Corralling a black swan: natural range of variation in a forest landscape driven by rare, extreme events

Abstract

The natural range of variation (NRV) is an important reference for ecosystem management, but has been scarcely quantified for forest landscapes driven by infrequent, severe disturbances. Extreme events such as large, stand‐replacing wildfires at multi‐century intervals are typical for these regimes; however, data on their characteristics are inherently scarce, and, for land management, these events are commonly considered too large and unpredictable to integrate into planning efforts (the proverbial “Black Swan”). Here, we estimate the NRV of late‐seral (mature/old‐growth) and early‐seral (post‐disturbance, pre‐canopy‐closure) conditions in a forest landscape driven by episodic, large, stand‐replacing wildfires: the Western Cascade Range of Washington, USA (2.7 million ha). These two seral stages are focal points for conservation and restoration objectives in many regions. Using a state‐and‐transition simulation approach incorporating uncertainty, we assess the degree to which NRV estimates differ under a broad range of literature‐derived inputs regarding (1) overall fire rotations and (2) how fire area is distributed through time, as relatively frequent smaller events (less episodic), or fewer but larger events (more episodic). All combinations of literature‐derived fire rotations and temporal distributions (i.e., “scenarios”) indicate that the largest wildfire events (or episodes) burned up to 105–106 ha. Under most scenarios, wildfire dynamics produced 5th–95th percentile ranges for late‐seral forests of ~47–90% of the region (median 70%), with structurally complex early‐seral conditions composing ~1–30% (median 6%). Fire rotation was the main determinant of NRV, but temporal distribution was also important, with more episodic (temporally clustered) fire yielding wider NRV. In smaller landscapes (20,000 ha; typical of conservation reserves and management districts), ranges were 0–100% because fires commonly exceeded the landscape size. Current conditions are outside the estimated NRV, with the majority of the region instead covered by dense mid‐seral forests (i.e., a regional landscape with no historical analog). Broad consistency in NRV estimates among widely varied fire regime parameters suggests these ranges are likely relevant even under changing climatic conditions, both historical and future. These results indicate management‐relevant NRV estimates can be derived for seral stages of interest in extreme‐event landscapes, even when incorporating inherent uncertainties in disturbance regimes.

Vegetation dynamics models: a comprehensive set for natural resource assessment and planning in the United States

“…When working at scales finer than the map zone, we advise users to consider adding temporal variability and spatial processes as needed to create more realistic ranges of reference landscape conditions. Such an approach has been developed using STSMs to estimate the natural range of variability for the Western Cascade Range of Washington (Donato et al 2020). In addition, ST-Sim can be run spatially or linked with other models to incorporate spatial dynamics such as fire spread (Jarnevich et al 2019).…”

“…LANDFIRE recognized the need for a national set of reference conditions despite the caveats associated with choosing a particular reference time period where the range of vegetation drivers did not include many current stressors, such as exotic invasive species, human land use, and climate change (Keane et al 2007). Despite these limitations, depictions of the relative proportion of different vegetation classes under historical conditions (or HRV) have been promoted in the resource management field as a reference for understanding patterns of degradation, and as a tool for informing restoration goals (Keane et al 2007, Haugo et al 2015, Donato et al 2020). While historical ecological information does not provide a prescription for restoration or a blueprint for desired conditions, it offers insight into the spatial and temporal variability of processes that have shaped ecosystems, which can support land management planning (Millar 1997, Keane et al 2009), even under a changing climate (Fulé 2008, Stephens et al 2013, Higgs et al 2014).…”

“…Second, all BpS were simulated based on average annual probabilities at the scale of a simulation cell. Therefore, for BpS with very infrequent but large episodic disturbances (e.g., wet temperate forests in the Pacific Northwest), simulated disturbance events were more frequent and affected smaller extents (i.e., fewer simulation cells) than actual historical events (Donato et al 2020). For these reasons, LANDFIRE's STSMs exhibit less variability than real landscapes, especially those with infrequent, large disturbances.…”

“…For the USA, HRV is often developed from the pre‐European‐American settlement era, which had substantial differences from the current time frame including an anomalous cold period, known at the Little Ice Age (Millar and Woolfenden 1999). Despite its limitations (Keane et al 2009, Millar 2014), managers have found the HRV useful in supporting a number of land management objectives including establishing restoration goals, understanding current conditions, informing desired conditions, and characterizing resilience (Hessburg et al 2015, DeMeo et al 2018, Keane et al 2018, McGarigal et al 2018, Donato et al 2020). Beyond these applications, HRV provides context for understanding current landscapes and offers insight into the drivers of ecosystem change over time and space (Fulé 2008, Keane et al 2009, Higgs et al 2014, Millar 2014).…”

Highlighting Complex Long-Term Succession Pathways in Mixed Forests of the Pacific Northwest: A Markov Chain Modelling Approach

“…Due to this temporal variation in stand type ratios over the last millennia, there is no way of validating this model outside of a detailed history of past stand and landscape disturbances, which unfortunately is not available. In addition, due to the permanent stochasticity of ecological processes, there is no "correct" landscape ratio to be compared with the model s predictions, as the real landscape have always been fluctuating [66]. Therefore, one way to evaluate forest successional models such as this one is by projecting long periods of forest change and then comparing the proportion of each ecological stage estimated by the model with those of the distribution observed in the field [67], an approach used since the early development of successional Markovian models [68].…”