LDSim (Landscape Disturbance-Succession Simulator) is a disturbance and succession model used to understand and predict changes in forest ecosystems over time. It is a process-based model that incorporates information about forest disturbances (such as fire, windthrow, and insect outbreaks) and the subsequent recovery and succession of the forest.
The model was developed by Dr. Kevin McGarigal and his colleagues at the University of Massachusetts Amherst and is based on the premise that natural disturbances and succession processes are important drivers of forest ecosystem dynamics. The model uses a spatially-explicit approach, meaning that it simulates forest changes at multiple scales of spatial resolution (for example, from the site to the Subbasin scale) to meet different management needs and use cases.
LDSim modeling begins with an assessment of the historical range of variability (HRV) in the forest ecosystem, which is the range of natural variation in ecosystem structure and function that has occurred over a long period of time. This provides a baseline for evaluating the degree to which current forest conditions have deviated from historical patterns. The model then simulates the occurrence of various disturbances, such as fire, and predicts the subsequent recovery and succession of the forest over time.
LDSim modeling can be used to assess the effects of different management interventions, such as prescribed burning or thinning, on forest ecosystem dynamics. It can also be used to evaluate the potential impacts of future disturbances, such as changes in climate or land use.
Overall, LDSim disturbance and succession modeling is a powerful tool for understanding and predicting changes in forest ecosystems over time and guiding management and conservation efforts to maintain or restore more natural conditions.
LDSim helps us better understand the complex dynamics of forest ecosystems and provides a tool for managers to make informed decisions about how to manage these ecosystems to meet their goals. For example:
The LDSim model is designed to simulate the long-term dynamics of forest ecosystems, which is important for understanding how forests change over time and how they respond to disturbances such as fire, insect outbreaks, and human activities like logging. This allows managers to assess the impacts of different management scenarios on forest ecosystems and make informed decisions about how to achieve their objectives.
The LDSim model is a landscape-level model, which means that it accounts for spatial variation in forest characteristics and disturbance regimes. This is important because disturbances like fire and insect outbreaks often occur at a landscape scale, and the effects of these disturbances can be influenced by factors like topography, climate, and soil properties.
The LDSim model can be used to test different hypotheses and theories about how forest ecosystems respond to disturbances and other factors. By comparing model outputs to real-world data, researchers can refine their understanding of the mechanisms underlying forest change and develop new theories and management approaches that better account for these dynamics.
The LDSim model provides a framework for forest managers to assess the impacts of different management scenarios on forest ecosystems.
LDSim data can provide a reference point for how well treatments may be working when the objective is to return a landscape closer to its Historical Range of Variation (HRV). LDSim was used to demonstrate that it is possible to return to a HRV within 100 yrs with active management, or much sooner with more extensive and intensive management, as was shown by McGarigal et al (2018).
Historic Range of Variability (HRV)—HRV is an essential concept in LDSim disturbance and succession modeling. As mentioned above, HRV refers to the range of natural variation in ecosystem structure and function that has occurred over a long period of time, typically several centuries to millennia. This is important because it provides a baseline for evaluating the degree to which current forest conditions have deviated from historical patterns.
Departure data is important in LDSim disturbance and succession models because it allows researchers and managers to assess the degree to which current forest conditions deviate from historical vegetation patterns. Departure data compares the current state of the forest with the state that would be expected if natural disturbance and succession processes had been allowed to operate over the long term. This comparison is important for several reasons:
LDSim is a specific version of a Landscape Disturbance and Succession Model (LDSM), which is a type of computer model used to simulate the long-term dynamics of landscapes. There are several features that make LDSim unique from other LDSMs:
As of March, 2023 (the time of this writing) the analysis covers a portion of the northern Sierra Nevada ecoregion, and corresponds to the extent of the Tahoe Central Sierra Initiative (TCSI). It currently comprises six focal landscapes that were defined primarily on the basis of Hydrologic Unit Code (HUC) 8th-level watersheds.
(Extract from Northern Sierra Historical Range of Variability and Current Landscape Departure Executive Summary)
“We chose to quantify the Natural Range of Variability (NRV) in landscape structure for the selected focal landscapes based on the simulated HRV in several vegetation attributes relevant to forest management. To simulate disturbance and succession processes representative of the historical reference period within the ecoregion, we developed the landscape disturbance-succession simulator LDSim [emphasis added]. LDSim is a raster-based, spatially-explicit, stochastic simulator of disturbance and succession processes at the cell level (in this case, 5 m resolution), and is mostly a phenomenological process simulator that seeks to emulate the statistical properties of disturbance and succession processes consistent with the statistical properties of real-world observations.
LDSim simulates succession at the beginning of each timestep in the simulation, representing the establishment, gradual growth and/or development, and eventual senescence of vegetation over time. Disturbance follows succession in each timestep in the simulation and is implemented as a generic disturbance algorithm that can be meaningfully parameterized to represent a variety of natural disturbances, including fire, insects/pathogens and wind, although only fire was simulated in this assessment [emphasis added]. Disturbance is implemented as a multi-step process involving climate control, disturbance initiation, spread, termination and mortality, in which each step is optionally constrained by climate (and weather proxies), geophysical setting and vegetation conditions.
LDSim requires a variety of spatial inputs, including cover type, vegetation age and developmental stage, and a variety of geophysical variables. In addition, LDSim has a user-defined set of model parameters that govern the disturbance and succession processes for any particular user-defined application and scenario. Importantly, the parameterization of the disturbance and succession processes represents the set of user assumptions about the system being modeled, and the results are only meaningful in reference to this set of assumptions [emphasis added]. In addition, the parameterization of the model is application- and scenario-specific. For this assessment, we parameterized the model to reflect our best, empirically-based understanding of the succession and wildfire disturbance processes characteristic of the historical reference period in the region.Although LDSim is a process-oriented model with individual parameters sourced either from data, literature, or expert opinion, the model outcomes reflect the complex spatio-temporal interactions of the various stochastic disturbance and succession processes. Consequently, it is necessary to calibrate the model by tuning (i.e., adjusting up or down) one or more of the model parameters to produce selected outcomes that are consistent with the data, literature and expert opinion. Importantly, in this context, we considered the parameters associated with the disturbance process to be “independent” variables and the vegetation conditions (e.g., seral stage distribution, spatial patterns of vegetation states, etc.) to be “dependent” outcomes [emphasis added]. We calibrated selected disturbance parameters (i.e., independent variables) to achieve the target disturbance regime without any attention to the vegetation response, which we treated as the dependent outcome of the disturbance regime. Importantly, we did not tune the disturbance regime parameters to achieve desired vegetation outcomes. Specifically, we focused model calibration of five disturbance regime attributes that we considered emergent properties of the disturbance regime: (1) Fire Rotation Period (FRP) by cover type, (2) Fire Mortality Rate (FMR) by cover type, (3) fire size distribution, (4) variability in the total area burned over time, and (5) strength of the relationship between total area burned and climate [emphasis added].
For each focal landscape we calibrated the model to achieve target performance criteria for each of these emergent properties. For each focal landscape, we simulated succession and wildfire disturbance under historical reference period conditions (circa 1550 - 1850) for a 1,500-year period with a 5-year timestep (i.e., 300 timesteps) and output several spatial data layers representing disturbance and vegetation attributes at each timestep. We dropped the first 500 years or 100 timesteps as a liberal estimate of the model equilibration period (i.e., the time to reach dynamic equilibrium) and retained the remaining 1,000 years or 200 timesteps to quantify the range of variability in various landscape attributes (see below). Note, although we simulated landscape dynamics for 1,500 years and retained the last 1,000 years for the analysis of HRV, it is important to recognize that this does not conflict with our ~ 300-year historical reference period. We parameterized LDSim to represent disturbance processes characteristic of the 300-year historical reference period; the longer simulation merely allowed us to produce a larger sample of landscape snapshots that are characteristic of this historical disturbance regime.
It is absolutely essential to interpret the model results in the context of the model parameterization, which is essentially a set of assumptions about how disturbance and succession processes operate at a particular scale in the focal landscape under historical reference period conditions. The choice of scale in representing the landscape is a user assumption that is especially important, as the simulated range of variability and current departure in landscape structure is highly scale-dependent [emphasis added]. In other words, the measured variability and current departure depends on the grain and extent of the assessment in both the spatial and temporal domains. Thus, it is imperative that both the spatial and temporal scale, in terms of both grain and extent, of the assessment be made explicit. In this regard, we fixed by design the spatial grain of our assessment at 5 m – the approximate scale of a single mature tree – and the temporal grain at 5 years [emphasis added]. In addition, we fixed by design the temporal extent of our assessment at the ~ 300-year historical reference period (circa 1550 - 1850). The spatial extent of our assessment, however, was not fixed by design, and we chose to describe HRV and current departure at a few different scales, or rather “levels” in the hierarchical organization of landscape mosaics: (1) site level, (2) geophysical unit level, and (3) Subbasin level, as defined below. Importantly, any attempt to compare our results to those generated at a different scale should be done with extreme caution or, preferably, not at all [emphasis added].
Lastly, to summarize HRV and current departure in landscape structure at the Site and Subbasin levels (see below), we subdivided the landscape into a set of Biophysical Classes (BPCs) based on simulated site productivity [emphasis added]. Specifically, we modeled simulated cumulative tree growth (or biomass) as a function of a suite of geophysical predictors and classified predicted cumulative tree biomass under historical dynamic equilibrium conditions into four equal-area BPCs (plus a non-forest class), as depicted in Figure 2. We posited that cumulative tree biomass under historical dynamic equilibrium conditions represented an index of realized site productivity that effectively integrated the net effects of all the disturbance and succession processes in LDSim, including, in particular, the frequency and severity of wildfire, the rate of tree establishment and the rate of tree growth after establishment. It is important to acknowledge that BPCs were defined within each focal landscape independently; thus, the BPC classes were relative to the site productivity gradient within each focal landscape [emphasis added]. We created BPCs as a means of “packaging” the results in a manner that will be most relevant to management at the project level and would best discriminate differences in the landscape metrics, given the assumption that cumulative tree productivity relates to vegetation characteristics of major interest to land management (e.g., tree cover). We also sought a parsimonious suite of BPCs; i.e., the fewest number of BPCs that would best characterize meaningful differences in the landscape metrics.”
Don’t interpret the numbers as the truth; instead, focus on the comparative results (e.g., HRV versus CRV).
(Extract from Northern Sierra Historical Range of Variability and Current Landscape Departure Executive Summary)
Because this project relies on the use of computer models, it is paramount to understand the scope and limitations of the modeling up front.
We use a phenomenological modeling approach; therefore, we limit our simulations of wildfire disturbance to reflect the historical and modern fire record. This contrasts with a “mechanistic” modeling approach that tries to simulate the actual physical, chemical or biological mechanism of the process, such that the outcomes are emergent properties of the mechanism governing the process (Gustafson et al. 2013). Of course, there are advantages and disadvantages to both approaches. The major advantage of the phenomenological approach is that it does not require a complete understanding of the mechanisms associated with the processes, requires far fewer model parameters, and can be more easily parameterized to reflect the limited observations of real-world behavior (that are more often statistical rather than mechanical). The major disadvantage is that the algorithms can be somewhat arbitrary, and thus may not have an intuitive ecological interpretation or can be viewed as somewhat of a “black box”. The major advantage of the mechanistic approach is that, if the mechanisms are well understood and parameterized correctly, it allows for the projection of landscape changes under novel environmental conditions (e.g., future climate) for which we have no observational data. However, if the mechanics are poorly understood or parameterized incorrectly, the projections will be inaccurate and can be grossly misleading. Because we use a phenomenological modeling approach, we limited our simulations of wildfire disturbance to reflect the historical and modern fire record. In particular, we do not project wildfire disturbance under future climate conditions.
Model results should be viewed as “fuzzy” estimates, not as exact answers. While models have many uses and advantages over strictly empirical studies, they are fundamentally abstract and simplified representations of reality. This is especially true for LDSMs. The processes that drive real landscapes to change are far too many and complex to model comprehensively and accurately. Therefore, our goal in modeling these systems is to capture the most important drivers well enough that our results, in a very general sense, reflect our real world expectations. Although it is tempting to view the results as exact, given their quantitative nature and apparent numerical precision, we should strongly resist over-interpreting them. Moreover, because the results may not be accurate in an absolute sense, they may nonetheless be reliable in a relative sense, especially when the focus is on comparisons. Specifically, whereas the absolute predictions may not be accurate, if the processes are parameterized and calibrated well, and implemented consistently in the model, then the relative results may still be reliable, especially when used to compare among scenarios. For example, in the context of this application, the estimate of absolute departure for a unit of ground may not be accurate, but the relative degree of departure of one unit compared to another may be quite reliable.
As long as the model gets it right most of the time, it still can have great utility. The results of our model, as with any model, are constrained by the quality of input data, which are not perfect. For example, the vegetation cover layer is subject to human interpretation errors and objective classification errors, and is further limited by the spatial resolution of the grid. Consequently, there will be places where the model gets it wrong, not necessarily because the model itself is wrong, but rather because the input data are wrong. Getting it wrong in some places, however, should not undermine the utility of the results as a whole. In the end, the results should be used and interpreted with the appropriate degree of caution and an appreciation for the limits of the available data.
Model results are subject to change as new scientific understanding or better data become available. The LDSM we used here (LDSim), like all landscape disturbance-succession models, requires substantial parameterization before it can be applied to a particular landscape. To the extent possible, we utilize local empirical data. However, we also draw on relevant scientific studies, often from other geographic locations, and rely heavily on expert opinion when scientific studies and local empirical data are not available. Despite our best efforts to incorporate the most relevant data and scientific findings, it is clear that we have a very limited understanding of the disturbance and succession processes being modeled. This does not undermine the utility of the results. The key question is whether the results lead to more informed and thus better decisions. Moreover, the model should lead to new insights that might at first seem counter-intuitive or inconsistent with our limited observations, because the model is able to integrate a large amount of data over broad spatial scales and long time frames in a consistent manner and thus provide a perspective not easily obtained via direct observation.
Model results reflect landscape dynamics as driven by the selected succession and disturbance processes — in our case, wildfire. In this application, the results do not reflect the influence of other disturbance processes (e.g., insects and pathogens, wind-throw, ungulate and beaver herbivory, avalanches, and other forms of soil movement, etc.) or all of the complex interactions among them that characterize real landscapes and drive the full range of variability. Other kinds of natural disturbances also occur in the region, as noted above, but the impacts of these other natural disturbances tend to be localized in time or space and have far less impact on vegetation patterns over broad spatial and temporal scales than does fire. Thus, while the model is incomplete in the strict sense, it can still provide useful results because it reflects the dominant landscape change process.
Model results pertaining to individual cover types should be interpreted with caution for cover types having limited extent within the focal landscape. The Sierra Nevada vegetation is extremely diverse and complex in its spatial arrangement of ecological settings and conditions; accordingly, the accuracy of mapping each unique ecological setting (i.e., cover type) and condition varies considerably. In general, because the model results are statistical in nature, confidence in the model results should decline as the extent of a cover type declines. Consequently, we limit our interpretation of the results to cover types that extend across > 1,000 ha of the focal landscape.
Lastly, extrapolating model results from one landscape to another should be done with extreme caution. As noted above (Axiom 2), landscapes are idiosyncratic; in other words, they have a unique internal structure and history, such that no two landscapes are the same. Thus, extrapolating our findings for, say, the Yuba River watershed to other landscapes should be done with caution. Our general findings certainly pertain to other similar surrounding landscapes and can probably be safely extrapolated to neighboring landscapes to some extent. However, to the extent that detailed quantitative results are desired for other geographies or ecoregions, a separate modeling exercise should be undertaken.
Don’t interpret lines on the map as discrete and real; instead, all lines are fuzzy classifications of what is in reality continuous variation and are only delineated to facilitate packaging results.
The primary geospatial data derived from the LDSim model includes Subbasins, Geophysical Units (GPUs), Biophysical Classes (BPCs), and two historical wildfire raster datasets: Fire Return Interval and High Mortality Fire Return Interval.
There is no single “right” scale/level for NRV assessment; therefore, a multi-scale/level approach is strongly warranted
The LDSim data is designed to reflect the dynamic and scale-dependent nature of land processes, and it achieves this through geospatial data layers at different scales. Two important components of the model are the GPU and Subbasin units, which provide valuable information on changes in vegetation patterns from historical to current-day conditions at two different landscape scales.
Both the GPU and Subbasin units offer a range of attributes that explain the extent of these changes, with each unit being suited for a particular scale of analysis. The GPU units are best for targeted treatments at the stand level, with an average size of around 50 acres, while the Subbasin units are larger and more appropriate for identifying areas for larger treatment projects such as a NEPA project, averaging about 15,000 acres.
In addition to their primary purpose, the GPU and Subbasin units also provide valuable data on various landscape patterns and processes. These include the occurrence of large trees, fire return intervals, and open forests, which are relevant for the respective scales of analysis.
Furthermore, both units offer a composite departure score that provides an overall assessment of the extent of deviation from historical conditions across all variables. This score can be used to identify areas that require targeted treatments or larger-scale restoration projects, based on the degree of departure from historical conditions.
The composite negative departure score at the GPU-level is the result of combining and weighing the eleven individual departure metrics outputted by LDsim within each GPU unit. The final numeric values represent the percentage of maximum departure of current landscape metrics from the GPU’s historical modeled range of variability (HRV). Larger values indicate increasing degrees of departure.
The Percent of Large Trees (>30in dbh) Departure Index data identifies Subbasins and BPCs where fewer or greater numbers of large trees are present within the landscape relative to HRV. The departure index reveals the level and degree of departure from the historical range: yellow to red values indicate that the current landscape has fewer large trees than the HRV. Conversely, the range of green values show more area in large trees than the median HRV. To this point, the map above highlights that most of the Yuba River landscape has far less area in large trees than would be expected under the historical reference period.
The Biophysical classes, on the other hand, partition the landscape into a categorical classification of four predicted cumulative tree biomass levels under historical conditions and rate them from high to low:
The data is necessary for computing the departure at the site (not included in this package, contact Kevin McGarigal to inquire about access) and Subbasin scales since a historical biomass baseline is essential to compare to the current range of variability (CRV). Note that this differed from how the departure scores are calculated for GPUs. As the General Technical Report outlines, “Vegetation conditions are not considered in delineating GPUs so as to not bias the HRV and current departure results. Importantly, LDSim compares the current condition of a GPU to its own HRV because it is sufficiently large enough to do so. Thus, we derive an HRV and compute departure for each GPU independently (in contrast to the Site-level HRV and departure).”
Results are only as good as the user assumptions in the model; thus, interpret results in the context of the model parameterization and calibration.
