iLand News

I guess that’s a question most if not all modelers have to deal with at some point. Because quite contrary to the first intuition it’s not making a model more complex that’s challenging but vice versa (see for instance David Mladenoffs notion on the issue here). A model is per definition a simplified representation of reality, so modeling is always about reducing a complex reality into something more traceable, analytically solvable and computationally implementable. The gentle art is though to find the level of complexity that’s sufficient to address the system traits relevant for the question at hand (NB that every model is context-specific, so there is no universal ‘world model’, as it would have to be as complex as the ‘world’ itself and would thus be no more simplification of reality… [as a sideline, here is a nice piece about the utopian attempt to create exactly that, a full-blown model of the biosphere, and where it stands today after 25 years and 200 million $... but I’m deviating]).
So the decision about complexity is probably the single most important task in modeling, but what is the right level of complexity? Albert Einstein maintained that a model should be as simple as possible, but not simpler. As good as this axiom sounds, where does it leave us? Recently some colleagues used modular model designs to scrutinize this question quantitatively, i.e. add levels of complexity and scrutinize the influence on model behavior with regard to certain aspects of interest. In iLand we’ve been putting exactly this question of complexity at the core of model development.
Complexity in ecosystems is more and more recognized as not just another trait or way to describe systems. It is essentially at the core of their functioning (as has been recently shown by colleagues here at OSU). Consequently, complexity is even recommended to be a central issue in our management considerations; “managing for complexity” is what Klaus Puettmann and colleagues see as the main challenge for a modern silviculture. To develop a model that supports this task, we need to embrace complexity and its functional, structural and spatial dimensions in ecosystems. The iLand model design is inter alia motivated by the limitations of previous approaches to address these three levels of complexity simulatenously in a dynamic modeling framework (more on context and approaches in the review contained in the first iLand-related publication)… Stay tuned on how we tackled this challenge and why we think it is important!

Computers are getting faster and faster and memory is hardly ever a problem any more- and this technical fast track is definitely an important factor in making a project such as iLand possible.

But despite its headline this post is not about technical limitations of modeling. It is about our own limitations in dealing with the larger amounts of information that we are digesting from day to day. In the context of a modeling project- we're now 6 month down the road, and important parts of the basic model design are pieced together and already implemented. And although there is still a lot ahead of us and six months is not a very long period we find it increasingly important to organize and document the thinking, research and development that has been going on over the last month. Or, in other words: in modeling (and many other things) it is often all about the details, but our own limitations make us forget these details rather quickly, once we turn our attention to other things.

In order not to loose this detail, and to conserve the thoughts and reasoning we arrived at in working on the model, we put quite an emphasis on documentation and communication in our project. A centerpiece of this effort is this webpage. The project collaborators section by now contains 72 wiki pages as of today, documenting what we've been working on. The beauty of the wiki system lies in two aspects for this task:

• First, documentation as we see it is a continuous process rather then something you write once the model is "done". The reason for that view is, among other things, the above mentioned inability to keep all the details and reasoning present. What the wiki allows us is to start a documentation parallel to development, and track all changes that might happen down the road. As the project progresses we'll not just have the current state properly documented but also the development that led to this state.
• Secondly, lo and behold the beauty of hypertext. The wiki documentation currently contains more than 100 unique references to scientific literature, embedding the model concept into its scientific context, linking to further information on technical details or giving reference to previously developed concepts that we adopted and adapted in iLand. But rather than an ordinary reference list, the hypertext allows the reader is only one click away from the articles fulltext , opening up a new dimension of reading (thanks to such handy tools as the doi.)

In addition to this conceptual level of documentation, software versioning and documentation is evolving as the model code grows. In iLand we use the software version control system Subversion and the documentation system Doxygen. Code documentation frequently links to the description of the concept in the related wiki pages, coherently adding another level of documentation. Overall, this system allows us to extract, retrace, and analyze any state (previous or current) in the model development.

And yes, it takes some effort to engage in this continuous documentation. However, we feel that this pays back easily in not having to laboriously re-construct previous thoughts and decisions a few month down but having them readily at hand for re-evaluation, extension or adaptation. Furthermore, such a comprehensive suite of model documentation can significantly enhance the value of a tool for potential users and collaborators in the future.

How trees capture radiation energy, how efficient they are in converting it to carbon, what environmental factors (and interactions among factors) are influencing this process, for which organs (e.g., leaves, roots, stem) the produced biomass is used… these are just some of the questions addressed by tree physiology. Rather than approximating the process of tree growth statistically relying on models based on physiological principles is important particularly since such models are more robust in predicting the effects of climatic change on ecosystem processes.
A large variety of models based in physiology exist, the challenge in iLand is, to incorporate physiological knowledge on growth processes in a balanced way with other ecosystem processes (e.g., mortality, regeneration), addressing the level of individual trees but keeping it scalable to watershed and landscape scale. Currently, the idea is to use a radiation use efficiency approach in iLand, which calculates the canopy carbon gain based on intercepted radiation and an efficiency measure of converting this radiation into primary production, taking constraints by environmental factors explicitly into account. The seminal paper of Landsberg and Waring (1997), for instance, presents such a model, and their approach has also proven to be suitable in an ecosystem dynamics context (i.e., hybridizing physiological and successional concepts of forest modeling). I had the opportunity to discuss prospects and details of such an approach recently with Dick Waring here at OSU, and it seems that one more piece of the (iLand) puzzle slowly falls into place… I’m trying not to get too excited yet (as model building is an iterative process and we’re still in the first iteration loop), but we’re expecting the first trees to grow in our virtual environment in the near future!

Mortality (for this posting: tree death due to causes other than disturbance and management) is a very important process in forest ecosystem dynamics- yet our understanding of causes and triggers of individual tree mortality is still poor. Despite all efforts, we’re mostly limited to recording a trees’ demise, for instance in inventories or from remote sensing data, but its cause of death remains widely unknown. Consequently, modeling the process of tree mortality has received much less attention than, for instance, growth, and models have been mostly “well designed”, lacking both detailed process representation and empirical underpinning. Whereas the latter has been addressed in recent work, notably of colleagues at the ETH, the process-based design of iLand enables us to strive for a simple yet process-oriented representation of tree death. We’re currently looking into using a trees’ carbon balance, i.e. how well a tree is able to sustain its living tissue, as an indicator of its stress and predisposition to mortality. In other words, if the “spending” of a tree for supporting vital organs such as leaves and roots exceeds its gains from photosynthesis, it has to draw on its reserves and gets increasingly prone to pressures (e.g., pathogens), eventually succumbing to them. In iLand, such a C balance indicator is a process-oriented proxy of a trees’ environmental influences, its competitive status and life history- and thus a good integrating indicator for mortality. Yet, accounting for the limited experiences with such approaches, we’re currently looking into a probabilistic design model design, in response to the still limited understanding of the variability in observed mortality patterns. We’re quite eager to test such assumptions against both hypotheses and data, as mortality eventually frees resources for regeneration, thus being essential in the cycle of forest dynamics.

One particular vision and objective with iLand is to integrate the ideas and concepts of individual based, stand level forest modeling with those of landscape ecology and modeling. To facilitate that idea we are currently holding a workshop with the core LANDIS-II developers Rob Scheller and Jimm Domingo in Corvallis. Discussing landscape issues (e.g., spatial interactions) both at the conceptual/ecological level as well as at the technical/implementation-oriented level gives us the unique opportunity to learn from the long experience of the LANDIS-II team and ensure compatibility with regard to certain aspects of the highly modular LANDIS-II structure, e.g. to design interfaces for compatibility with LANDIS-II extensions. Their newly extracted spatial library (available here), but also their approaches to model spatio-temporal dynamics of seed dispersal, forest management and disturbances in particular are highly relevant for our work in iLand. Rather than re-inventing the wheel wrt these processes the cooperation with the LANDIS-II team allows us to tap into their knowhow, bringing us a step closer to actually bridging the gap between individual based ecology and landscape ecology. And besides, working with them is really inspiring and fun…

The summer high pressure has settled over the Pacific Northwest, temperatures are well in the 90ies (30°C) and campuses here at OSU and elsewhere are getting pretty empty… yet a quite summer is not whats on our minds as we dive deeper into the development of iLand. Werner Rammer, the principal technical developer of iLand, has arrived here in Corvallis for a summer of model development and we’ve started to structure the development cycles ahead of us (see an illustration here). Currently we’re spending some time over the question how individual trees compete for resources (light, water, nutrients) and how such processes can be efficiently abstracted for implementation at landscape scales. Our initial efforts to derive a process-oriented field of neighborhood patterns of individual tree interactions look quite promising… and is constantly exposed to real world comparisons as we hike through the stunning forests of the Pacific Northwest on the weekends. Speaking of which- happy Independence Day, we’re now going BBQ…

Around the official project start date, April 1st 2009, we held two kick-off meetings in order to get the project team together and the work on the project finally started. On March 30th the Austrian-based part of the project team (still including myself, at that date) gathered for a project kickoff, reviewing the proposed task and work structure and discussing approaches potentially relevant for the modeling tasks at hand. Having two weeks to digest this impulse (and also to get established in the lovely town of Corvallis, Or) iLand work started here in Oregon in mid April. The second half of the kick-off round together with Tom Spies and Rob Scheller was held soon thereafter, on May 6th. Overall, discussion and comments were very stimulating and diverse – and I’m looking forward to catalyzing some of them in bringing together ideas from individual-based population dynamics and landscape ecology through the project. Stay tuned for more…