iLand News

I wanted to share some more thoughts on scales and scaling in ecosystem management (and subsequently in modeling) for quite a while now, and will take the recent publication of a paper on the issue as an opportunity to finally do so. This'll be a two-part post with some more general thoughts in pt.1 and a practical example of scale effects in pt.2.

Before I start, I however also want to make clear that I do not attempt to provide an exhaustive summary on the vast literature on scaling in ecology here. Scaling is one of the central problems in ecology, and others have elaborated more deeply and clearly on it that I ever could (see e.g., the seminal papers here and here).

What I find interesting though, coming at it from an applied angle, is that scaling issues are for the most part not explicitly addressed in the applied disciplines, such as in forestry. Of course, many traditional concepts of forestry are inherently scale-dependent. Think sustainable yield, which is not achieved at the entity of actual management decision making (the stand scale) but at the hierarchically higher management unit level (the landscape scale). So while the importance of different scales is generally acknowledged in forestry, scaling up is oftentimes perceived as not much more than adding up trees. This naïve view of scaling, however, neglects that some processes change nonlinearly with scale, that others are relevant and can only be described at particular scales, and that interactions across scales can have strong impacts on ecosystem dynamics (if you care for the technical terms, check out transmutation, emergence, and path dependence for starters). Yet, I would argue that many of these issues are crucial in the context of current challenges of forest management, from managing to conserve biodiversity to mitigating climate change in forestry. This, imho, asks for a stronger consideration of scaling issues in management.

We’ve described some examples for scaling issues in forest management in more detail in the paper – here I want to come back to modeling though. The modelers amongst you will probably agree that design decisions about spatial and temporal scale are among the most important in developing a model. They ultimately determine which processes can be accommodated within the framework of a model, and which questions can be addressed with it (e.g., dispersal and migration cannot be modeled at the stand scale in any meaningful way). In many instances, different scales increase the complexity of the problem. Generally speaking, this is counter to the art of modeling, which is to reduce the real life complexity to (mathematically) tractable levels. On the other hand, scaling can be a modeler’s best friend, as often predictability of ecological processes increases with scale. So choosing scales wisely can make a great difference in modeling. The traditional way of doing it was to carefully select a focal scale to match the questions and processes to be addressed with a model. This usually leaves processes at underlying scales implicitly lumped into gross equations, while controlling for processes at higher hierarchical levels by assuming them constant (or random). This reductionist approach makes for very successful and powerful models at their particular scales. Yet, it foregoes one imho increasingly important application of models, and that is to understand and predict cross-scale phenomena. Ultimately, if you’re interested in an emergent property such as resilience, for instance, you need to capture exactly this emergence in order to be able to simulate and predict it. If, because focused on a single scale, you inhibit emergence in the model by design, your model will not do a good job in addressing such a property.

My main point here is that many of the current challenges in forest management are multi-scale issues. And in order to address these multi-scale issues we need multi-scale models, as only those will help us understand these complex properties better, and ultimately help managers to resolve their scaling-related management issues. Scaling is, after all, one of the major strengths of simulation models in the context of ecosystem management. Just to show you that this multi-scale modeling is more than just an intellectual exercise and can actually be done in practice, have a look here. Also, I promise that part 2 of the scaling post will be less theoretical and more hands-on, so stay tuned!

If it wasn't for another round of recent technological advances (the computer world never sleeps, it seems!) it would have been seriously quiet around here recently. One thing that's certainly not behind this quiet is the iLand world being idle... Au contraire - we're happily hacking away, both on the modeling side as well as on the publishing side of things, working on a few - as we think - quite novel ideas.

Ah novelty... it is such an ambivalent thing in science - everybody wants it, we all need it to get funded, or published (which we again need to get funded, which...). But it is hard to come by after 100+ years of systematic scientific inquiry, and most (if not all) scientific breakthroughs are incremental and build on a large body of previous work, if one takes a closer look. Yet, also funding agencies and journals need to sell their decisions these days, I guess, and can't always escape the trend towards a societal attention span equaling that of a toddler.

Anyway, enough with ranting already. We hope to get back with more details on our novel activities soon, but in the meanwhile, here's a link to a recent critique on the obsession with novelty in current academic publishing. It probably wins the "best paper title of the year" award too, so make sure to click through - if just for the title.

It’s a new year (well, sort of… by now we’re through 10% of 2013 already, sigh), and what better way to start off the new blogging season with introducing two new (iLand-related) research projects that are starting up here at BOKU, Vienna:

The first one, entitled “Agent-based MOdeling of Climate Change Adaptation in forest management” (MOCCA) and funded by the Austrian Climate and Energy Fund ACRP, will be focusing on the development of an agent-based management module for iLand. The aim is to better be able to simulate managed forest ecosystems, and include adaptive behavior of managers (e.g. to dynamically changing disturbance regimes) into the simulations. Our main motivation and hypothesis is that considering purely prescriptive management in simulations only (as we’ve done previously) might lead to an underestimation of resilience. The project thus will not only be a major next step in developing iLand further, but will also apply the model to address questions of how to robustly manage forest ecosystems under the uncertainty of changing climate and disturbance regimes. We will start MOCCA in March 2013, and are delighted that we’ll be collaborating with Bernhard Wolfslehner (EFICEEC-EFISEE) and Kristina Blennow (SLU) in its implementation.

The second grant we’ve just got accepted is “Simulating Adaptation of forest manaGement to changing climate and disturbancE regimes” (SAGE), and it’s an European Union FP7 Marie Curie Career Integration Grant (CIG) awarded to me (Rupert). As you can see already from the title, we’ll be working on issues closely related to MOCCA also under this grant – CIGs in general are conceived as additional funding, intended to bolster researchers starting up their own group at a new host/ new position. The CIG will allow us to go considerably beyond what we proposed for MOCCA. It will enable us to pursue investigations on disturbance interactions with iLand, study disturbance – climate relationships also at larger scales (in collaboration with Mart-Jan Schelhaas, Alterra), and continue our iLand-based work on resilience and complexity in forests ecosystems of the US Pacific Northwest (together with Tom Spies). So SAGE will contribute considerably to advancing our research on disturbance dynamics and resilience to disturbances in forest management, and will also bring about a number of new and exciting iLand applications.

So overall we’ve had a quite good start into 2013, and are very much looking forward to diving into the questions and materials of these two new projects. We’ll keep you updated on the progress as we go. And, just in case you were wondering: We’ll of course continue our strict open source policy also under MOCCA and SAGE, and will eventually make our new developments (code, executables, etc.) publicly available through this website – stay tuned!

Well, the headline might not sound as exhilarating as, for instance, Paris is burning would, but it has Werner and me quite excited in any case. After months of preparing the data, testing and evaluating the model, and discussing about the simulation design we've just started a new iLand simulation experiment.

In a nutshell: We'll be testing in silico what the (long-term) effects of disturbance legacies are on forest structure, composition, and functioning. Once more using the HJ Andrews Experimental Forest as our study system, we are conducting simulations with different levels of green-tree legacies after high severity disturbance, in order to gain insights into the magnitude and temporal persistence of legacy effects. In addition, we ask how subsequent disturbance is altering the legacy effect, i.e. if wildfire frequency were to increase in the future, would legacy effects be dampened?

These questions are motivated by a growing focus on retention and legacies in ecosystem management, as well as by the observation that disturbance legacies can influence successional trajectories profoundly and might even induce path dependence in forest dynamics.

Oh, and by the way, in order to ask these questions and conduct this study we have implemented and tested a dynamic wildfire module in iLand (see an older blog post about some ideas here), which is able to simulate fire regimes as an emerging property of vegetation, weather/ climate, and topography.

We're curious as to what our findings will be, and will be back with an update in due time. For now we enjoy having the warmest office of the institute, with all the CPUs running at 100%, which - on a cold and grey December day like this - is a nice co-benefit of being a simulation modeler!

So, as already mentioned by Werner over at the tech blog, we're happy to announce the release of a new iLand version (v0.69). This release marks the publication of another major iLand paper (read more about it here). While my previous post should give you an idea about what this new version is capable of and what it can be used for, I briefly want to highlight the major changes from the previously released version in this post.
In a nutshell, iLand v0.69 is the first release that has full dynamic landscape modeling capabilities (i.e., simulates the major demographic processes of growth, mortality, and regeneration), and simulates a closed carbon (C) cycle (i.e., accounts of both above- and belowground forest C compartments). The major changes from the previously released version thus relate mostly to the inclusion of dynamic regeneration and soil modules.

Regeneration modeling

We followed the general recommendation of Price et al. (2001) to address the main processes associated with forest regeneration explicitly. This means that iLand explicitly addresses the processes of seed dispersal (in space and time), the germination and establishment of trees, as well as the growth and competition of saplings. The technical details can be found at the respective model documentation wiki pages. Here are some more general thoughts on the regeneration modeling in iLand. One of the big challenges was how to address the potentially very large number of seedlings and saplings at the landscape scale in an efficient manner, while still maintaining the ability to capture the structure heterogeneity and spatial complexity associated with regeneration dynamics (cf. the initial objectives of model design in the iLand wiki history section.). In order to address this challenge we applied a (computationally efficient) mean tree approach to model seedlings and saplings at a high spatial resolution of 2 x 2 meters. This approach allows us to model inter-specific height growth competition in the regeneration layer explicitly, while it is very sensitive to the type (e.g., management, wildfire), size (small gap vs. large clearing), and location (relative to seed trees) of canopy disturbances. To account for the effect of the latter on regeneration we again used iLand's detailed light simulation routine and the continuous light influence field calculated by the model at the landscape scale - with great success, as attested by the evaluation conducted in Seidl et al. (2012b). Overall, the model is sensitive to climate, and is able to reproduce both the temporal patterns of species succession as well as the spatial patterns of light-driven regeneration in temperate forest ecosystems.

Soil and decomposition modeling

The main objectives with regard to modeling decomposition and soil processes were to select an approach that

• consistently simulates effects of climate change, management and disturbances on soil C stocks, and
• allows first order plant-soil feedbacks with regard to nutrient availability.

The challenge here was to select an approach that is robust and tractable in landscape scale simulations. In this regard, the selection of a relatively simple model was supported by a recent meta-analysis by Manzoni and Proporato (2009), who found that for modeling general soil processes <5 state variables suffice (whereas a considerably higher model complexity is required for modeling certain soil aspects with a high level of detail). After reviewing the literature and analyzing seven soil modeling approaches in detail (i.e. Standcarb, TRACE, YASSO07, CENTURY, Biome-BGC, LPJ, the ICBM-family), particularly contrasting their designs and abilities with the above mentioned objectives, we adopted the ICBM/2N approach as the soil C and N cycling module in iLand. ICBM/2N was developed as a relatively simple, analytically solvable model to study climate and management effects on soil C (Andrén and Kätterer 1997). The iLand soil and decomposition model accounts for eight detritus pools, i.e., snags, downed woody debris, litter, and soil organic matter for both C and N.
An important issue in the context of simulating forest soil dynamics is the initialization and parameterization of soil models, since high quality soil data with continuous coverage are rarely available in potential study landscapes. In this regard the new iLand soil module was designed to flexibly use the available data while not constraining the models' applicability in cases where wall-to-wall soil data coverage is not available for a study region. The model can thus be initialized with empirical data where available, but can also be spun-up to derive initial C and N pools from the simulation itself (see here for more details). Also, it can be used both with N feedbacks on plant growth simulated dynamically and using an externally derived fertility rating (which can be hand for analytical purposes, see e.g. here). To test the model we compared simulations against observed C stocks at the HJ Andrews Experimental Forest, finding good agreement between simulation results and empirical data (Seidl et al. 2012b).

In summary, iLand v0.69 marks a major milestone in the development of iLand, and presents the first release that is capable of simulating closed C cycles and full forest landscape dynamics. If you want to try it out just head over here and download a copy - we've included a test landscape which allows you to run the model out of the box! Next up in iLand world: a focus on disturbance modules (e.g, wind, wildfire), and more applications, finally getting around to using the model to answer some of the questions it was originally developed for. Stay tuned!

For those of you who didn’t catch the iLand presentation at ESA, here’s a short summary of what is the first fully-fledged landscape scale application of iLand (you can read more about it in a recently accepted paper in Ecosystems). The background and motivation to the study was that forest ecosystems store large amounts of carbon (C), and take up C from the atmosphere, mitigating the anthropogenic greenhouse effect. The amount of C stored in forests, however, varies considerably in the landscape, and our understanding of what causes this variation is still limited.

We used iLand to unravel the main drivers of forest C stocks at the HJ Andrews Experimental Forest (HJA) in Oregon. A key methodological issues that iLand helped to resolve was that a multitude of - hierarchically nested - drivers affect the processes in forest ecosystems, making it anything else but straight forward to interpret correlations of individual factors with forest C stocks. Climate, for example, influences carbon-relevant process rates like heterotrophic respiration (i.e., the flux of C from the soil back to the atmosphere) directly, but also affects the species composition of a forest (which in turn can affect the C cycle). The reason why we in particular tried to understand the role of climate on forest C is that climate will likely change drastically in the future, making it important to understand how sensitive forest C stocks are to such changes. On the other hand, aspects like species composition and stand structure are the factors that we can influence directly by means of forest management. Managing forests for climate change mitigation thus requires an understanding of what features can positively influence the forest C balance, and to what degree.

Using iLand, we disentangled drivers of forest C storage at the HJA, with quite interesting results. Contrary to other study we found that variation climate only accounted for approximately half of the variation in C, despite the fact that our study landscape is characterized by strong environmental gradients and complex terrain. The effects of forest structure and composition were found to be in the same order of magnitude than that of environmental drivers. In particular, we found that diversity in species and structure was positively associated with higher C stocks. In other words, old-growth features like diversity in tree species and sizes are not only important for species depending on forests for habitat, but they also enhance forest C stocks – adding one more reason to managing for complexity! Read more and download the full article here.

For those of you who happen to be in Portland for this years’ annual meeting of the Ecological Society of America: We’ll be presenting an iLand application in the contributed oral session on linking community structure and ecosystem function. The title of the paper is “Drivers of spatial variation in old-growth forest carbon density disentangled with LiDAR and an individual-based landscape model”. In the paper we’ve used iLand to
(i) estimate C stocks of old-growth forests at the HJ Andrews experimental forest, and
(ii) conduct a hierarchical simulation experiment to disentangle environmental and community drivers of spatial C variation.
I’m looking forward to an interesting session and discussion!

It’s a few weeks now since the project funding the initial development of iLand, an European Commission FP7 Marie Curie International Outgoing Fellowship (IOF), came to a close. Time to reflect some more on the past and future of iLand…

Looking back, the IOF was a great opportunity to develop a new simulation tool, particularly considering that grant programs funding the development of new methodologies are becoming fewer in favor of research directly addressing the “grand challenges” of today (I would argue that we direly need the former to do the latter, but that’s just my two cents, for what they are worth). Also, three years was a great time frame for such an endeavor, with enough time to really dive into the topic, study the literature extensively, and play around with approaches developed previously before sitting down and starting to sketch out new ideas (the iLand history pages give a little glimpse into the past of the model).

I just once more read through the project proposal and the reviews thereof, contrasting what we had proposed four years ago with what we have accomplished. To put it with the words of a Reviewer of the proposal: “The plan is ambitious”. It certainly was, and it certainly took a lot of energy to get where we are now. But, as E.H. Land said: "Don't undertake a project unless it's manifestly important and nearly impossible.” Considering that we have accomplished all the main objectives of the project is thus all the more satisfying – a big thank you to everybody who contributed to making this success possible.

Which brings me to the future of iLand: The end of the initial IOF is of course not the end of the model, it on the contrary marks a new phase in the “life” of iLand. After three years of basic model development we’re eager to see iLand being put to use. With the first application paper already in review, and iLand being a central part of several proposals currently being evaluated we’re confident to see the model addressing current questions of ecosystem dynamics and management soon. To highlight just one activity: iLand is currently used to assess biodiversity – ecosystem functioning relationships across forested landscapes within the frame of the EC FP7 collaborative project FunDivEUROPE.

But of course we also have plenty of ideas on how to further improve the model, including (but not limited to) adding to the range of disturbance agents (currently modules for wildfire and wind are operational, with a bark beetle module being in the planning stage) and thinking about spatially explicit competition for resources other than light. To conclude this post with another quote of the proposal review: “This project should be considered as a first step in the development of a new analysis/ modeling tool; […] this fellowship is the beginning of a longer-term effort in forest ecosystem model development”. I couldn’t have said it better myself - the “fellowship of iLand” is well on its way to new quests (and we’d be more than happy to hear from you if you’d want to join us on this journey, or have ideas with regard to its direction).

We're delighted to let you know that after more than three years of research, discussion, coding, and testing iLand is finally published. A paper describing the core components of the iLand approach with regard to individual-tree competition for resources, growth, and mortality, and demonstrating the models' ability to simulate both even-aged and complex stands, has recently been published in Ecological Modelling. In line with our open source strategy it'll also become "open access" soon, i.e. available to everyone (with access to the internet).

In parallel to this peer-reviewed publication we have also published the model code and software version that was used in the simulation exercises presented in the paper. You can download the package here, and since we even supply some example simulation files running iLand on your machine is only a click away.

But there's even more. Despite publishing a 36 page appendix alongside the paper, to describe our model logic in more detail, we have also made the respective iLand wiki documentation pages public over the last days. On this 70+ pages you can browse interactively through the iLand world, search for key words you're interested in, or link yourself directly to relevant references we relied on in developing iLand.

For us this is a great moment (and we'll go celebrating it later this week ;-). Let me take this opportunity to thank everybody who contributed to making this possible, particularly the amazing colleagues who worked with us on iLand over the last three years, but also the funding sources making all of this possible, above all the EC's FP7 Marie Curie program.

So is this it? No, quite the contrary. For us, this is only the beginning. We're currently working on a number of additions (think disturbance modules) for iLand, so iLand development is far from finished (hence also the v0.3 of the currently public model version). And, of course, we didn't develop the model for the sake of model development itself (although that's a fun and interesting process, I admit): We're currently conducting a number of studies applying the model to a variety of questions, including an investigation into the drivers of C storage in complex mountain forest landscapes, the interactions between disturbances and forest vegetation, and the link between biodiversity and ecosystem functioning.

Publishing a model – what it is*

Put the executable online, says the (potential future) user
Put the code online, says the fellow model developer
Write a paper about it, says your tenure committee
Put the technical documentation online, says the PhD student
Write a paper about it, says the evaluator of your next proposal
Write two papers about it (because there is too much information), says Reviewer #1
Write a paper about it (but in a different journal), says Reviewer #2
[Nothing], says the subject editor (and forgets the paper on a pile somewhere for half a year)
Blog about it, says the grad student
Do all of the above, says the model

On the eve of finally publishing the core conceptual approach of iLand, I’ve been pondering quite a bit what “publishing a model” actually means. Of course, peer-reviewed journal publications are the currency of our “publish-or-perish” science world, but are they really the best way publish (as in: to make what you’ve developed available to the community) a simulation model? From personal experience I’m inclined to say “no”, or at least “yes, but not exclusively”, because of the following factors:

• Time: It takes on average a year to get a paper published from the first submission to its final publication. Add three to six months for writing and internally reviewing the manuscript before submission, and six to 24 months (depending on the complexity of the endeavor) for model development and testing, and you’ll end up taking between 21 and 42 months (that’s a range of 1.75 to 3.5 years) between having an idea and communicating it with your peers. Needless to say there’s a lot of progress in 3.5 years, particularly in a strongly technology-driven field such as modeling, i.e. your initial problem might already be obsolete (e.g., technological advances might have relaxed prior limitations), somebody else might have put out a similar idea in the meantime, etc.

• Level of detail: More and more journals put page restrictions on papers, and more and more reviewers demand the lean studies that we’ve all gotten so used to in this day and age of the “least publishable unit”. So you basically end up having to gloss over a lot of details to publish a complex model in a journal article. Or, if you’re like me, and have been trying to re-trace the models of other people from their papers A LOT, you end up writing online appendices that are (almost) as extensive as the original paper. In general, these online appendices are a good opportunity to communicate more detail about a model without hampering the narrative of the paper. But let’s be honest, they are rarely looked at with the same rigor by reviewers, are even less read by your random reader, and are not at all included/ considered in the current system of quantifying the value and quality of a contribution (think number of downloads, citiations, etc.).

• Utility: So even if you do end up writing a detailed online supplement (somebody once told me I’m wasting a perfectly good second paper on an Appendix), it is in many cases still not enough to make your modeling work really useful. First, some of the details only reveal themselves in a technical documentation or, even better, in the code. Second, if you’ve been sold by a convincing paper and want to try the model, an appendix doesn’t really help you. What you’ll need in the latter case is an executable, a documentation of inputs and drivers, and ideally some templates and examples for you to just hit “play” and toy around a bit to get an idea about what the model actually does…

• Visibility: We scientists oftentimes tend to reduce visibility to the impact factor of a journal. But let's face it, if we were to do an initial search on a subject, even we'd start out by using everybodys tool of the trade for searching the world (wide web). Particularly for simulation modeling platforms, which are designed to be applicable over a wide range of conditions and to a wide range of questions (and would thus (at least in theory) be of interest to a wide range of people - a major difference to case-specific empirical models), there should be more to it than a proprietary measure of journal importance. What I'm saying is: A paper in a well-esteemed journal of course doesn’t hurt for a models visibility, but it’s the 21st century (and I’ve lost track if its web 2.0, 3.0 or any other version out there currently). Foreign affairs cables are being blogged. Revolutions are being televised. The news of the day are being tweeted from every corner in real time (and thereby transforming the course of actions). So why not use these channels also to spread the word on your latest model? To engage into a discussion with the world beyond the selected readers of the big syndicated publishers? To enable a discussion of ideas more immediately and directly?

You guessed it already, but addressing these issues is what we’re aiming for with the “model publication strategy 2.0” we’re pursuing with iLand. To have high quality papers in esteemed journals. To have the code and documentation available only a mouse click away. And, of course, to blog about it (case in point) – so stay tuned.

*modified from Erich Fried. All characters and statements are completely fictional and are not ment to be a faithful reproduction of our publishing experience with iLand.