MarProb: conservation planning for the real (uncertain) world
Accounting for risk and dynamics in our spatial planning
There is a need to better account for risk and dynamics in conservation spatial planning
MarProb is risk averse. It either targets sites where we are more certain that the species or habitat is (or will be) there, or it targets sites that have a lower risk of being destroyed by a threatening process
If uncertainty is not explicitly incorporated in all stages of the decision-making process, then the cost of not representing this uncertainty increases leading to failures in conservation planning
Spatial conservation planning is all about using available information to weigh up options about which bits of the landscape (or seascape) will be put aside into conservation reserves – protected areas. But what if the information we are feeding into our planning processes contains errors or uncertainty (think remote-sensed reef data or population models) or there is a chance that threatening processes destroy habitats or ecosystems (think climate-change impacts into the future). What if we don’t account for these uncertainties? Then our management decisions may be suboptimal or ineffective; which translates to the reserves we select being in the wrong place, or the wrong size.
Unfortunately, decision makers are often required to make planning decisions based on inadequate or uncertain knowledge. It’s not a matter of simply collecting better information, this takes time and money that is usually not available. Given this reality, it’s important to incorporate risk or uncertainty associated with the outcome of protected area decisions into the planning process itself and a recent version of the conservation-planning software Marxan does just that. It’s called MarProb (or Marxan with Probability) and to demonstrate its utility (and the importance of taking uncertainty in planning into account) I’d like to share with you three recent research projects I’ve been involved with in which MarProb was applied.
The first describes a marine spatial planning framework which targets good condition coral reefs for protection, given the chance of oil-palm development and associated runoff degrading reefs (in Papua New Guinea). The second accounts for data uncertainty by evaluating trade-offs between accuracy and resolution of coral-reef habitat data derived from remote sensing (in Fiji). And the final example discusses how conservation planning in boreal forests can incorporate threats posed by future climate change (in Canada).
Three conservation challenges dealing with different forms of uncertainty in three very different parts of the world. What they demonstrate together is that by acknowledging and incorporating uncertainty up front it’s possible to generate significantly improved conservation outcomes.
1. Oil palm and conservation planning
Mention oil-palm plantations and most people think of the impact to tropical rainforests that usually get cleared to make way for the plantations. What’s often overlooked is that clearing forests for oil-palm plantations is also a major threat to tropical freshwater biodiversity, and can potentially affect downstream marine ecosystems such as coral reefs. Planners aiming to protect coastal coral reefs (and other marine ecosystems) should account for the impacts of current and possible future development on the land using ‘ridge-to-reef’ planning. The aim is to avoid placing marine reserves in areas that might be degraded from runoff (meaning the reserves would fail to protect the values they were set up for).
We also need to find ways to modify oil-palm development to reduce negative impacts on connected marine ecosystems. In many cases this requires knowledge on how coral reefs respond to runoff from oil palm, which we don’t have because most oil-palm plantations are in remote tropical areas where data is scarce.
We developed a holistic ridge-to-reef approach for marine conservation that predicts the response of coral and seagrass ecosystems to changing land-based threats (current land-use, and three oil-palm development scenarios ranging from extreme to best practice).
We applied this in a data-limited region in Papua New Guinea (Tulloch et al, 2016). We then developed a network of marine reserves that avoid highly degraded reefs from possible oil palm expansion in the future. Marxan with Probability offers a new approach to considering threatening processes, allowing planners to incorporate information on the probability that the feature exists but is degraded from threatening processes and can not contribute towards conservation goals.
Almost 60% of coastal ecosystems were predicted to be substantially degraded in five years’ time if all suitable land was converted to oil palm.
Strategic planning for palm oil can deliver substantial benefits to reefs in PNG – we should be placing new marine reserves in turbid areas containing coral reefs that are more tolerant to high levels of sediment in the water, as these areas are more likely to be in good condition even if oil palm expansion occurs in the future.
Importantly, we evaluated global sustainability guidelines for oil-palm development, and show that these guidelines cannot be truly ecologically sustainable unless they are modified to account for the impacts of oil palm on coastal marine ecosystems. Substantially reducing the impact of oil palm development on marine ecosystems requires limiting new plantings to hill slopes below 15°, a more stringent restriction than currently allowed for in the RSPO (Roundtable on Sustainable Oil Palm) guidelines.
The work is critically important as conservation organisations continue to work in regions where there are substantial trade-offs between economic development and conservation activities, but data are scarce to evaluate options for sustainability.
Tulloch VJD, CJ Brown, HP Possingham, SD Jupiter, JM Maina & C Klein (2016). Improving conservation outcomes for coral reefs affected by future oil palm development in Papua New Guinea. Biological Conservation 203: 43-54. http://www.sciencedirect.com/science/article/pii/S0006320716303160
2. Mapping uncertainty when planning MPAs in Fiji
One of the challenges faced by researchers and conservation practitioners in making decisions is reconciling the trade-offs associated with using biodiversity data of differing qualities, particularly when funds are limited.
Conservation planners often have to use habitat maps derived from remote sensing because there is limited information on the distribution of species, but these maps can be very inaccurate, and vary in their ability to represent other species distributions.
Coarse-scale habitat maps are commonly used as proxies to design marine reserve networks when detailed biodiversity data are incomplete or unavailable. Although finer-resolution habitat data provide more detail, they may have more errors due to misclassification, a common problem with remotely sensed data (see Decision Point #79). Despite this, planners usually don’t think about habitat map quality or the effect of mapping errors in when planning reserves.
How much difference can this make? We attempted to find this out. We evaluated how habitat-mapping accuracy at different spatial scales affects reserve priorities and costs by comparing reserves designed using Marxan with Probability that used two different map classifications – one highly accurate simple map of nine coarse-scale seabed structures (slope, reef crest, etc), the other more complex habitat map describing 33 small-scale coral reef with high classification error (Tulloch et al, 2017).
By making this comparison we highlighted the trade-offs between the choice of habitat data, costs, and surrogacy value. These are all important variables when making decisions in conservation planning as they affect the location, size, and cost-effectiveness of the selected priority conservation areas.
Our analysis demonstrates that it is cheaper to use simple maps of reef structure to design reserves due to the high costs of obtaining and verifying more detailed habitat data. Using these simpler maps, however, means the selected reserves might not contain the full diversity of coral reef habitats.
Using more detailed or complex habitat information ensures that we are targeting the full diversity of coral habitats, but these data are more expensive to obtain, leading to a more costly reserve design process.
Using more detailed or complex habitat information also results in larger reserves being selected. This is because these data have more misclassification errors so we need to protect more area to buffer our uncertainty in terms of what we are getting.
Our study highlights the need for error information to be provided with any habitat map, and this information needs to be included in the decision-making process. This is particularly the case when maps of high thematic complexity and error are used.
Tulloch VJ, CJ Klein, SD Jupiter, AIT Tulloch, C Roelfsema & HP Possingham (2017). Trade-offs between data resolution, accuracy, and cost when choosing information to plan reserves for coral reef ecosystems, Journal of Environmental Management 188: 108-119. http://dx.doi.org/10.1016/j.
3. Designing forest reserves for future climate
Canada’s boreal forest is one of Earth’s largest remaining wilderness areas but changing climate is expected to have large impacts on its function and structure due to altered temperature, rainfall and seasonality. When planning conservation reserves in such landscapes, how do we incorporate knowledge about such change when there is considerable uncertainty around future conditions?
Marxan with Probability offers new approaches to planning under climate change. It is especially useful for dealing with the uncertainty around the exact response of species and habitats to climate change, which may vary from place to place.
In this research we showed how ecologically based strategies for climate change adaptation could be integrated (Powers et al, 2016). We used predicted spatial distributions of biodiversity in Canada’s boreal for the year 2080 based on vegetation productivity.
We targeted areas that minimise variability in projected vegetation productivity, as these may represent a less risky conservation investment by reducing the amount of anticipated environmental change.
We developed hypothetical protected area networks designed for future vegetation variability under a range of different IPCC climate scenarios (least change (B1), business as usual (A1B) and most extreme change (A2)).
Including future climate change impacts into national or boreal-wide conservation assessments increases the total area and cost of reserve networks. But failing to do so risks the conservation value of the network. Reserve networks designed for current or least change (B1) climate scenarios will likely not achieve conservation targets when faced with more severe conditions, and will require additional sites.
We can use assessments like these to provide recommendations for adaptive conservation for future climate change that support ongoing boreal conservation and land-use planning.
Powers RP, NC Coops, VJ Tulloch, SE Gergel, TA Nelson, & MA Wulder (2016). A conservation assessment of Canada’s boreal forest incorporating alternate climate change scenarios. Remote Sens Ecol Conserv. doi:10.1002/rse2.34 http://onlinelibrary.wiley.com/doi/10.1002/rse2.34/abstract
How MarProb works
The traditional version of Marxan aims to achieve biodiversity objectives for the lowest cost (see Decision Point #62). MarProb solves the same problem but is also able to maximise the probability of protecting every conservation feature given uncertainty in its distribution, or the chance that features in a site are lost or degraded due to threatening processes. It does this by applying a penalty to habitats or planning units that do not meet a certain level of confidence or certainty, and so is less likely to include them in the final solution as doing so will increase the objective function score.
Essentially, this makes Marprob risk averse – we either target sites where we are more certain that the species or habitat is there, or we target sites that have a lower risk of being destroyed by a threatening process (such as climate change). But if we need to target species or habitats associated with high risk (eg, coral reefs), Marxan chooses more. In this sense MarProb is hedging your bets – making sure you don’t end up with nothing at the end of the day.
More info: Vivitskaia Tulloch email@example.com