Combining high-tech and low-tech to turn satellite images into action

  • Remote sensing is not only a science, it's also an art. And in order to interpret the imagery, you need a deep understanding of the land you're studying.
  • Satellite images can show us where the forest patches are, but they don't tell us why. That information comes from communities.
  • Information doesn't automatically transform into a management action, and conservation scientists must learn how to turn scientific information and big data into engaging stories.


Dr. Lilian Pintea, Vice President of conservation science at the Jane Goodall Institute. Photo courtesy of JGI/Jeff Kerby.

Since its founding, the Jane Goodall Institute (JGI) has been on the forefront of conservation science, and in recent years the group has been applying remote sensing, mobile phone technologies, and cloud-based mapping tools to its forest and ape habitat conservation work. Today, JGI scientists use a suite of technologies to monitor chimpanzee habitat, from high-resolution imagery from DigitalGlobe, to Unmanned Aerial Vehicles (UAVs) or drones. But the largest source comes from Landsat imagery at 30-meter resolution, which can be used for long-term analysis going back in time or for habitat monitoring covering large areas. Very recently, it has become possible to use it to monitor change in forest cover in near-real time, and in some cases even once per week.

To learn more about the chimpanzee habitat conservation work JGI is tackling and the future of remote sensing, WildTech spoke with Dr. Lilian Pintea, JGI’s Vice President of Conservation Science.

Answers have been edited for clarity and length.

We’re starting to see cloud mapping platforms like ArcGIS online and Google Earth Engine make large-scale modeling much faster. Could you give an example of how these cloud mapping tools work with remote sensing projects?

Cloud platforms are transforming the ways we access, process, and collaborate to use remote sensing imagery. It opens big data for conservation applications, making remote sensing imagery more feasible and cost-effective to use. For example, biomass monitoring and mapping is a very important part of any REDD project. Colleagues at the Woods Hole Research Center developed a biomass mapping approach that uses MODIS imagery, GLAS and other datasets (Baccini et al. 2012)

However, one of the challenges with running such large-scale models is that you need to download and manage the data. Data acquisition and management can consume 60 to 80 percent of the cost of a GIS project. So as part of the JGI REDD preparedness project in Tanzania we partnered with Google Earth Outreach and Woods Hole and demonstrated for the first time in 2012 that you could take such a complex biomass model and run it in the Google Earth Engine cloud using only a computer web browser and internet access; no need to download and manage large satellite imagery datasets and software packages. Remote sensing specialists could spend more time and resources to interpret the data and communicate it to decision makers. Plus, while slow on the local computer workstations, the Baccini et al. model automatically refreshed on the fly in Google Earth Engine.

Another example is accessing up-to-date very high-resolution satellite imagery. I recently conducted a training with Global Forest Watch and Uganda Wildlife Authority in Kibale National Park. I didn’t have the latest imagery with me, but I was able to access it through a mobile modem via DigitalGlobe’s ImageConnect add-on for Esri ArcMap desktop. It was amazing to be in the field and see that cloud technologies had reached a point that could truly open the imagery to conservation practitioners and partners on the ground.

How has the increasing availability of satellite imagery and remote sensing data affected traditional habitat monitoring and ground surveys?

Local communities are a major focus of JGI’s work. Many of our conservation strategies and actions are happening on community land with their participation, so high-resolution imagery opens the possibility to use the imagery as base maps for documenting, ground-truthing, and geo referencing community local knowledge, values and uses of the place. And this is important because, often in places where we work in Africa topographic maps can be very outdated, and high resolution satellite imagery overlaid with local names of the places are the best basemaps that we have. We started doing participatory interpretation of high resolution satellite imagery with the local communities in 2003, and it has been incredible to have [the imagery] as a common language. From mapping sacred sites to understanding why certain forest patches have avoided deforestation, local knowledge puts the imagery in the local context. Satellite imagery on its own is great for showing where the forest patches are, but it doesn’t tell you why those forests are protected or not by the local people. So [the answer to] that “why” question is coming from communities.

In conservation, forest loss [per se] is not the threat, the threat is conversion of forest to incompatible agriculture, or illegal logging, or incompatible charcoal production. If you don’t know why the forest loss occurred, then you cannot develop effective conservation strategies. An example of how this plays out is in riverine forest in Masito-Ugalla and Greater Mahale Ecosystems in western Tanzania, which is critical habitat for chimpanzees. Being able to have high-resolution imagery, zoom in to riverine forest and see that a lot of this forest has been converted to rice fields or other crops, where you can see scattered human houses being 20-30 km away from the nearby village, and the human foot paths. It allows you to not only better quantify forest loss but also to have a much better understanding of the sources of the threats such as incompatible expansion of scattered settlements and farming that led to the forest loss.

Mwamgongo land use mapping
Community members consult a map in Mwamgongo village in western Tanzania. Detailed maps made from satellite imagery are used  by JGI for community land use mapping. Photo courtesy of the Jane Goodall Institute.

Do you see remote sensing replacing traditional ground surveys?

It depends. Take, for example, if you’re doing a classification with Landsat imagery at 30 meters resolution and your goal is to map forest and farmland land cover. In the past you’d have to go in the field and ground-truth to develop an understanding how that spectral response, color, or shape in the imagery is related to different land cover types. Now, if you have high-resolution imagery collected in near real-time, you can just zoom in and see if that is a farm or a forest.

However, remote sensing is not only a science, it’s also an art. And to fully interpret the imagery, you have to have a deeper and more holistic understanding of how numerous factors – ecological, social, economic, historic, etc. – all play together to shape the way the landscape looks and changes. So if you don’t have this understanding that comes from spending time on the ground, I don’t think you are really discovering the real meaning in the remote sensing pixels. So I’m a big supporter that you have to mine the pixel, not just using computer software, but mine its meaning by looking at it from a variety of perspectives. You need the local context.

Do you think that Open Data Kit and other mobile technologies can help translate some of those local aspects into data points to help remote sensing?

It’s fascinating. Since 2009 we have been empowering village forest monitors from the local communities in Tanzania and Uganda to use Open Data Kit (ODK), a field data collection app running on Android smartphones or tablets. It’s clear that there is huge potential for the local communities to engage and contribute citizen science data using mobile apps like ODK that could help communities to better manage their forests while also contributing ground-truth data to remote sensing scientists. We recently launched a mobile app in Uganda called Forest Watcher that works with ODK to enable local users in the field to access and use offline GFW forest loss alerts while potentially contributing their field georeferenced observations and pictures as validation data to improve remote sensing models.

I am also excited to explore how other technologies could be used for validating remote sensing data. In 2014, with Google Earth Outreach, we went to Gombe National Park and did a Street View project. We did it for a number of reasons; one is to give people around the world virtual tours of Gombe… but also to document different habitat types – as a record so that five years later we can redo these transects to see how they’ve changed. It’s monitoring and ground-truthing. It’s a different perspective which could complement ongoing monitoring efforts from satellites, UAVs, and on the ground by the local communities using mobile apps.

High-resolution satellite imagery shows forest cover regeneration in the Kigalye Village Forest in (2005) and (2014). The Forest Reserve was established in 2005-2009 as part of Kigalye participatory village land use planning process. Photos courtesy Lilian Pintea/JGI, DigitalGlobe, and Esri.

How is JGI integrating satellite imagery into on-the-ground conservation and enforcement?

We’re using [the imagery] in a variety of ways. First as a common language to capture and integrate traditional knowledge with science knowledge.  We use satellite data as input to predict and model potential chimpanzee habitats and distributions. Most importantly, we now use imagery along with Geodesign and Esri’s ArcGIS platform not only to support creation of the village land use and conservation action plans, but also to measure enforcement. It’s important to develop a land use plan, but without implementation it’s a plan on a paper. At what extent do these paper plans and paper protected areas actually transform into real protected areas with impact on the ground? So monitoring and evaluating conservation success of habitat conservation efforts is important.

Can you tell us more about the Decision Support System (DSS) that JGI is building? What sparked the idea and what overarching problems is it trying to address?

Chimpanzees are threatened by habitat loss and fragmentation, illegal bushmeat hunting and trade, disease, and the illegal pet trade. Habitat loss and bushmeat are the major ones, and habitat loss is one of the most irreversible threats out there. Once you lose the habitat, it’s very expensive to get it back. Remote sensing has been very advantageous in our study sites. However, the entire chimpanzee range is covered by more than one hundred individual Landsat satellite scenes. Therefore, most of the analysis was done at the site or regional level using few Landsat scenes. As the result conservation practitioners end up with different vegetation and habitat maps done by different groups and organizations across Africa that you can’t easily compare because of different methodologies and classification systems. Another challenge is timing between image acquisition and development of a mapping product. Even if you classify your entire site, by the time you make another map, you’ll find deforestation already happened three years ago, and you’re behind in terms of developing an effective conservation strategy.

So when this breakthrough with Matt Hansen’s team [at University of Maryland] happened, in terms of being able to mosaic the image scenes and do the analysis at the pixel level, we immediately recognized the opportunity to explore if we can use it to monitor chimpanzee habitats from the village to species range wide scales in Africa.

We asked ourselves, “can we use a combination of species modeling with Landsat satellite imagery and crowd-sourced data to systematically monitor habitat at scales locally relevant and consistent across the entire chimpanzee range in Africa?” We wanted to focus on continuously updated data derived directly from remote sensing in order to detect and map change in habitat suitability and health in near real-time, as satellite imagery becomes increasingly available.

Fig 3 - Geoplanner screenshot
A screenshot from Esri’s ArcGIS Online Geoplanner app shows a dashboard view of the same forest change in Kigalye Village Forest Reserve. Photo courtesy of the Jane Goodall Institute, DigitalGlobe and Esri.

What types of data will be used with the DSS? How will the community-based aspects of JGI’s work feed into the DSS?

For the habitat health index project supported by NASA, JGI and University of Maryland looked at around 23 potential variables and picked 12. In addition to Landsat ETM+ bands, we also used forest data derived from remote sensing, like canopy cover and canopy height, along with distance to steep slope, and elevation – chimpanzees like to nest on steep slopes in certain regions like western Tanzania and eastern [Democratic Republic of Congo]. Then we used a number of landscape indicators calculated in GIS, such as distance to the forest edge and forest loss and edge in 25 km buffers.

We have trained around 100 village forest monitors to collect field data [including data on chimpanzee presence] in Tanzania, and around 200 forest monitors in Uganda. That’s still relatively small, but with time we hope to scale up and possibly use crowd-sourced data as an input to the model. Citizen science data nicely complement survey efforts collected by researchers. In places where we work, scientists tend to focus on relatively intact areas away from human settlements, while village forest monitors are patrolling their community forest reserves as part of their village boundaries. So you end up with a very nice way to complement community and science data in terms of geographical coverage, and get a better understanding of species presence.

Lilian Pintea, JGI-USA (first from right) and Timothy Akugizibwe, JGI-Uganda (first from left) work with NFA rangers on an Open Data Kit mapping project in Budongo Central Forest Reserve, Uganda. Photo courtesy of Peter Apell/JGI

What are some of the gaps in information for the Decision Support System?

There are big gaps in chimpanzee survey efforts and presence data across the range and because of this we can’t use this data in the model just yet. It doesn’t work. And that’s why there is a need to do more surveys across the entire chimp range in Africa. We also don’t have consistent and updatable information on human land uses and infrastructure. Roads, for example, could have been a good input to the chimpanzee suitability model, but because we do not have yet access to an annually updated road layer across Africa, we decided not to include it in the model yet.

What type of information will JGI and its partners be able to glean from the DSS data to influence conservation planning?

JGI uses Open Standards for the Practice of Conservation (OSPC) to guide the development and implementation of its conservation projects and strategies. We also use OSPC with partners to develop a common understanding of the conservation needs, threats and cooperate on joint implementation of conservation actions in specific regions (e.g. Great Apes CAP in Eastern DRC). In the case of chimpanzees, a key component outlined in the OSPC is an overall assessment of the viability or health of chimpanzee habitats and the establishment of a monitoring plan that requires indicators that are measurable, precise, consistent and sensitive.

The goal is to develop a practical and operational DSS that will annually inform this adaptive management cycle and OSPC process at all scales. The focus is on standard and actionable chimpanzee habitat viability and change information that is most likely to trigger a management decision and action.  We are also exploring ways to measure success using forecasting. By looking back at the habitat suitability change over the last 12-13 years, we can also forecast what habitat would look like by 2030 if it’s just “business as usual”. So we could measure our success on what was predicted versus what really happened.

Women focal group mapping2
Women consult a map during a focal group mapping session to identify cultural sites and other community landmarks. Photo courtesy of Lilian Pintea/JGI

How do you see remote sensing data and near-real time monitoring systems developing in the future?

It’s going to have profound effects on forest protection… be more action-oriented, and it will transform the way decision makers use remote sensing. Let’s say you have an alert that there is deforestation happening within the last week, you are more willing to go and see what happened, as opposed to seeing that deforestation happened two years ago. The near real-time is important, because it can almost be used as intelligence to guide law enforcement on the ground.

There are a lot of near-real-time platforms cropping up, whether global like GFW and Global Fishing Watch, or more local like MAAP in the Amazon and one-off habitat assessments. How do we translate all that data into easily-digestible formats and give it the context for decision makers and park managers to use it?

I think we have to recognize first that information doesn’t automatically transform into a management action – even if the information is potentially useful and delivered in time. There’s many other factors that influence how information is used in a specific decision-making process. At the most basic level, I think we need standards. As long as we agree how we will define and measure habitat viability or condition, and what degree of change should trigger management, it shouldn’t matter if you use Google Earth Engine, ArcGIS Online, Azure clouds, GFW, or any other platform. We have to learn how to convert remote sensing and other big data into actionable information to support conservation decisions.

Decision makers are people. And people learn best through storytelling. Through stories we could reach not only the minds, but also the hearts of decision makers. Only then, I believe, we could hope that they will make the right decisions. It doesn’t matter how big the big data is going to be, we need to find a way to properly tell an engaging story.

Article published by
, , , , , , , , , ,