Forests reflect the disturbances that have shaped them: fires, droughts, insect outbreaks, and harvesting all shape forest composition and structure (White and Jentsch, 2001). But under climate change, disturbance regimes will shift, and current forest conditions may no longer be in a “safe-operating space” and can change rapidly (Johnstone et al., 2016; Serra-Diaz et al., 2018). Across California climate change is projected to increase drought (Diffenbaugh et al., 2015; Crockett and Westerling, 2018), forest fire activity (Westerling, 2016), and insect outbreaks (Fettig et al., 2019). With such disturbance pressures, shifts in forest composition and structure will have long-term consequences for the types and levels of ecosystem services provided by future forests.

The 2012–2015 drought in California led to a mass tree mortality event in the southern Sierra Nevada caused by direct water stress (via C storage, hydrologic cavitation, or both; Sevanto et al., 2014), insect attacks on drought stressed trees, and increased fire activity. Historically, pre-European settlement forests across the Sierras were of lower density and experienced a short fire return interval (FRI) marked by frequent fires (every 11–16 years), but low to moderate fire severity (Hessburg et al., 2019). A history of fire suppression and timber harvesting over the past century resulted in a denser forest as shade tolerant species filled in, leading to anomalously high fuel loadings and larger, more severe fire patterns (Keeley and Syphard, 2019). While current forests are at or near a self-thinning phase due to the forest densification, pre-European settlement forests were likely sufficiently sparse that trees experienced little resource competition (North et al., 2022). Forest densification amplified the mass tree mortality event by increasing competition for soil water leading to increased insect outbreak severity (Safford and Stevens, 2017). Forest treatments reduce stand density and fuel loads and increase spatial and structural heterogeneity, which increases the capacity of the forest to resist mortality events (Restaino et al., 2019; Knapp et al., 2020) and can further enhance the resilience of the forest by allowing it to recover after a fire to pre-suppression conditions (Coop et al., 2020).

Because of budget limitations and competing objectives, there has been limited opportunity to implement widespread thinning and prescribed fire treatments despite the compelling need. As such, managers will need to rely on wildfires to reduce fuels and stand density (North et al., 2012). It is, however, unknown whether such treatments can sufficiently tame wildfire to effectively contribute to the objectives of a lower severity fire regime, improved drought and insect mortality resistance, and resilience in response to disturbance. Previous studies have suggested that early and aggressive, large-scale thinning treatments can reduce fire severity and increase carbon storage across the whole Sierra Nevada range (Liang et al., 2018). Accelerated forest treatments on 14% of the Sierra Nevada ecoregion per decade reduced the risk of tree mortality from fire and held more carbon (Liang et al., 2018). Within the Lake Tahoe Basin (a subset of the larger Sierra Nevada), Loudermilk et al. (2017) found that strategically placed fuel treatments in high ignition areas covering less than 7% of the forested area can reduce wildfire risk, increased the fire resiliency of forest, and benefitted carbon storage. However, Scheller et al. (2018) found that fuel management practices in Lake Tahoe based on Loudermilk et al. (2017) would not reduce landscape-scale forest mortality from beetle outbreaks.

We hypothesized that mimicking the historic fire-return interval, by matching it with the combined frequency of natural disturbances (wildfire) and management (i.e., thinning and prescribed fire), will maintain forest resilience despite a changing climate. We measured this resilience by tracking several metrics through time: (1) tree-mortality due to wildfire, (2) tree mortality due to insect outbreaks, (3) carbon storage and sequestration, and (4) avoidance of forest conversion to shrub. To test our hypothesis, we deployed a forest landscape simulation model, LANDIS-II, to test a broad range of management scenarios under climate change. LANDIS-II is able to simulate multiple processes within forests, including forest growth and succession, wildfire, insects, management, and climate change. This allowed us to test a variety of management scenarios ranging in intensity and methodology. We specifically focused on scenarios that encapsulated a range of proposed actions for returning the landscape to the historic disturbance interval. Our scenarios allowed us to parse the contributions of specific and plausible management actions and therefore to evaluate their relative performance at maintaining forest resilience over the 21st century. Those management scenarios were paired with climate change projections selected to highlight a broad range of potential future climate conditions.

While the management scenarios were designed to integrate management activities with wildfire to achieve a disturbance return interval that matched the historic fire return interval, the target historic interval was not achieved (Figure 2). These differences are due in part to restrictions on management activities. Scenarios that used high levels of thinning (like Scenario 5) were further restricted by other disturbance processes (mainly insects) that were dispersed across the landscape resulting in a decline in the amount of area harvested or thinned through time as those stands were not eligible to be treated. Between the first decade, 2020–2030, and the last decade, 2090–2100, the area treated by management declined by 60% for scenario one and 40% for scenario six (Supplementary Appendix 2; Supplementary Figure 9). The forest treatment goal for scenarios one through six was 11–63% of the landscape per decade, while the actual treatment modeled was 4–59%. Moreover, as the amount of wildfire increased over time, especially around model year 2050, the treatment area (thinning and prescribed fire) declined because individual cells were already disturbed and within their historic disturbance return interval and therefore not eligible to be treated.

Future forest change will depend on the climatic stressors going forward. Our climate projections indicate that the climatic water deficit will increase, driven mainly by increases in temperature. However, in four out of the five future climate projections selected, despite those increases in temperature, there was also a moderate to substantial increase in projected precipitation, which lowers the likelihood of extreme fire weather conditions (Supplementary Appendix 1; Supplementary Figures 1–4). Recent droughts (2003–2004, 2012–2015, 2020+) are exceptional and are not recreated in any of our climate projections, with the MIROC5 projection coming closest in the 2050s. While projected climatic water deficit increased into the future, this did not necessarily translate to increases in projected fire weather due to timing of the precipitation. As a result, area burned did not exceed recent trends until the end of the century under the non-MIROC5 projections; but extreme fire events (events greater than 100,000 ha), which is only slightly larger than a fire that happened in the study area in the fall of 2021 (Caldor Fire at 90,000 ha) had a chance of occurring with the MIROC5 projection. Regardless of climate projection, and even for historic climate, insect outbreaks were spatially pervasive and the leading source of forest mortality (Figure 3 and Supplementary Appendix 1; Supplementary Figure 6). Only the MIROC5 future climate projection had increases in fire mortality beyond the levels generated from the historic contemporary climate and projected resampled contemporary climate. Repeating historic climate into the future produced more variability in tree mortality and source of mortality given the random resampling of the climate data.

Because of the rapid onset of outbreaks, regardless of treatments conducted prior to the period simulated, simulated forest management had limited effect on reducing insect caused forest mortality (Figure 3). While insect mortality rapidly increases flammable fuels (duff, litter, and dead wood), the limited frequency of wildfire meant that large patches of high severity fires were rare except under the MIROC5 projection (Supplementary Appendix 1; Supplementary Figure 7). The primary effect of higher intensity management scenarios (Scenarios 3–6, beyond the business-as-usual Scenario 2) was lowering the cumulative mortality by end of century caused by wildfires compared to Scenarios 1 and 2 for only the MIROC5 climate projection due to the overall higher levels of area burned in that projection (Figure 4) and those differences were statistically significant (Supplementary Appendix 1; Supplementary Table 2).

Simulated forest composition change was substantial and rapid due to disturbance and climate change. Most of the insects we modeled have various species of the genus Pinus as their preferred host, and because of the high mortality rate from these insects, other tree species from within the Sierran mixed conifer forest type (Douglas-fir, incense-cedar) that were already extant in those cells became the dominant tree species in outbreak-prone areas (Figure 5 and Supplementary Appendix 1; Supplementary Figure 8). At higher elevations, tree species such as mountain hemlock and red fir declined due to reduced regeneration as they were no longer in their ideal climate envelope by the end of the century (Supplementary Appendix 1; Supplementary Figure 9) but this was balanced by upslope movement of the mixed conifer type by end of century (Supplementary Appendix 1; Supplementary Figure 10). Tree cover was generally maintained as a result—only a small percentage of cells in the modeled forested landscape converted to non-forested (shrub dominated)—and mainly in areas that had experienced high severity fire rather than insect mortality (Figure 5). For historic climate projected into the future, there was a 2% increase in non-forested area after 30 years (from 2% of the landscape to 4%). For the non-MIROC5 future climate projections, there was a 4–6% increase, depending on management scenario by end-of-century. And for the MIROC5 climate projection, there was an 8% increase of non-forested area.

In spite of the compositional shift of forest cover, the total amount of carbon in the forest remained relatively stable across the duration of our simulations, except for a modest decline under the MIROC5 projection (Figures 6, 7). With prescribed fire being the central treatment in Scenario 6, and prescribed fire removing more down and dead material, Scenario 6 maintained less total C overall compared to the other management scenarios. Nevertheless, the landscape switched from being a moderate carbon sink in 2020 to becoming a carbon source by year 2050, regardless of management scenario.

We hypothesized that mimicking the historic fire-return interval by combining natural disturbances (wildfire) and management (i.e., thinning and prescribed fire) will maintain forest resilience despite a changing climate. We assessed our hypothesis by tracking: (1) tree mortality due to wildfire, (2) tree mortality caused by insect outbreaks, (3) carbon storage and sequestration, and (4) the maintenance of tree cover.

Our simulations show that tree mortality due to wildfire can be reduced by management, particularly under more extreme weather conditions, as expressed by MIROC5. Under this climate future, Scenarios 3 and 4 reduced tree mortality and this reduction exceeded the mortality generated by management itself, while Scenario 5 was a more even substitution of mortality agents and Scenario 6 had a net increase in mortality due to higher mortality associated with prescribed fires over thinning. The other climate scenarios generally did not show a strong effect of management. Our MIROC5 results are consistent with prior research that suggested that fuels management would become more effective as with climate change due to more frequent intersections between wildfire and treated areas (Syphard et al., 2011; Loudermilk et al., 2014).

Our simulations further suggest that insect mortality was not substantially affected by management as was simulated. Regardless of management scenario, and with little variation among climate futures, our simulations indicate that insect mortality will become the dominant disturbance agent over the next century. While these treatments were in line with current levels of removal, treatments specifically designed to mitigate wildfire hazard may not confer drought or insect resistance. Modeling the impact of more aggressive treatments to reduce insect mortality by thinning the overstory, up to 75% rather than the ∼20% currently in the model (Fettig et al., 2007), and intentionally creating structural forest diversity over large areas (Fettig et al., 2012) with greater biomass removed per hectare could provide insights into how management may affect insect mortality.

Carbon storage in this landscape will be limited by the frequency, severity, and type of future disturbances as well as the potential interactions among disturbances. While forest management activities can offset the potential C losses from wildfire (Liang et al., 2017), the high levels of projected insect mortality are not offset by the simulated management activities (Figure 3). Unlike fire, where C is released through combustion, the dead materials resulting from insect outbreaks remain in the forest until they decompose, though it is at risk of combustion at a later point. The timing of fire after the insect disturbance is important. Meigs et al. (2016) found that fire mortality eventually declined due to reductions in the live vegetation that was susceptible to fire, though this effect was most pronounced several years after fire, while Hart et al. (2015) found negligible links between insect mortality and fire area. This process of insect mortality offsetting future fire mortality might explain why C stocks were able to increase even under the MIROC5 projection through 2040. Later in our simulations, as the effects of climate change intensify in the latter half of the century, C stocks declined as disturbances increased (Figure 6).

Finally, our hypothesis that forest management would increase the retention of tree cover was not confirmed. Although there has been concern over the possible loss of forest to shrub or grassland due to large high severity fire events (Serra-Diaz et al., 2018; Coop et al., 2020), we found only a limited area of conversion from forest to shrub (Figure 5). This has been noted in certain environments due to large patches of high severity fire and the climatic conditions post-fire that limit successful establishment (Harvey et al., 2016; Coop et al., 2020). Such conversions were not found in our results at scale (Figure 5), mainly because of the limited increase in proportion of high severity fire within the model. Although our wildfire sub-model simulated large, high-severity fires under the MIROC5 projection, in aggregate they were not a large driver of forest conversion. Because insect outbreaks are the dominant mortality agent into the future but do not result in 100% mortality, the remaining understory allowed for the maintenance of forest cover and depending on the dominant species and presence of remaining adult trees after a wildfire or insect outbreak, such as yellow pines in the Sierra Nevada, the potential for a drought induced forest conversion to shrubland (Young et al., 2019) may be limited.

Climate change will drive direct and indirect substantial changes to forest composition and structure. It may not be possible to maintain the forest as we know it today (Schoennagel et al., 2017). Fettig et al. (2019) found in the southern Sierras that, after the insect outbreaks in response to the 2012–2015 drought, there was a shift in the understory competition toward more shade tolerant species in mixed conifer stands, similar to our findings of a transition from mixed conifer to increasing dominance by Douglas-fir at low to mid-elevations (Figure 5). This trend also reflects historical trends based on climate envelope modeling: from 1930–1996 a 240,800-hectare area within the study area increased in cover of Douglas-fir, hardwoods, and grassland, while upper elevation conifers decreased (Thorne et al., 2008); and from 2001 to 2018, Douglas-fir saw an increase in population expansion within the Sierra Nevada (Stanke et al., 2021). In looking at potential future climate refugia, Thorne et al. (2020) found that 27% of Douglas-fir forest in Sierra Nevada was within a climate refugia compared to just 2% of red fir. This decline in red fir and other montane species is reflected in our findings (Figure 5) and moving upslope to occupy that space were primarily Douglas-fir and other components of the Sierra mixed conifer forest type.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://github.com/LANDIS-II-Foundation/Project-Tahoe-Central-Sierra-2019 and archived through zenodo at: doi: 10.5281/zenodo.7199459.

RS designed the model. CM implemented the simulations and took the lead in writing the manuscript. CM and KW analyzed the data. All authors provided critical feedback and helped shape the research, analysis, and manuscript.