1 Introduction

Disturbance plays an important role in the dynamics of natural forests. Biotic and abiotic changes can disrupt stand structure, resource availability, and/or the physical environment (Pickett and White 1985). Such disturbances can range spatially from small-scale to large-scale, stand-to-landscape replacing events. The rotation interval (RI) for disturbance (not necessarily stand replacing) varies widely with many factors including disturbance type, forest cover type, successional stage, geographic location, ownership, and management regime. Moreover, these rates of disturbance may vary with time, depending on long-term climate trends, anthropogenic land use patterns, wildlife population cycles, and other factors. In total, these disturbances drive successional change in the forest (Guyette and Kabrick 2002; Reilly and Spies 2016) by, for example, determining where and when canopy openings occur, which seedlings and sprouts grow free of browse, and the average time between disturbance events.

Increasing disturbances associated with climate change, invasive species (both plant and animal), invasive tree pests like emerald ash borer (Agrilus planipennis Fair.), and diseases like Dutch elm disease (Ophiostoma ulmi Buism) have been reported in some parts of the world, including our study area. Extreme wind and precipitation-related events (e.g., either too much or too little) also appear to be on the rise. Indeed, it seems we must now plan for increased disturbance in the forest (Seidl et al. 2017). Thus, assessment of regional patterns and trends is needed.

Disturbance rates are relevant to timber production and harvest scheduling, but also to carbon sequestration, habitat management, biodiversity, and ecosystem services considerations. (For a few examples, see Rogers (1996), Thom and Seidl (2016), and Seidl et al. (2016).) Here, we use data from the ongoing cyclical forest inventory effort of the United States Department of Agriculture (USDA) Forest Service’s Forest Inventory and Analysis Unit (FIA) to analyze recent disturbance trends. We believe this is also the first study to utilize the cyclic FIA data to estimate RIs for different types of disturbance.

Prior efforts to determine RIs have typically looked at a handful of disturbance and cover types across a modest geographic region using intensive field methods to assess dates and types of disturbance (Heinselman 1973; Frelich and Lorimer 1991; Frelich and Reich 1995). Those efforts typically examined tree data obtained using increment borer samples from the trunks of standing or downed trees. These samples allowed the assessment of fire dates, drought stress, times of release from suppression, or other indicators of disturbance across relatively long periods of one to four centuries.

However, forest inventory data can be used to assess disturbance-related mortality across large spatial extents (Reilly and Spies 2016). This research used data collected for periodic inventory efforts with a sample design unique to the Pacific Northwest Region. Others have used satellite imagery (Landsat) and light ranging and detection (lidar) to assess combined change from harvesting and natural disturbance (Vogeler et al. 2018; Matasci et al. 2018). The more recent, and nationally consistent, cyclic data collected by FIA (a 5-year cycle of data collection contributes to each estimate of forestland area in Minnesota) hold promise for informing similar analytic approaches. However, these data present various difficulties stemming from the timing and sequence of observations. We develop an approach, utilizing a broad-based representative forest inventory to inform a range of interests including estimates of natural and unplanned human-induced disturbance. While our methods are applicable nationally and elsewhere, our specific objective was to devise a method for using FIA estimates of area disturbed to study RIs and trends for various types of natural disturbances across the forest. We use Minnesota as a case study to highlight the utility of the technique. We also test the hypotheses that disturbance rates (other than harvesting, which represents a planned treatment rather than a disturbance) have increased, decreased, or remained the same during the period of observation.

2 Materials and methods

2.1 Study area

A combination of natural and anthropogenic disturbance, coupled with periods of regrowth, has shaped Minnesota’s forests over time. The exact nature of these dynamics varies widely depending upon timber markets, soils, physiography, and the specifics of disturbance leading to change. Frequent fires, windstorms, insect infestations, and diseases have all played a role in shaping Minnesota’s forested landscape.

Heinselman (1973, 1996) spent much of his career mapping historic fires which shaped the mosaic of even-aged and multi-aged coniferous forests of Minnesota’s Boundary Waters Canoe Area Wilderness. Fires also played a large role in shaping the forest following the removal of most white and red pine from accessible portions of Minnesota’s northern forest between 1890 and 1920. These fires consumed the dense slash and damaged less desirable trees left on site. Aspen responded strongly to clearcuts followed by fire (Friedman and Reich 2005), dominating regrowth across much of Minnesota.

The famous blowdown of 1999, which affected a large swath of southern boreal forest, and mixed hardwood forest stretching from Minnesota across Southern Quebec, Canada, is another example of disturbance shaping Minnesota’s forests. This massive straight-line windstorm toppled mature trees across 193,000 ha (an area roughly 48 km long and 6-18 km wide) in Minnesota alone (USDA Forest Service 2001). The blowdown released younger or smaller diameter stems of shade-tolerant species present in the understory. Regrowth in the path of the blowdown leans heavily towards northern white cedar (Thuja occidentalis L.), balsam fir (Abies balsamea (L.) Mill.), and black spruce (Picea mariana Mil.), a distinct change from the mature jack pine (Pinus banksiana (Lamb.)), red pine (Pinus resinosa Ait.), eastern white pine (Pinus strobus L.) and aspen (Populus tremuloides (Michx.), Populus balsamifera (L.), and Populus grandidentata (Michx.)) originally present on most of the landscape (Rich et al. 2007).

An explanatory mechanism for recent increases in insect outbreaks is their accelerated development rate and increased reproductive potential related to increasing temperatures (Porter et al. 1991; Ayres and Lombardero 2000). Additionally, the establishment of exotic insect and pathogen species in new locations (Aukema et al. 2010) may be more likely in a warmer climate (Virtanen and Neuvonen 1999; Lesk et al. 2017). The sporulation and colonization success of some forest pathogens may also respond to changes in temperature, precipitation, soil moisture, and relative humidity (Chakraborty et al. 1998). Hence, increasing temperatures also likely contribute to increased severity and shorter RIs for insect outbreaks, disease, fire, rot, and other events (Frelich and Reich 2010). Our methods may contribute to better understanding these processes and their implications for forest management.

2.2 FIA disturbance observations

Disturbances recorded during plot visits (FIA methods are summarized in Appendix 2) include biotic and abiotic agents such as insects, diseases, animals, fire, weather, and geologic events affecting individual trees (Appendix 1) (O’Connell et al. 2017). Following FIA procedure, we do not consider intentional disturbances such as harvesting or thinning here. In that context, human-induced disturbance includes things like tree removal not associated with a planned harvest, brush clearing that damages or removes trees, or other unplanned human activities resulting in damage or mortality of the specified threshold.

Per FIA instructions, disturbances must have occurred since the last plot visit (5 years prior in Minnesota, but 7 or 10 years for some states) or within a cycle length (a defined 5-year sample period) for new plots. FIA records the estimated year of disturbance along with the type. A disturbance must be at least 0.405 ha (1 acre) in size (clearly, this observation extends beyond the plot and includes assessment of conditions in the plot vicinity), with mortality or damage to 25% of the trees on the physical plot (e.g., the equivalent of at least one full subplot must be affected in most cases). Observations of disturbance are associated with the entire plot.

When a plot becomes inaccessible for any reason, it is replaced by a new observation (possibly from a different location) during the following field season. For this reason, observations of disturbances greater than 5 years old (Fig. 1 and Appendix 3 Table 4) can be recorded. There are two caveats: (1) Observations from five sequential panels must be used to produce estimates of total and disturbed area. (2) A temporal lag between disturbance events and their observation limits the sightability, relevance, or completeness of some observations.

Fig. 1
figure 1

Distribution of time delay for observation of disturbances via FIA inventory in Minnesota. Note the observation interval corresponding to the largest number of disturbance observations (e.g., 2 years following the event). We calculate the focal year as the annum in which disturbances observed in a given year were most likely to have occurred

Full size image

Recorded disturbances occurred between 1995 and 2016. Hence, we have a continuous set of observations, including any visible disturbance, spanning 21 years and ~ 7 million ha. Disturbances more than a few years old are clearly difficult to observe, with evidence of different disturbances likely persisting for different lengths of time. The record is reasonably complete for disturbance events occurring during the period 1998 to 2014 (Appendix 3 Table 4). Because most disturbances more than a few years old likely go unnoticed, disturbance RIs derived from these estimates are biased upwards (e.g., some level of disturbance went unnoticed).

2.3 Model for estimating disturbance

The standard defined by FIA is a 0.405-ha (1-acre) disturbance affecting at least 25% of the trees present on and around a subplot. Thus, our prediction capability using these data is for disturbances affecting at least 0.405 ha.

At times, more than one disturbance occurs between visits to a plot. We counted these disturbances (142 records) as unique disturbance observations. We assigned these secondary disturbances a distinct disturbance code and year and counted the disturbed area a second time in summation of disturbed area. Correlation of repeated measures was ignored, but may be an issue for these 142 observations (0.76% of our total plot observations; 142/18,789), where the nature of a first disturbance may have altered susceptibility to a future disturbance. Regardless, a downward bias in estimates of disturbed area resulting from loss of disturbance visibility over time (Fig. 2) likely makes our estimates conservative in terms of the RI calculated.

Fig. 2
figure 2

Timeline for observation of disturbances by FIA in Minnesota. Percentages indicate the cumulative proportion of total observations for a given disturbance year made with each panel of a 5-year cycle

Full size image

The rotation (or recurrence) interval is the usual measure used to discuss the frequency with which a defined portion of the landscape will experience one or more disturbances (Swift and Ran 2012; Pickett and White 1985). The forestry literature further provides a distinction between the disturbance return interval (a point-based concept) and the rotation interval (an area-based concept) (Bond and Keeley 2005). Our methods rely on area estimation and therefore are parallel to the rotation interval concept and terminology. Comparison of area disturbed over time against total area (e.g., for a cover type, or for the forest as a whole) enables calculation of a RI for disturbance. In this case, the RI is the expected period required for each 0.405 ha in a specified cover type, or stratum, to experience one or more disturbances. We also discuss the RI for specific types of disturbance. Those intervals reference the entire forest, rather than area of a specific cover type, unless otherwise noted.

For each cover type group, and for the forest as a whole, we calculated RI as:

$$ {\mathrm{RI}}_t=\frac{{\mathrm{Total}\ \mathrm{Area}}_t}{\mathrm{Disturbed}\kern0.5em {\mathrm{Area}}_t}, $$

where t corresponds to the focal year for reporting of disturbances, and Total Area(t) is the then current estimate for total forest. In Minnesota, the focal period for observation of disturbed area lags 2 years behind the observation period for total area. An ordinary least squares regression model using focal year as the predictor and RI as the response was developed. We used Welch’s t test (Welch 1951) and linear regression (R Core Team 2018) to test for significance of changes in disturbance frequency over the period of observation.

To assess likely the variability of disturbance rates, we use the prop.test function (R Core Team 2018) to calculate the 95% binomial confidence interval for each forest type group/disturbance type group combination, and for the forest as a whole. The size of the set of responses generated at each iteration was proportional to (disturbed area sampled / mean expansion factor) for that forest cover type. This method ensures that perceived precision of the estimate is proportional to the number of disturbance observations contributing to the estimate for a forest type. In addition to trend analysis, we present cross-tabulations of:

  1. 1.

    Disturbance RI (all disturbances combined) by forest cover type,

  2. 2.

    Disturbance RI (all cover types combined) by disturbance type group, and

  3. 3.

    Disturbance RI by forest cover type-disturbance type group combinations.

2.4 Statistical computing tools

Here, we propose a reproducible methodology for estimating disturbance RIs. We compile our scripts (Appendix 3) as R markdown documents (.Rmd) constructed and run using {knitr} (Xie 2014, 2015, 2018) in RStudio (RStudio Team 2016). Scripts used to obtain our results summarized by disturbance group (Table 2) are available upon request. Similar methods were used to compile the disturbance data by forest cover type group (Table 3) and for the cross-tabulation of forest cover type group and disturbance type group (Appendix 3 Table 5).

3 Results

3.1 Total disturbance rates: USDA-FIA inventory years 1999–2016

We have records of 1780 disturbance events from 18,759 discrete observations (1999–2016) distributed across ~7 million ha of forestland. Of these, we used 1677 disturbance observations for events thought to have occurred from 1998–2014 to develop complete estimates of disturbance and to calculate RIs (Appendix 3 Table 4, bold). After excluding incomplete observations for events occurring at the extremes of the period of interest, we have a usable set of observations (Appendix 3 Table 4, shaded and bold) to develop estimates of disturbance RIs.

We depict frequency of disturbance as a RI by focal year (2001–2014) in Table 1 and Fig. 3. The expected RI declines from almost 23 years, to roughly 8 years over the course of observation. A linear model of the trend over time yields a highly significant slope (p < 0.0000012, R2 = 0.8584) suggesting an average change in overall RI of − 1.28 years per year of observation. Similarly, one-sided Welch’s t test (p = 0.01786 with 172 degrees of freedom) suggests rejection of the null hypothesis that initial RIs for forest cover type-disturbance type combinations were either less than or equal to those observed at the end of the observation period. We therefore accept the alternative hypothesis that disturbance RIs have declined since 2001. Sample size limits our ability to estimate RI and confidence bounds for disturbance/cover type pairs with very few disturbance occurrences. Those with sufficient observations appear in Appendix 3 Table 5 along with binomial confidence bounds (e.g., RI.025 and RI.975). Examination of the temporal trend (Figs. 3, 4, and 5) shows that variation in the RI observed for different disturbances and cover types has decreased along with the interval.

Table 1 Total area of forestland and disturbance for evaluation groups* 2003–2016
Full size table
Fig. 3
figure 3

Disturbance rotation interval (RI) trend, 2001–2014. The trend line is calculated as: E[RI] = 26.41 − 1.28 × ΔFocal year, with variance weighted by number of plots contributing to each 5-year estimate. The expected reduction in RI for each 1-year increase in focal year is 1.28 years (p < 0.0001)

Full size image
Fig. 4
figure 4

Rotation interval (RI) (years) by disturbance type for focal years 2001–2014 (focal year refers to the annum in which most disturbances observed in the 5-panel evaluation group are thought to have occurred.). Geologic and unknown disturbances are omitted for visual simplicity

Full size image
Fig. 5
figure 5

Disturbance rotation interval (RI) for major forest cover types occurring in Minnesota

Full size image

3.2 Details by disturbance and forest cover type groups

Frequency of disturbance and uncertainty associated with RIs varies greatly by disturbance type (Table 2). The most frequent disturbances were weather, animal, and human. RIs have generally declined for all disturbance types except human and geologic. The greatest decreases have occurred for disease, vegetation, insects, and fire (Fig. 4).

Table 2 Forest disturbance (1998–2014) observed by USDA-FIA in Minnesota (1999–2015)
Full size table

There was an almost 3-fold difference in disturbance RIs among forest cover types (Table 3). Although the aspen-birch group has suffered the greatest amount of absolute disturbance (hectares), both the oak-hickory and lowland hardwood groups have shorter RIs for disturbance. RIs have generally decreased over the study period for all forest type groups (Fig. 5). The RIs for total disturbance in a given cover type (Table 3) are much shorter than those for a given disturbance type (Table 2). For example, in order for the entire forest to experience insect damage or mortality to at least 25% of trees across at least an acre, we would need to wait approximately 180 years (Table 2). On the other hand, for all aspen to experience any kind of disturbance, we would only need to wait about 14 years (Table 3).

Table 3 Total disturbance by forest cover type group
Full size table

For reference, we estimate the RI for fire between 24 and 148 years (Appendix 3 Table 5) in the mixed pine woodlands of northern Minnesota with a mean of 60.5 years. Our overall estimate of RI for fire in the Border Lakes ecological subsection encompassing the Boundary Waters Canoe Area and Wilderness is 54.6 years.

4 Discussion

Previous research supports non-harvest tree fall recurrence intervals of 50–200+ years in mesic hardwood and mixed forests (Runkle 1982; Frelich and Lorimer 1991; Seymour et al. 2002). Frelich and Lorimer report 145–175 years for canopy residence times in sugar maple and hemlock, while at the 0.5-ha scale, recurrence intervals range from 69 years for > 10% canopy removal to 1920 years for > 60% canopy removal, also similar to results from this study (using a 25% canopy damage, not removal, threshold). Somewhat shorter RIs (~ 129 years) predominate across Canada’s western Boreal Shield (interpreted from White et al. 2017), which extends south into Minnesota’s Border Lakes ecological subsection. Intervals are even shorter (50–100 years) in the southern boreal forests of northern Minnesota (Heinselman 1973). These observations are consistent with the higher frequency of minor disturbance events compared with major events resulting in substantial tree mortality.

Results of the current study, as well as casual observation, support the conclusion that more frequent disturbances are now shaping Minnesota’s forests. For example, emerald ash borer, oak wilt, Dutch elm disease, tent caterpillar, a severe drought, and several large wind and fire events have all occurred in Minnesota during the period analyzed.

Continued characterization of disturbance patterns using spatially and temporally representative data, like FIA, is needed to interpret changes in the forest, and processes affecting the forest, across larger spatial and temporal scales. The picture we can construct from FIA will improve as the length of time with consistently spaced repeat observations increases. The upper confidence bounds for many uncommon disturbance type/cover type combinations (Appendix 3, Table 5) will likely decrease as we acquire additional observations.

Disturbance frequencies fluctuate over time, especially those for large infrequent disturbances, making variability of estimates quite high. This complicates interpretation of the observed trend towards declining RIs. Nevertheless, the trend is significant, and the change is substantial (~ 50%), reaching very low recurrence intervals in the latest 5-year period (even considering the confidence intervals). This trend may indicate a shift in response to warming climate (Dale et al. 2001; Seidl et al. 2017), or change in other factors including anthropogenic species introductions/extinctions, fire suppression/ignition, or land clearing for agriculture. These changes could also be part of a natural fluctuation. If it is a regime shift characterized by increased frequency of storms, droughts, and invasive insect pests and tree diseases, it may still be temporary and shift back to longer recurrence intervals because susceptible trees die off, in a self-regulating, homeostatic process described by Runkle (1982).

The Generic Environmental Impact Statement on Timber Harvesting in Minnesota (Jaakko Pöyry Consulting, Inc. 1992), as well as more recent cross-tabulations of cover type area (Wilson and Ek 2018), shows that there is a regular and somewhat predictable exchange of area among cover types. Disturbance interacts with age-related growth and susceptibility patterns to determine the success of individual stems and hence the direction and velocity of forest succession. This hypothesis is supported by much experimental and observational data and may be corroborated through comparison of disturbance rates against successional and demographic changes among forest cover types and ecoregions. To this end, we enable assessment of disturbance, as well as differences among forest cover types, across relevant spatial and temporal scales.

5 Conclusions

Further investigation of disturbance in other regions may corroborate the trend towards shorter disturbance intervals observed here (e.g., Reilly and Spies 2016) as part of a broader pattern, or show that they are a local phenomenon. Trends reported here appear to be both meaningful in terms of their magnitude and are significant from a statistical perspective. Methods presented are transferrable to other states or regions monitored by systematic forest inventory and may aid in illustrating the spatial and temporal scale of observed changes. Finally, regional or national application of this methodology could help estimate RIs for different forest and disturbance types across their geographic ranges.