New open-source software supports land-cover monitoring

By Victor Maus, IIASA Ecosystems Services and Management Program

Nowadays, satellite images are an abundant supply of data which we can use to get information about our planet and its changes. Satellite images can, for example,  help us detect an approaching storm, measure the expansion of a city, identify deforested areas, or estimate how crop areas change over time. Usually, we are interested in extracting information from large areas, for example, deforestation in the Amazon Rainforest (5.5 million km², around 15 times the area of Germany). It would be challenging for us to monitor and map such vast areas without combining satellite images with automated and semi-automated computer programs.

Aerial view of the Amazon Rainforest, near Manaus, Brazil. Monitoring deforestation in the Amazon is difficult because the area is massive and remote. ©Neil Palmer | CIAT

To address this problem, I developed — along with my colleagues Gilberto Camara from the Brazilian National Institute for Space Research and Marius Appel and Edzer Pebesma from the University of Münster, Germany — a new open source software to extract information about land-cover changes from satellite images. The tool maps different crop types (e.g., soybean, maize, and wheat), forests, and grassland, and can be used to support land-use monitoring and planning.

Our software, called dtwSat, is open-source and can be freely installed and used for academic and commercial purposes. It builds upon on other graphical and statistical open-source extensions of the statistical program R. Adding to that, our article in press in Journal of Statistical Software is completely reproducible and provides a step-by-step example of how to use the tool to produce land-cover maps. Given that we have public access to an extensive amount satellite images, we also get much benefit from tools that are openly available, reproducible, and comparable. These, in particular, can contribute to rapid scientific development.

The software dtwSat is based on a method widely used for speech recognition called Dynamic Time Warping (DTW). Instead of spoken words, we adapted DTW to identify ‘phenological cycles’ of the vegetation. These encompass the plants’ life cycle events, such as how deciduous trees lose their leaves in the fall.  The software compares a set of phenological cycles of the vegetation measured from satellite images (just like a dictionary of spoken words) with all pixels in successive satellite images, taken at different times. After comparing the satellite time series with all phenological cycles in the dictionary, dtwSat builds a sequence of the land-cover maps according to similarity to the phenological cycles.

The series of maps produced by dtwSat allows for land-cover change monitoring and can help answer questions such as how much of the Amazon rainforest has been replaced with soy or grass for cattle grazing during the last decade? It could also help study the effects of policies and international agreements, such Brazil’s Soy Moratorium, where soybean traders agreed not to buy soy from areas deforested after 2006 in the Brazilian Amazon. If soy farming cannot expand over areas deforested after 2006, it might expand to areas formerly used for cattle grazing deforested before 2006, and force the cattle grazing farmers to open new areas that have been cleared more recently. Therefore, besides monitoring changes, the land-cover information can help better understand direct and indirect drivers of deforestation and support new land-use policy.

Further info: dtwSat is distributed under the GPL (≥2) license. The software is available from the IIASA repository PURE pure.iiasa.ac.at/14514/. Precompiled binary available from CRAN at cran.r-project.org/web/packages/dtwSat/index.html

dtwSat development version available from GitHub at github.com/vwmaus/dtwSat

Reference:

Maus V, Camara G, Appel M, & Pebesma E (2017). dtwSat: Time-Weighted Dynamic Time Warping for Satellite Image Time Series Analysis in R. Journal of Statistical Software (In Press).

Maus, V, Camara, G, Cartaxo, R, Sanchez, A, Ramos, FM, & de Queiroz, GR (2016). A Time-Weighted Dynamic Time Warping Method for Land-Use and Land-Cover Mapping. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 9 (8): 3729–39.

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.

Picture Pile: Gaming for Science

By Dilek Fraisl, IIASA Ecosystems Services and Management Program

In October 2015, we launched our latest game, Picture Pile. The idea is simple: look at a pair of satellite images from different  years and tell us if you can see any evidence of deforestation. Thanks to the participation of many volunteers, 2.69 million pictures have already been sorted in our pile of 5 million pairs. But we still have a long way to go, and we need your help to get us there!

PicturePileScreen

Screenshot from the game: click for more information (Image credit Tobias Sturn)

Deforestation is one of the most serious environmental problems in the world today. Forests cover a third of the land area on Earth, producing vital oxygen, habitats for a diversity of wildlife, and important ecosystem services. According to the World Wildlife Fund (WWF), some 46,000 to 58,000 square miles of forest are lost each year, which is equivalent to 48 football fields every minute. But this is a rough estimate since deforestation is very difficult to track. Reasons why are that satellite imagery can be of insufficient spatial resolution to map deforestation accurately, deforestation mostly occurs in small chunks that may not be visible from medium-resolution imagery, and very high-resolution data sets are expensive and can require big data processing capabilities, so can only be used for limited areas.

To help contribute to better mapping of deforestation, researchers in IIASA’s Earth Observation Systems (EOS) group, led by Steffen Fritz, have been working on novel projects to engage citizens in scientific data collection that can complement satellite-based traditional deforestation monitoring. One of the latest applications is Picture Pile, a game that makes use of very high-resolution satellite images spanning the last decade. Designed by Tobias Sturn, the aim is to provide data that can help researchers build a better map of deforestation. Players are provided with a pair of images that span two time periods and are then asked to answer a simple question:  “Do you see tree loss over time?” After examining the image, the player drags the images to the right for “yes,” left for “no,” or down to indicate “maybe” when the deforestation is not clearly visible.

Every image is sorted multiple times by numerous independent players, in order to build confidence in the results, and also to gain an understanding of how good the players are at recognizing visible patterns of deforestation. Once enough data are collected at a single location, the images are taken out of the game and new ones are added, thereby increasing the spatial coverage of our mapped area over time. Right now we are focusing on Tanzania and Indonesia, two regions where we know there are problems with existing maps of deforestation.

Picture Pile is focusing first on Indonesia and Tanzania - two regions where there are problems with existing maps of deforestation. Photo (cc) Aulia Erlangga for Center for International Forestry Research (CIFOR).

Picture Pile is focusing first on Indonesia (pictured) and Tanzania – two regions where there are problems with existing maps of deforestation. Photo (cc) Aulia Erlangga for Center for International Forestry Research (CIFOR).

Once the pile is fully sorted, the 5 million photos in the data set will be used to develop better maps of forest cover and forest loss using hybrid techniques developed by the group as well as inputs to classification algorithms. We will also use the data to validate the accuracy of existing global land cover maps. Finally, we will mine the data set to look for patterns regarding quality (for example, how many samples do we need to provide to the “crowd” before we can be confident enough to use their data in further research). In short, by integrating citizens in scientific research, Picture Pile will also help us improve the science of land cover monitoring through crowdsourcing mechanisms.

So please join in and help us get to the finish line. You can play Picture Pile in your browser or you can download the free iOS/Android app from the Apple and Google Play stores and play on your smartphone or tablet. Your contributions will help scientists like those at IIASA to tackle global problems such as deforestation and environmental degradation. At the same time you may win some great prizes: a brand new smartphone, a tablet, or a mini tablet.

More information:

Reference
Schepaschenko D, See L, Lesiv M, McCallum I, Fritz S, Salk C, Perger C, Schepaschenko M, Shvidenko A, Kovalevskyi S, Albrecht F, Kraxner F, Bun A, Maksyutov S, Sokolov A,  Dürauer M, Obersteiner M. (2015) Global hybrid forest mask: synergy of remote sensing, crowd sourcing and statistics. Remote Sensing of the Environment, 162, 208-220. doi:10.1016/j.rse.2015.02.011

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.

Scientific decision support systems: One step beyond bridging science to policy

By Ping Yowargana, IIASA Ecosystems Services and Management Program

Recently, Indonesia has been combating its most severe forest fire of the decade. Around 43 million Indonesians have been exposed to hazardous fumes, and countless loss of biodiversity and ecosystem services has occurred. An estimated 1 billion tonnes of carbon emission has been released to the atmosphere. Within three months, Indonesia’s daily carbon emission has surpassed the average daily emissions of the whole US economy.

Firefighters outside Palangka Raya, Central Kalimantan, 15 October 2015.  Photo by Aulia Erlangga/ CIFOR

Firefighters outside Palangka Raya, Central Kalimantan, 15 October 2015.
Photo by Aulia Erlangga/ CIFOR

Forest fires in Indonesia are closely related to unsustainable agricultural practices spreading out throughout the country. Indonesia is the world’s largest producer of palm oil, with 8 million hectares of plantation area. Other than its significant contribution to the economy, and broadly debated effects on the environment, oil palm is also regarded as a promising solution to the country’s effort to achieve energy security. The current administration has set ambitious targets to increase national biofuel production, and to consume it domestically.

In this landscape of uncertainty and crisis, scientific support for Indonesian energy policy is more urgently needed than ever. That’s why it is one of our main focuses in the IIASA Tropical Forests Initiative (TFI).

“Scientific authority has to be the basis of our future energy policies,” said Mr. Sudirman Said, Indonesia’s Minister of Energy and Mineral Resources, at the opening session of our first collaborative screening workshop in September in Bandung, Java. In the workshop, jointly organized by IIASA and the ministry, we aimed at laying out a plan to establish a new decision support system for the ministry, based on IIASA’s energy systems models such as the renewable energy systems optimization model, BeWhere and the Model for Energy Supply Strategy Alternatives and their General Environmental Impact (MESSAGE).

Scientific decision support systems (DSS) are a tangible crystallization of bridging science to policy. A decision support system gathers information and analytical expertise in order to improve the quality of policy making, using feedback and evaluation from previous planning and policy implementation. As a practical approach in dealing with what scientists refer as complex adaptive systems, such DSS should be able to integrate visions of long-term planning with technical details that are important for daily executed policies.

The IIASA and the Ministry of Energy and Mineral Resources of Indonesia screening workshop took place from 15-17 September. ©MEMR

The IIASA and the Ministry of Energy and Mineral Resources of Indonesia screening workshop took place from 15-17 September. ©MEMR

Indonesia’s energy sector is a typical example of a highly complex system. Currently, challenges of the sector are more cross-sectoral than ever. Issues that seem to have limited scopes, such as bioenergy, actually influence a broad swath of other areas including agriculture, land use change, air pollution, climate change and social equity.

For that reason, the approach we brought to the recent meeting relies on multiple models. BeWhere brings a snapshot perspective to explore energy supply options that best meet the objective set by policy makers, such as cost efficiency or least CO2 emissions, based on location specific energy demand, resource and infrastructure availability. On the other hand, MESSAGE brings a more macroscopic perspective, looking at various scenarios that project optimal solutions of meeting long-term energy demand in a certain region or country.

To have a truly systems perspective, the above approach cannot stand alone. Before we started looking at Indonesia’s energy sector, we had engaged local researchers in the tropics to localize IIASA’s Gobal Biosphere Management Model (GLOBIOM). GLOBIOM is used to analyze the competition for land use between agriculture, forestry, and bioenergy, which are the main land-based production sectors. Clearly, investigating further into the energy system will allow us to grasp a more holistic understanding and develop solutions to tackle challenges in tropical countries.

As one of IIASA’s pilot countries in the budding TFI, Indonesia represents conflicting realities of the tropics, which are essential to the planet’s well-being. Tropical forests help regulate the Earth’s climate system, while being home to huge biodiversity, millions of plant and animal species. However, the region is also highly challenged by domestic development needs and the growing consequences of a globalized economy. Abundant natural resources and land-intensive agricultural commodities, together with intensified infiltration of global supply chains and complicated socio-economic structures, have resulted in severe ecological pressures that are harmful to the region as well as the planet.

The TFI aspires to address such complexity by applying systems analysis together with regional policymakers. Such application implies a two-fold challenge. The first one is to put together IIASA’s various scientific tools to understand the broader picture that comes out from the integration of interrelated aspects of domestic development. Secondly, working together with policymakers leads to a mutual learning process. Policymakers learn to use scientific models and tools in their decision making process. In this process, fitting the models into the local context is an inevitable step that requires intense communication between scientist and practitioners. Eventually, this process will also benefit researchers by giving them a better understanding of the issue, and opening opportunities for further scientific investigation on new topics.

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.

Mapping the global palm oil boom

By Johannes Pirker and Aline Mosnier, IIASA Ecosystems Services and Management Research Program

In the late 2000s, Palm oil became a major target of environmental NGO’s working to save tropical forests. Bleak images of recently cleared forest sites left no doubt about the harmful impact of the commodity which today is omnipresent in our food—palm oil today is used in many everyday products such as chocolate bars, shampoo, and margarine.  Campaigning against palm oil is a safe bet for NGO’s; it has become the North Korea among the vegetable oils.

The facts are clear: oil palm cultivation has expanded tremendously in recent years. Indonesia and Malaysia were and continue to be the epicenter of this expansion. In these two countries, new plantations have led to at least 3.7 million hectares of forest loss during the last 20 years, an area bigger than Belgium. This deforestation threatens not only animal and plant species, but  expansion of palm oil plantations is increasingly occurring into carbon-rich peat soils, leading  to  the release of tremendous amounts of climate-warming  CO2  into the atmosphere.

Palm oil fruit. Photo Credit: Aline Soterroni

Palm oil mill of CDC in the South-West province in Cameroon, taken by Aline Soterroni, a 2010 IIASA YSSP participant and collaborator on the REDD-PAC project.  Photo Credit: Aline Soterroni

On the other hand, even environmental NGOs cannot deny that the palm-oil boom has brought major benefits to the economies of producer countries. In Indonesia, the sector is estimated to employ on average 0.4 persons per hectare – at  least 3.2 million jobs in a country where about 30 million people live in poverty. Unskilled slave labor? Well, no. Almost half of the plantations in the country are owned and managed by smallholders. In Thailand the share is as high as three quarters of the total plantation area.

Demand for palm oil remains high and there is now evidence that the palm oil boom might spill over to Central and Western Africa, where about 800,000 hectares of plantation concession have been granted to companies in recent years. Latin American countries too see the opportunity to benefit from the boom, such as Brazil, which has recently included oil palm in its reforestation plans, bolstered by a generous subsidy scheme for smallholders. So is the way inevitably paved for the palm oil industry to embark on a new round of forest-destroying plantation expansion?

Land use planning as a way forward
In many countries land is available – mainly degraded forests and grassland – to satisfy the future demand for palm oil in a less damaging way. Earmarking the right sites for palm plantations requires a good deal of capacity and knowledge by local authorities about where natural conditions are suitable for oil palm, which environmental and social safeguards need to be considered and at which place – a land use planning process.

Global palm oil suitability map

The map above shows where palm oil production is possible on the globe when taking into account climate, soil, and topography features.

The first step toward more sustainable oil production is a map indicating where bio-physical conditions are suitable for oil palm cultivation. To that end, we constructed a global bio-physical suitability map, building on climate, soil and topography data at the resolution of 1 km. The map reveals that in fact the Amazon basin – the better part of it is located in Brazil – harbors by far the biggest stretch of suitable land, followed by the Democratic Republic of the Congo (DRC) and Indonesia.

IIASA supports the MOABI platform, a collaborative mapping initiative that aims to increase transparency and accountability on resource issues in DRC. Our oil palm suitability map will help to inform this process by providing insight to the sustainability of the expected expansion of oil palm in DRC in the coming years.

This map shows areas that are potentially suitable for oil palm cultivation in the Democratic Republic of the Congo.

This map shows areas that are potentially suitable for oil palm cultivation in the Democratic Republic of the Congo. The map was developed from our data, and made interactive and zoomable by our partners the Moabi Project.

Biophysical suitability is not all
However, if and where plantations will start to appear will depend on many factors, most of which are economic :

  • Availability, productivity and costs of land and labor
  • The institutional set-up and support for the sector
  • Accessibility to refinery plants and markets is a key determinant for oil palm plantations profitability

In order to address these issues, an economic model such as IIASA’s Global Biosphere Management Model (GLOBIOM)  model can be deployed to gain insights in the likely development of the sector, help land use planning and explicitly show the trade-off between economic development and biodiversity protection.

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.

REDD+: Cutting emissions, not trees, in the Congo Basin

By Aline Mosnier, IIASA Research Scholar

Aline Mosnier

Aline Mosnier

Deforestation and forest degradation contribute substantially to greenhouse gas emissions, particularly in developing countries. The Reducing Emissions from Deforestation and Forest Degradation plus forest conservation, sustainable management of forests and enhancement of forest carbon stocks (REDD+) Initiative, launched in 2008 by the United National Framework Convention on Climate Change (UNFCCC), aims to help developing countries prevent such deforestation and degradation. It creates a mechanism that would provide financial compensation to developing countries that make efforts to address these problems. Some funding has started to flow to build REDD+ readiness plans and forest monitoring capacity. However, many methodological issues stand in the way of reaching agreements and attracting enough funding for the initiative to succeed.

One of the core ideas of REDD+ is that payments should be based on results. But particularly in Congo Basin countries, where I recently spent three weeks meeting with stakeholders and policymakers on REDD+ plans and goals, determining results is not an easy task.

How do we measure performance? First, we must agree on a benchmark to which the future efforts can be compared. The simplest benchmark is perhaps just to compare current efforts to the past: using past data has the advantage of being based on facts and consequently less prone to inflation. But for this to work, one has to believe that the past is the best predictor of the future.

The Congo Basin countries have a problem: they have high forest cover and low historical deforestation rates… but fast-growing needs.

Yaounde, Cameron. Photo credit:  Aline Mosnier.

Yaounde, Cameron. Photo credit: Aline Mosnier.

The low historical deforestation rates in the Congo Basin countries result from several factors. Some argue that conflicts, unfavorable investment climate, lack of infrastructure, and low levels of economic development have led to a “passive protection” of the forests. But the context is changing. Presidents of the Congo Basin countries have big plans–they want to become emerging countries within the next two decades–and they are looking for new opportunities. Foreign investment projects in mining, oil, agro-industrial plantations, and large-scale agriculture are now flourishing in the Congo Basin, and protected areas are under threat. Local communities could be threatened by expropriation and pollution from large scale projects, but at the same time these communities are also eager to see new employment opportunities.

What does this situation tell us about REDD for the Congo Basin? First, payments for living forests are necessary to avoid deforestation because this is the only way to convince developing countries that forests are valuable. These payments have to benefit both local communities who are living next to the forest, and governments who are making the decisions about large-scale conversion of forests.

Second, if payments are conditional to reduction compared to past deforestation, we can’t expect much from REDD in the Congo Basin countries. If payments are delivered based on lower future deforestation rates and are not underestimated compared to what could be foreseen according to countries development needs, the international community has a chance to make a change.

But this needs trust. Trying to quantify future emissions from deforestation and forest degradation is challenging and undoubtedly involves large uncertainties. However, by engaging with stakeholders to understand the local context while having independent funding, by building the models under the necessary scrutiny and scientific rigor, and by clearly communicating the results to the international community, scientists could play an important role in finding a fair deal to fight against future deforestation.

At IIASA, we are contributing to this objective under our REDD-PAC project by combining land use and systems analysis tools from IIASA, regional expertise of the National Institute for Space Research (INPE) in Brazil and the Central African Forest Commission (COMIFAC) in the Congo Basin, and the experience of UNEP-WCMC on the multiple benefits of REDD+.

Aline Mosnier contributed to work that will be presented at a special session organized by UNEP-WCMC and IIASA at the Global Landscapes Forum (GLF) at the COP 19th in Warsaw, highlighting the role of land use change models in supporting landscape-scale planning. She recently returned from travels through the Congo Basin, where she met with stakeholders and policymakers.