By Linda See and Ian McCallum, IIASA Ecosystems Services and Management Program, Earth Observation Team
Land cover is of fundamental importance for environmental research. It serves as critical baseline information for many large-scale models, for example in developing future scenarios of land use and climate change. However, current land cover products are not accurate enough for many applications and to improve them we need better and more accessible validation data. We recently argued this point in a Nature correspondence, and here we take the opportunity to expand on our brief letter.
In the last decade, multiple global land cover data products have been developed. But when these products are compared, there are significant amounts of spatial disagreement across land cover types. Where one map shows cropland, another might show forest domains. These discrepancies persist even when you take differences in the legend definitions into account. The reasons for this disagreement include the use of different satellite sensors, different classification methodologies, and the lack of sufficient data from the ground, which are needed to train, calibrate, and validate land cover maps.
An artist’s illustration of the NASA Landsat Data Continuity Mission spacecraft, one of the many satellites that collects data about Earth’s surface. Credit: NASA/GSFC/Landsat
A recent Comment in Nature (Nature, 513, 30-31; 2014) argued that freely available satellite imagery will improve science and environmental-monitoring products. Although we fully agree that greater open access and sharing of satellite imagery is urgently needed, we believe that this plea neglects a crucial component of land cover generation: the data required to calibrate and validate these products.
At present, remotely sensed global land cover is not accurate enough for monitoring biodiversity loss and ecosystem dynamics or for many of the other applications for which baseline land cover and change over time are critical inputs. When Sentinel-2–a new Earth observation satellite to be launched in 2015 by the European Space Agency–comes online, it will be possible to produce land cover maps at a resolution of 10 meters. Although this has incredible potential for society as a whole, these products will only be useful if they represent the land cover more accurately than the current products available. To improve accuracy, more calibration and validation data are required. Although more investment is clearly needed in ground-based measurements, there are other, complementary solutions to this problem.
Map showing cropland disagreement between two different land cover maps, GlobCover and GLC2000: all colors represent disagreement. Credit: Geo-Wiki.org, Google Earth
Not only should governments and research institutes be urged to share imagery, they should also share their calibration and validation data. Some efforts have been made by the Global Observation for Forest Cover and Land Dynamics (GOFC-GOLD) in this direction, but there is an incredible amount of data that remains locked within institutes and agencies. The atmospheric community shares their data much more readily than the Earth Observation (EO) community, even though we would only benefit by doing so.
Crowdsourcing of calibration and validation data also has real potential for vastly increasing the amount of data available to improve classification algorithms and the accuracy of land cover products. The IIASA Geo-Wiki project is one example of a growing community of crowdsourcing applications that aim to improve the mapping of the Earth’s surface.
New apps developed by IIASA’s Earth Observation Team aim to involve people around the world in on-the-ground data validation efforts.
Geo-Wiki is a platform which provides citizens with the means to engage in environmental monitoring of the earth by providing feedback on existing spatial information overlaid on satellite imagery or by contributing entirely new data. Data can be input via the traditional desktop platform or mobile devices, with campaigns and games used to incentivize input. Resulting data are available without restriction.
Another major research projects we are using to address many of these issues identified above is the ERC Project Crowdland .
Note: This article gives the views of the authors, and not the position of the Nexus blog, nor of the International Institute for Applied Systems Analysis.
Steffen Fritz has just been awarded an ERC Consolidator Grant to fund a research project on crowdsourcing and ground data collection on land-use and land cover. In this interview he talks about his plans for the new project, CrowdLand.
Farmers in Kenya are one group which the Crowdland Project aims to involve in their data gathering. Photo credit: Neil Palmer, CIAT
What’s the problem with current land cover data?
There are discrepancies between current land cover products, especially in cropland data. It’s all based on satellite data, and in these data, it is extremely difficult to distinguish between cropland and natural vegetation in certain parts of the world if you do not use so-called very high resolution imagery, similar to a picture you take from space. With this high-resolution data you can see structures like fields and so on, which you can then use to distinguish between natural vegetation and cropland. But this is a task where currently people are still better at than computers–and there is a huge amount of data to look at.
In our Geo-Wiki project and related efforts such as the Cropland Capture game, we have asked volunteers to look at these high-resolution images and classify the ground cover as cropland or not cropland. The efforts have been quite successful, but our new project will take this even further.
How will the new project expand on what you’ve already done in Geo-Wiki?
The big addition is to go on the ground. Most of the exercises we currently do are based on the desktop or the phones, or tablets, asking volunteers to classify imagery that they see on a screen.
What this project aims to do is to improve data you collect on the ground, known as in-situ data. You can use photography, GPS sensors, but also your knowledge you have about what you see. We will use volunteers to collect basic land cover data such as tree cover, cropland, and wetlands, but also much more detailed land-use information. With this type of data we can document what crops are grown where, whether they are irrigated, if the fields are fertilized, what exact type of crops are growing, and other crop management information which you cannot see in satellite imagery. And there are some things you can’t even see when you’re on the ground, thus you need to ask the farmer or recruit the farmer as a data provider. That’s an additional element this project will bring, that we will work closely with farmers and people on the ground.
For the study, you have chosen Austria and Kenya. Why these two countries?
In Austria we have much better in situ data. For example, the Land Use Change Analysis System (LUCAS) in Europe collects in situ data according to a consistent protocol. But this program is very expensive, and the agency that runs it, Eurostat, is discussing how to reduce costs. Additionally the survey is only repeated every three years so fast changes are not immediately recorded. Some countries are not in favor of LUCAS and they prefer to undertake their own surveys. Then however you lose the overall consistency and there is no Europe-wide harmonized database which allows for comparison between countries. Our plan is to use gaming, social incentives, and also small financial incentives to conduct a crowdsourced LUCAS survey. Then we will examine what results you get when you pay volunteers or trained volunteers compared to the data collected by experts.
In Kenya, the idea is similar, but in general in the developing world we have very limited information, and the resources are not there for major surveys like in Europe. In order to remedy that the idea is again to use crowdsourcing and use a “bounded crowd” which means people who have a certain level of expertise, and know about land cover and land use, for example people with a surveyor background, university students, or interested citizens who can be trained. But in developing countries in particular it’s important to use financial incentives. Financial incentives, even small ones, could probably help to collect much larger amounts of data. Kenya is a good choice also because it has quite a good internet connection, a 3G network, and a lot of new technologies evolving around mobile phones and smartphone technology.
What will happen with the data you collect during this project?
First, we will analyze the data in terms of quality. One of our research questions is how good are the data collected by volunteers compared to data collected by experts. Another research question is how can imperfect but large data collected by volunteers be filtered and combined so that it becomes useful and fulfills the scientific accuracy requirements.
Then we will use these data and integrate them into currently existing land use and land cover data, and find ways to make better use of it. For example, in order to make projections about future land-use and to better quantify current yield gaps it is crucial to get accurate current information on land-use, including spatially explicit information on crop types, crop management information and other data.
Once we have done some quality checks we will also make these data available for other researchers or interested groups of people.
Crowdsourcing for land cover is in its infancy. There have been lots of crowdsourcing projects in astronomy, archaeology, and biology, for example, but there hasn’t been much on land use, and there is huge potential there. ”We need to not only better understand the quality of the data we collect, but also expand the network of institutions who are working on this topic.”
Note: This article gives the views of the interviewee, and not the position of the Nexus blog, nor of the International Institute for Applied Systems Analysis.
By Linda See, Research Scholar, IIASA Ecosystems Services and Management Program
Humans have a long history of map-making that can be traced back to cave paintings older than 20,000 years, and detailed maps made by the ancient Romans, Greeks, and Chinese. These maps tell the story of exploration and changing borders of states, countries, and populations.
This screenshot shows our Geo-Wiki tool for collecting data from the crowd.
Until recently, military and government mapping agencies have been entirely in control of mapping, but this is changing. The rise of neogeography and user-generated geo-referenced content online has led to a new generation of community-based maps such as OpenStreetMap. Enabled by interactive web technology (Web 2.0) and the GPS in mobile phones, people are now mapping different aspects of the Earth’s surface through crowdsourcing. This new model has proved its worth in cases like the post-disaster recovery, e.g. the devastating earthquake in Haiti.
The trouble with maps
Even in this age of satellites and space technology, it is far from easy to generate good automated representations of the Earth’s surface. While satellite imagery has allowed us to create global maps of land cover—the various materials such as grass, trees, water, and cities that cover the Earth’s surface—at various resolutions from 10 km to 30 m, there are two main problems with all the different products that are now currently available. The first is that these products have accuracies that are only between 65 to 75%. Secondly, when they are compared with one another, there are large spatial disagreements between them. If you are a user of these products, which one should you choose? How can you trust any one of these products when they have uncertainties as large as 25 to 35%? And more importantly, without good baseline information about the Earth’s land cover, such as the amount of forest or cropland, how can we possibly predict what will happen in the future?
The Geo-Wiki Project
Our Geo-Wiki project aims to solve this problem through crowdsourcing. With open access to satellite imagery through Google Earth and Bing Maps, citizens and interested experts can help us better characterize land cover, to correct existing land cover maps or build new ones. Geo-Wiki is a simple set of tools to sample the Earth’s surface, which allows a network of Geo-Wiki volunteers to tell us what type of land cover is visible from Google Earth or Bing Maps.
This map of cropland in Ethiopia was created from crowd-sourced data.
One example of our crowd-sourcing campaigns was focused on mapping cropland in Ethiopia. Over a three week period, we collected more than 80,000 samples across the country, roughly 5% of the area of Ethiopia. Using simple interpolation, we have demonstrated that a cropland map of Ethiopia, a key type of land cover, can be created very easily, with just a small crowd of volunteers. We validated the map using an official validation data set from the GOFC/GOLD reference portal as well as other crowdsourced data collected through Geo-Wiki. The results of this study showed that the map is considerably more accurate than global land cover maps for Ethiopia when considering only cropland. You can find more details about this research at:
See, L. McCallum, I., Fritz, S., Perger, C., Kraxner, F., Obersteiner, M., Deka Baruah, U., Mili, N. and Ram Kalita, N. 2013. Mapping Cropland in Ethiopia using Crowdsourcing. International Journal of Geosciences, 4(6A1), 6-13 http://dx.doi.org/10.4236/ijg.2013.46A1002.
The Ethiopian example is just the tip of the mapping iceberg. As more citizens get involved in mapping land cover online—for example with our Geo-Wiki Pictures app, we could revolutionize land cover mapping in the future.
If you are interested in helping us improve land cover, register at http://www.geo-wiki.org or find us on Facebook to join our crowdsourcing network and learn more about upcoming crowdsourcing campaigns.
Note: This article gives the views of the author, and not the position of the Nexus blog, nor of the International Institute for Applied Systems Analysis.