HJA newt logoHJ Andrews
Login Donate
  Home > Research > LTER Program > LTER6 Focus Areas > Future Change > Research Summaries


The individual-based forest landscape and disturbance model (iLand) was recently developed by Rupert Seidl, an Andrews LTER collaborator, to address how changing climate and disturbance regimes might influence forest ecosystem dynamics and consequently the provision of ecosystem services and is thus a highly relevant tool for the research questions at HJ Andrews under LTER 6.Forest ecosystems are modeled from the perspective of complex adaptive systems in iLand (Seidl et al. 2011a, Seidl and Rammer 2011), with ecosystem dynamics an emerging property of interactions between agents and processes across multiple scales. The core agents modeled in iLand are individual trees (Grimm et al. 2005). Their spatially explicit competition for light, water, and nutrients is simulated based on ecological field theory, accounting for each individual's ability to locally compete for these resources (Figure 4, cf. also Berger et al. 2008). Generalized physiological principles are applied to derive tree growth and mortality from these captured resources. iLand applies a radiation use efficiency approach to derive primary production (Medlyn et al. 2003). Response functions to daily weather conditions are used to account for environmental effects on resource utilization efficiency. The model furthermore employs a cascading sequence of allometric ratios to calculate allocation to tree compartments (Landsberg and Waring 1997), with environmental factors affecting the allocation to root vs. shoot biomass as well as to height vs. diameter growth. Utilizing an individuals' carbon budget, tree mortality is simulated probabilistically for trees experiencing carbon starvation (Guneralp and Gertner 2003). The fate of dead organic matter is tracked in a decomposition module, that distinguishes standing and downed deadwood, litter, and soil organic matter pools (Katterer and Andren 2001, Magnani et al. 2007). iLand thus simulates process-driven estimates of the C exchange between forest landscapes and the atmosphere.

Figure 4: iLand is a dynamic, process-based forest landscape model. The left panel illustrates the light interference patterns modeled for every individual tree, and their aggregation to a continuous field of light competition in every timestep in the model. The right panel highlights the main processes modeled in iLand and their respective scales. Source: Seidl et al. (2011a).

To scale from individual trees to forest landscapes, iLand simplifies the competitive influence between trees to size- and species-specific interference patterns pre-computed via ray tracing (Canham et al. 1988). It furthermore harnesses a hierarchical multi-scale approach, in which higher level processes (e.g. water availability at the stand scale, disturbances at the landscape scale) constrain lower level dynamics (e.g. growth at the level of individuals) (Wu and David 2002). This model design allows us to simulate individual-based forest dynamics at the scale of the HJ Andrews watershed in a computationally efficient manner. At the landscape scale, modeled spatial processes include seed dispersal and disturbance processes, the latter modeled by means of a cellular automaton approach. Spatially explicit seed dispersal kernels are used to calculate seed distribution over the landscape (Lischke et al. 2006), and a species' success in establishing at a new site is calculated using a phenology-based approach (Nitschke and Innes 2008). Spatially explicit wildfire and windthrow modules are currently in development (Seidl et al. 2011b).

To rigorously test a multi-scale simulation model such as iLand, the variety of documented long-term datasets at the HJ Andrews has proved invaluable. To evaluate model performance from the individual tree level all the way to the landscape level we made use of the HJ Andrews long-term vegetation plot data, soil inventory, detailed climate data, disturbance history information, and Lidar data. The conducted suite of tests showed that iLand's scalable approach to model individual-tree competition was able to simulate the complex light competition regime in old-growth stands at the HJ Andrews (Figure 5). They furthermore revealed the ability of the model to simulate forest C cycle processes (Figure 6), and the spatial distribution of species within the landscape (Figure 7) with satisfactorily accuracy. Addressing Goal I within the LTER 6 proposal, iLand is currently applied to unravel the drivers of spatial heterogeneity in the carbon storage at the HJ Andrews. Due to its high spatial and process resolution iLand is an important addition to the HJ Andrews simulation model arsenal in addressing this issue. In the near future, the model will also be used to address the impact of climate change on vegetation dynamics at the HJ Andrews (Goal II under LTER 6). iLand was developed by Rupert Seidl (supported by a EU Marie Curie Fellowship) in close collaboration with HJ Andrews PI Tom Spies, with further input from collaborates at the University of Natural Resources and Life Sciences (BOKU) Vienna, Austria, Portland State University, Portland, Oregon, Oregon State University, Corvallis, Oregon, and the HJ Andrews community.

Figure 5: Observed (dark grey) versus simulated (light grey) diameter distribution for HJ Andrews old-growth reference stands 20 (upper panels) and 22 (lower panels) at the end of the 22 to 24 year observation period. Boxplots indicate the species-specific individual-tree diameter increment residuals. Source: Seidl et al. (2011a).

Figure 6: Comparing observed carbon storage in HJ Andrews reference stands of different vegetation zones to iLand model results after a 500 year undisturbed model run. Observed reference data are from Smithwick et al. (2002).

Figure 7: Example for the evaluation of the simulated species composition at the HJ Andrews experimental forest. Simulation results from the individual-based forest landscape and disturbance model (iLand) after a 500 year simulation run are compared to a the gradient nearest neighbor (GNN) imputation of inventory data (Ohmann and Gregory 2002) for the high elevation species Abies amabilis.