and data sets. That is a fundamental objective of new analysis environments coming on the scene, such as PANGEO and Open Data Cube. And because these are open source, they can operate on any cloud or on-prem platform. In late 2019, we (NLI, EROS, and CHS folks) began discussing these kinds of joint development opportunities with NASA and NOAA, which are continuing. That’s a pretty exciting prospect for EROS.”
Is there anything else you want to add that I didn’t ask you about this new role you’ve taken?
“Yes, a few words on the Landsat Science Team (LST). Over the years, the LST had made several recommendations to EROS Science that have contributed significantly to developments, such as Landsat data distribution over the internet, standardized atmospheric correction, and cloud detection algorithms, Analysis Ready Data, CCDC, LCMAP, and most recently, an Aquatic Reflectance product. As we move to broaden our integration goals, we need to consider ways to include the LST to the greatest extent possible in that process, given its solid track record of inspired ideas. My plan is to actively pursue these kinds of opportunities where available. And one last item for longer-term consideration—the collection and availability of hyperspectral imagery (HSI) and synthetic aperture radar (SAR) may ramp up considerably this coming decade, and we need to expand our science capabilities to assess their utility toward many of our projects. Utilizing different collection modalities is also consistent to doing ‘integrated science.’ ”