Eyes on Earth Episode 33 – Global Land Change

Download
Right-click and save to download

Detailed Description

Mapping land cover across the United States using Landsat satellite data is a difficult, time-intensive job, but there are jobs far larger. Matt Hansen, a Maryland-based professor and member of the Landsat Science Team focuses his efforts on mapping land cover and change on a global scale. In this episode of Eyes on Earth, we hear from Matt on how he and his team use the Landsat archive to map change at such a wide scale.
 

Details

Episode Number: 33

Date Taken:

Length: 00:15:48

Location Taken: Sioux Falls, SD, US

Credits

Guest: Matt Hansen, Department of Geographical Sciences at the University of Maryland, College Park

Host: Steve Young
 

Transcript

STEVE YOUNG
Hello, everyone. Welcome to another episode of Eyes on Earth. This is a podcast that focuses on our ever-changing planet and on the people here at EROS and across the globe who use remote sensing to monitor and study the health of Earth. I am your host Steve Young. Today's guest is Matt Hansen, a professor in the Department of Geographical Sciences at the University of Maryland in College Park. Matt is a remote sensing scientist whose research specializes in large area landcover and land use change mapping. His research involves developing improved algorithms, data inputs and thematic outputs which enable the mapping of landcover change at regional, continental and global scales. Matt's work contributes to the management of natural resources, looking at such things as deforestation and bio-diversity monitoring. On top of that, Matt's exhaustive mining of the global Landsat archive has led to a greater understanding of global landcover change, global forest gains and losses, annual agricultural monitoring and many other issues. Welcome, Matt. 
MATT HANSEN:
Thanks, Steve. Appreciate it, glad to be here.
YOUNG:
Let's start at the beginning. What made you pursue a career as a remote sensing researcher and an expert in the geographic aspects of global land change? 
HANSEN: 
I was an engineer undergrad. I did Peace Corps. I was a fish farmer, fish aquaculture extension agent in Zaire. I came back from that and I had a blank slate. The geographers hate when I saw say that I love maps, that is why I am a geographer. But that is basically the idea. Spatial representations of our world are really attractive to me and they make a lot of sense, and they answer a lot of questions. Obviously, if we look at them over time we can see where we are going. I went to the local university and talked about getting a masters in geography, and in that I took one remote sensing course, and with that course, I was able to start my career getting a job at the University of Maryland after I graduated. 
YOUNG:
Your early research involved mapping studies using data from daily global polar orbiting satellites such as AVHRR and MODIS. You gradually moved to large area studies using Landsat and other multi-resolution imagery. What sparked this shift? 
HANSEN:
AVHRR and MODIS, with the daily acquisitions and consistent records, you could extrapolate algorithms very easily, or comparatively easily across the globe. That was like a work bench with these big blurry pixels and most of those products because they were big and blurry were mainly for big climate models. The reason we didn't work with Landsat was data limitations, data costs, compute. But, NASA Pathfinder program had a tropical deforestation project. It had a node at University of Maryland. I wasn't working on it. But, out of the Pathfinder program, there were these great results from Landsat. In the Amazon, you have a dry season where you get a single good look every year and you can map deforestation. But you can't do that in the Congo. The scale change is very small, cloud cover is persistent. We had to start thinking about how we process AVHRR and MODIS data and apply those methods to Landsat. Because you have to start compositing. You have to look at each pixel and say, "Is that a cloud? Is that a shadow? Is that haze? Throw it away. Look for another one." So it becomes this more data intensive approach, as opposed to single images. In the early 2000s, we begged and borrowed and got several hundred, 300, 400 images over the Congo Basin. We started processing these data per pixel using the methods we had developed with AVHRR and MODIS. And, voila, we could map all of the forest loss across the country like Democratic Republic of Congo, the second-largest rain forest country in the world. The shift really was about a particular context, the tropical rain forests of the world and how important they are, them being the site of huge changes in land use. In that context, we got into the per-pixel Landsat monitoring. We published our first large area map at the time of the opening of the Landsat archive in 2008. We had just published our first results. We were ready, absolutely, to go with Landsat at scale based on that experience.
YOUNG:
In your investigations of global land change, are there any surprises that have changed the way you think about the state of the Earth?
HANSEN:
It seems like our appropriation of land, either to convert it from a natural land cover to a land use, or to intensify existing land uses, appears inexorable. We are really efficient at squeezing out economic productivity out of the land. I guess that is sort of surprising, maybe not. But it is remarkable. One of the biggest challenges is, "How much of that can the planet sustain without feedbacks like climate change from the emissions of tropical forests burning, or water quality degradation due to intensification of agricultural landscapes and the like?" I think that is the biggest surprise is how industrious we are, I guess in terms of taking economic advantage of land.
YOUNG:
I am told you have contributed many innovations that are now common in the land remote sensing community, including continuous vegetation fields, global scale applications, multi-resolution monitoring, cloud computing and the blending of sampling and wall to wall mapping to quantify land surface characteristics. Of these many "firsts", which one or ones are you most proud of?
HANSEN:
I really like working at the global scale. There is no global government. The United Nations is the only agency or bureaucracy responsible for global information. But it is a member organization so each country submits their data separately and then the UN has to shoehorn them. The great thing about the satellite is that it is taking the same picture with the same calibration around the globe so you can really track global resources in a very consistent way. If you are a civil servant, you have a set geography that you are responsible for. I think global scale is really something that we have gotten good at processing. We processed the entire Landsat archive in our lab. We are up to over 20 Petabytes of storage. It is high access storage. We have built this for a purpose. I am really proud of the infrastructure and our capability to map the globe. And, more recently this idea of sampling, we were very happy to make maps at the very beginning. Now, with our sampling, which is more into the idea that we use the maps as a targeting mechanism to actually estimate areas with statistical uncertainties. Wow, Now we are cooking. Now we've got definitive information. I really do like that, as well.  
YOUNG:
You are well known for the marathon road trips you make annually that support your various monitoring studies. Why do you put so much emphasis on field observation? How do these trips improve the quality of your research?
HANSEN:
Well, there is no question that if you do remote sensing and all you do is sit at a desk, you're missing something. I am more and more convinced that the map is a point of departure. You have to go to the field for certain things. When I talk about it being a targeting mechanism, if we are going to do soybeans across the United States or across Brazil, you are going to have cultivars that are confused with it. You cannot resolve that at the desk. When you have dry beans, sometimes peanuts, sometimes cotton, that with all the different planting dates and inter-annual variability, you cannot nail soybeans sitting at your desk. You can do a very good job of targeting it. But when you go to the field, you can with your eyes, say soy, soy, soy. And idea of the map as a targeting mechanism, means we can do this very efficiently. We can go to the field and in two weeks have a national estimate. I have a post-doc who did winter wheat in Punjab, Pakistan by himself in two weeks. Well, he had a partner, sorry, so two guys. Punjab is the breadbasket of Pakistan. They get an estimate for which they had a couple thousand enumerators at village level doing it officially. But they do it by themselves because they know where to look. The satellite tells them where to go. So, there is a couple of things. One is it is great feedback to help you understand what you are looking at in the imagery. But it is also the deal in terms of really targeting your theme of interest and getting precise estimates. I would even go further and say that the satellite will tell us where intact habitats are. The satellite tells us where the tall forests are. We cannot easily kludge together existing networks of field data, that are a lot of times not placed statistically in landscapes but they are quite biased. The satellite can construct it with which to target a whole host of  critical variables that we can then visit in situ. So, it's a big deal, the field work. There are some cases where you don't need it, but a lot, a lot, a lot where you do. 
YOUNG: 
What can you tell us about what you are working on right now?
HANSEN: 
We always try to push the envelope with new data, trying to go down in spatial scale so you can say Landsat 30 meters, Sentinel 10-20 meters, Planet 3 meters. It's all really the same processing on our side. We are doing these optical time series so we can get better and better. I think the biggest deal is really the effect that our work is having on policy and management of natural resources. I find it increasingly frustrating that we have ongoing emphasis on technology but, Brazil effected a policy that slowed their deforestation from almost 30,000 sq. kilometers per year down to below 5,000 sq. kilometers with a single Landsat, Landsat 5. Landsat 5 was the instrument with which they quantified the huge rate of deforestation. When they put in their policy to balance development with protection of ecosystem services, Landsat 5 was the evidence for their success. Aside from what I am working on now and my personal curiosities, I am a little discouraged that our data are not really being used to more regularly inform and support policy like the one instance in Brazil. No other country has done what Brazil has done. And, Brazil itself, has been back sliding since that success. My whole focus now is really how relevant can we be? We can be a closed shop and pat each other on the back at the really cool things we are doing, but it is more about, "What is the material impact on society?"
YOUNG:
Matt, take out your crystal ball and share some of your ideas on future trends in global land remote sensing.
HANSEN:
Global land remote sensing ... man, if I were in charge it would really be at the top of a lot of important themes. I would try to do a more top-down approach. I alluded to this before. I think land change is the big one. And, understanding the impacts of land change on environmental systems and human systems is the biggest deal. One of the interesting things for the future is, "To what degree can we put ourselves out of business?" If you are in research, you want to do research operations. Operations means it just runs and we are solving problems and we are supporting decisions. That means we have solved some problems methodologically. A really important trend that I would like to see is that we solve some problems and we move out of R&D for some basic monitoring tasks. The other future trend which I think is up in the air right now is the idea of whether or not we need people with geographic backgrounds or we just need computer programmers. This is a point of contention with me where a machine learning computer programmer can outsource calibration of a model and run map making processes without being geographically knowledgeable and understanding whether or not the map that's expressed in front of him or her makes sense.  When you talk about deep learning and who's in charge of algorithms and who's running algorithms and producing information products...is it going to be people with no geographic background at all? Or, is a geographer going to be in charge? I think geographers are still needed. There is an iterative, active learning process to mapping that I think is still required. You have to have geographers to trouble shoot the data inputs and also to critique the outputs and iterate. That is the active learning side of it. In the future trends, I am very curious which path it goes down. I feel like domain experts and geography, knowing environments and knowing drivers and putting them into landscapes and understanding that is critical to making maps. But, conversely, maybe there are algorithms smart enough that with training that is outsourced you can do the same thing at the same level of accuracy.
YOUNG:
I know you are a member of USGS NASA Landsat Science Team. What do you consider to be the priorities for improving the value of the nearly 50-year-old Landsat archive?
HANSEN:
MODIS to me was a fantastic model because MODIS built all of these grid, different composite time series that were ready for the shelf for users. I think Landsat science team is working towards that, this idea of analysis ready data modeled I think really on the MODIS approach. One of the tricks with Landsat that we face when we are doing our long-time series studies is the tension between continuity and improving the observations themselves in terms of quality. Because the Landsat instruments themselves are slightly different. The strength of Landsat is the time series. It is our baseline. This is what we will use as a reference to compare what's going to happen now and into the future. I am a strong advocate for continuity. That the observations don't get so far off of a standard that you can't compare them. That being said, it's hard to resist improving your technology, right? I think this is a challenge, "What do we do to maintain the ability to integrate the entire time series so that we can track the dynamics?" For me, Landsat is a piece of global input infrastructure. There are very few things, you can talk about GPS and a few other systems, like our Navy keeps the ocean trading lanes free and open. There are examples of global public goods that individual countries pay for and everybody benefits. Landsat is one of those. I think that is such a great gift and we need to maintain that. It is great to have a bunch of people looking at the same data and working on the same problems at different scales, different time periods, maybe different themes. But this is how we come to consensus and we solve problems.
YOUNG:
We've been talking to Matt Hansen, a professor in the Department of Geographical Sciences at the University of Maryland and remote sensing scientist whose research has given the world a much better understanding of large area landcover and land use change mapping. Thanks for joining us, Matt. 
HANSEN:
Thank you, Steve. Appreciate it.
YOUNG:
We hope you come back for the next episode of Eyes on Earth. This podcast is a product of the US Geological Survey Department of the Interior.