Forest dynamics in Westland, New Zealand: the importance of large, infrequent earthquake-induced disturbance


  • Andrew Wells,

    Corresponding author
    1. Ecology and Entomology Group, Soil, Plant and Ecological Sciences Division, Lincoln University, PO Box 84, Canterbury, New Zealand
      *Present address and correspondence: PO Box 14, Franz Josef Glacier, South Westland, New Zealand (e-mail
    Search for more papers by this author
  • Richard P. Duncan,

    1. Ecology and Entomology Group, Soil, Plant and Ecological Sciences Division, Lincoln University, PO Box 84, Canterbury, New Zealand
    Search for more papers by this author
  • Glenn H. Stewart

    1. Ecology and Entomology Group, Soil, Plant and Ecological Sciences Division, Lincoln University, PO Box 84, Canterbury, New Zealand
    Search for more papers by this author

*Present address and correspondence: PO Box 14, Franz Josef Glacier, South Westland, New Zealand (e-mail


  • 1Large, infrequent disturbances can exert a dominant influence on the structure and dynamics of forested landscapes and in Westland, New Zealand there is evidence of at least three massive earthquakes within the last 650 years. We reconstructed the history of forest disturbance in two study areas, totalling 1412 ha, to quantify the role of such disturbance in structuring the conifer/hardwood forests.
  • 2We divided the study area into different land-surface types, aged trees on each land surface and identified cohorts of trees established in response to past disturbance. The type of disturbance (tree fall or erosion/sedimentation event) that initiated cohort establishment was determined from the type of land surface and other physical evidence. We also dated abrupt growth releases or suppressions in tree rings to aid disturbance history reconstruction.
  • 3Erosion and sedimentation events dominated the disturbance regime, affecting 86% of the study area in the last 650 years. Catchment-wide episodes of forest cohort-initiating disturbances were centred around 1820–30, 1710–20, 1610–20 and 1460 ad. Of the 51 cohorts identified in the study area, 47 were initiated during one of these episodes, when disturbance by erosion or sedimentation affected from 10–50% of the study area. Consequently, over 80% of the forested area currently comprises simple, first generation cohorts of trees established after catastrophic disturbance. Only 14% of the study area is more complex, all-aged forest.
  • 4Three disturbance episodes coincide with the three most recent Alpine fault earthquakes (c. 1717, 1630 and 1460 ad), while one coincides with earthquakes recorded to the south of the study catchment in 1826 ad. Age structures from throughout Westland show that extensive, similar-aged, post-earthquake cohorts of trees are a feature of the region, suggesting that infrequent, massive earthquakes are the dominant coarse-scale disturbance agent, triggering episodes of major erosion and sedimentation and leaving a strong imprint on the forest structure.


In some forests, the disturbance regime is characterized by frequent, small-scale gap formation, so that the forest is constantly turning over and, at any one time, a significant proportion of the landscape is in a recently disturbed state. Insights into the structure and dynamics of such forests can be gained by studying contemporary processes of gap formation and tree regeneration (Runkle 1982). Other forests, however, are affected by large, infrequent disturbances with return times of decades to centuries. Examples of these disturbances include catastrophic fires (Heinselman 1973), floods (Duncan 1993), hurricanes (Foster & Boose 1995), earthquakes (Veblen et al. 1992) and volcanic eruptions (Turner et al. 1997). Because these disturbances tend to be severe, often causing complete devastation of parts of the landscape, and to impact over large areas, they may exert a dominant influence on the pattern and function of forested landscapes even when they occur very infrequently relative to other types of disturbance (Garwood et al. 1979; Foster et al. 1998; Turner & Dale 1998). Indeed, because of the great longevity of many trees, it is possible for much of the landscape-level pattern in forest structure to be a consequence of the impact of previous, large disturbances that have no counterpart in the recent past (Foster et al. 1998). Despite their significance, the importance of large, infrequent disturbances in the dynamics of forested landscapes can be overlooked if the disturbances occurred sufficiently long ago so that direct evidence of their impact has been obscured by forest regrowth.

Much of the forested landscape of Westland, New Zealand is predisposed to large disturbances, particularly landslides and floods, due to high rainfall, steep topography, weak schist rock masses and the presence of a large, active fault (the Alpine fault) that runs the length of the region (Fig. 1). Large disturbances are important in initiating the regeneration of the major canopy tree species in this region, particularly long-lived conifers in the families Podocarpaceae and Cupressaceae (Veblen & Stewart 1982a; Norton 1983; Stewart & Rose 1989; Duncan 1993; Rogers 1999). Evidence is also accumulating that many of the forest-initiating disturbances in Westland may have occurred synchronously in a few brief episodes during the last 650 years. The published tree-age data from forest stands sampled throughout Westland show that in most stands the canopy trees regenerated as one or more distinct cohorts following major disturbance caused by tree fall or erosional processes (Wells et al. 1998). The dates of the disturbances that initiated each cohort (estimated from the age of the oldest tree in each cohort plus an interval to allow for colonization and growth to increment-coring height) cluster into three distinct episodes centred around 1717, 1630 and 1460 ad (Wells et al. 1999). These dates coincide with the dates of the most recent Alpine fault earthquakes inferred from radiocarbon-dated material excavated from trenches dug across the active fault trace, and from other indirect indicators of earthquake impacts, including tree-ring growth anomalies (Yetton 1998; Wells et al. 1999). Hence, there is evidence that large earthquakes, with associated land-sliding, flooding and tree-toppling, have periodically caused large-scale tree mortality and initiated widespread forest regeneration across much of Westland in the last 650 years.

Figure 1.

Map of the Karangarua River catchment, south Westland, New Zealand. The boundary of the catchment is indicated by the dashed black line. The two sites selected for detailed study (Welcome Flat and Lower Karangarua study sites) are marked. The inset map shows the location of Karangarua catchment and Alpine fault within the South Island of New Zealand.

Nevertheless, evidence that large, infrequent earthquakes have significantly impacted Westland forests derives from tree-age data collated from numerous, often widely scattered forest stands, most of which were subjectively chosen for sampling by the authors of the original studies. As such, the existing data do not permit us to quantify with any certainty the landscape-level extent or nature of these earthquake impacts, and therefore to assess the importance of large, infrequent earthquakes relative to other forms of disturbance in influencing forest pattern and processes in this region.

In this study we reconstruct forest disturbance history over the last 650 years in two study areas totalling 1412 ha in a Westland river catchment. Our aim is to quantify the importance of large, infrequent disturbances in structuring forests of this region. We do this by mapping and dating disturbances that have initiated new cohorts of canopy trees in disturbed areas. Previous studies indicate that land-sliding and flooding (due to storms and earthquakes) and tree-toppling (due to wind and seismic shaking) are the major forms of forest disturbance in this region (Adams 1980b; Wardle 1980; Norton 1983; Stewart & Rose 1989; Duncan 1993; Wells et al. 1998).

The study area

The study was carried out in the Karangarua River catchment, south Westland (Fig. 1), which drains a stretch of over 20 km of the Main Divide of the Southern Alps. Precipitation in this catchment is frequent and heavy throughout the year: mean annual rainfall at Fox Glacier (25 km north of Karangarua) is 4540 mm (New Zealand Meteorological Service 1983), although annual rainfall in the catchment is likely to vary from about 3500 mm at the coast to over 10 000 mm in the frontal ranges (Griffiths & McSaveney 1983).

The topography of the catchment is typical of large valleys in the region and is strongly influenced by the plate boundary at the Alpine fault (Whitehouse 1988). West of the fault a broad, gently sloping piedmont flood plain extends to the Tasman Sea, whereas east of the fault the Southern Alps rise abruptly, resulting in the central and upper reaches of the catchment having narrow boulder-choked river beds and steep-sided valleys, although occasionally they open out onto broad flats.

Geologically, the region is composed of relatively young, weak schist rock masses that are prone to slumps and failures (Tonkin & Basher 1990). The combination of high rainfall, steep topography and weak rocks makes the area very susceptible to erosion and sedimentation events. Consequently, rivers in Westland have among the highest sediment loads in the world (Griffiths 1979).

Reconstructing disturbance history in these catchments is made difficult by steep-sided valleys and large, often unfordable, rivers, where a major earthquake is likely to cause such extensive damage that it would erase any record of all but the most recent disturbance. We therefore selected for study two areas of broad flats, both east of the Alpine fault, where a series of river terraces and large fan deposits has been preserved at the base of the steep hillsides (Welcome Flat and lower Karangarua study sites, Fig. 1). These study sites are less likely to suffer complete devastation during a major earthquake, and so should preserve a record of past disturbances, although the pattern may not be representative of the entire catchment.

The Welcome Flat study site (625 ha) is in the middle reaches of the Copland River, extending 5 km along the Flat from Shiels Creek to Bluewater Stream. It is surrounded by steep rocky peaks up to 2200 m high and extends from 450 to 700 m a.s.l.; above this, forest cover is usually absent (due to steep rock faces) or seral (due to frequent snow and rock avalanches and flooding).

The lower Karangarua study site (787 ha) extends along 8 km of the Karangarua River valley and lower hillsides on the true right from McTaggart Creek to State Highway 6. The area consists of broad river terraces, with numerous side-stream fans and debris deposits spilling onto these terraces from the steep hillsides above. Above the upper limit (about 300 m a.s.l.) the hill slopes were steep and dissected with few stands of trees accessible for sampling.

The forest types of the catchment are described in detail by Wardle (1977, 1980). The long-lived conifer species Dacrydium cupressinum (rimu), Dacrycarpus dacrydioides (kahikatea) and Prumnopitys taxifolia (matai) dominate the canopy below about 500 m elevation, with the angiosperm tree Plagianthus regius (mountain ribbonwood), Olearia species, and the conifers Podocarpus hallii (mountain totara) and Libocedrus bidwillii (cedar) dominant at higher elevations. All of these species form a typically open upper canopy, often as scattered trees emerging over a lower tier of small angiosperm trees or shrubs. Steep hillsides throughout the catchment carry dense stands of the angiosperm trees Weinmannia racemosa (kamahi) and Metrosideros umbellata (rata). Human modification of the study sites is limited to stock grazing on recent river terraces in the lower Karangarua and the construction of walking tracks and huts, and did not present a problem for reconstructing disturbance history.


Land surface and forest stand mapping

We identified eight land-surface types in the two study sites, five of which would have been formed by erosion or sedimentation events (landslide, debris fan, debris avalanche, terrace, sedimentation surface; Table 1). The land-surface types were clearly distinguishable in the field, with boundaries between adjoining surfaces marked by sharp changes in topography, substrate or forest structure. We mapped the location and extent of each type from aerial photographs and extensive field inspection before dividing the two study sites into different forest stands, with each stand comprising a continuous block of forest that had been mapped on a single land-surface type. Stands were further subdivided where different areas of the same land surface had obviously formed at different times, as indicated by abrupt changes in forest structure or ground features (e.g. separate but adjacent landslides) to give a total of 44 stands that ranged in area from 1 to 120 ha.

Table 1.  The eight land-surface types identified in the two study sites in the Karangarua study area, south Westland, New Zealand, and the key features that distinguish each type
Land-surface typeArea (ha)Distinguishing features
Forested surfaces
Stable hill-slope 77Stable sites, well developed soil, mature all-aged forest
Landslide 100Visible landslide scars and deposits; even-aged forest
Debris fan 289Coarse unsorted debris deposits at base of steep slopes
Debris avalanche 215Large debris deposits, below obvious source zone
Terrace 352Flat surfaces by rivers; well sorted alluvium
Sedimentation surface 139At base of hillsides and streams; layered silts/gravels indicate periodic deposition
Unforested, frequently disturbed surfaces
Present riverbed 170Bare gravels and rocks in river flood range
Exposed bedrock slope 70Bare rock surfaces beside ephemeral watercourses and avalanche paths
Total area1412 

Reconstructing disturbance history

Six conifer species (L. bidwillii, P. hallii, P. taxifolia, D. dacrydioides, D. cupressinum and Prumnopitys ferruginea (miro)) and four angiosperm species (W. racemosa, M. umbellata, P. regius and Olearia ilicifolia) formed the upper canopy of the mapped forest stands. All of these species are early colonisers of disturbed areas and all of the conifers are long-lived (500–1000 years; Ogden & Stewart 1995). We used annual rings to age canopy trees of all species, except M. umbellata (which does not have visible rings), in each forest stand and used these ages to reconstruct the history of tree establishment and past disturbance.

Canopy trees in all but six of the forest stands were scattered across the land surface or clumped in patches of 5–10 trees. In these stands, we increment-cored at 1 m above ground all trees > 10 cm d.b.h., up to a maximum of 150 trees per stand (and at least 15 trees except for four small stands which had only 7–12 trees). Each of the remaining six stands, where even-sized conifer trees formed a continuous dense canopy, were sampled by coring the largest trees encountered on a randomly located transect, up to a maximum of 150 trees, and all trees > 10 cm d.b.h. in randomly located rectangular sample plots of 0.1–0.2 ha.

Tree cores were prepared using standard methods (Fritts & Swetnam 1989), and the number of rings counted under a stereo microscope. We used the method of Duncan (1989) to estimate the number of rings in the missing part of cores that failed to intersect the tree's pith but where the growth ring arcs were visible. Cores that missed the pith by > 5 cm (estimated from the curvature of the arcs) or that had < 70% of the radius present (for rotten trees) were excluded from all analyses. Excluded cores accounted for just under 10% of all trees sampled.

Tree ages from each stand were grouped into 25-year age classes for interpreting disturbance history. A sliding 25-year age class was used to examine regeneration patterns and identify cohorts. Single-cohort stands (Fig. 2a) were characterized by a single, relatively even-aged cohort of trees, indicating a pulse of establishment following a major disturbance event. Two or more cohorts were evident in other stands. Such multiple-cohort stands (Fig. 2b) indicate establishment following two or more major disturbance events, defined by a > 100% increase in the number of trees in one 25-year age class compared with the previous age class, where in addition > 25% of trees sampled in the stand had established in the pulse of recruitment associated with the > 100% increase. The third type of age structure (Fig. 2c) had trees established at various times in the past, indicating an absence of major cohort-initiating disturbance in the last 650 years and more frequent small-scale disturbance and tree replacement.

Figure 2.

Examples of the three key forest stand age-structures identified in the Karangarua catchment, south Westland, New Zealand. (a) Single-cohort stand (McTaggart's Fan, n = 32), (b) Multiple-cohort stand (Shiels Ck fan, n = 33), and (c) Relatively all-aged stand (Rough Ck terrace, n = 15). Disturbances in the stands took place about 280 years ago in (a), and 280 and 370 years ago in (b).

We dated cohort initiation in each stand by examining the age-class distributions. For single-cohort stands, the oldest tree in the stand was used to estimate the date of cohort initiation (e.g. 266 years ago for the cohort in Fig. 2a). In multiple-cohort stands, the oldest tree in the age class in which high tree establishment commenced in each cohort was used to estimate the date of cohort initiation (e.g. 239 and 345 years ago for the cohorts in Fig. 2b). The age of the oldest tree in a cohort will underestimate the true date of disturbance because there will be a delay due to colonization and growth to tree-coring height. A previous study of the time taken for trees to colonize and grow to 1 m height on 19 surfaces with a known date of disturbance found that this delay ranged from 5 to 40 years for the species included in this study, with most close to the median delay of 28 years (Wells et al. 1999). We therefore added 28 years to the age of the oldest tree in each cohort to estimate the date of disturbance.

Cohort-initiating disturbances in the study area were most likely caused by tree fall (due to windstorms or earthquakes), or by erosion and sedimentation events such as land-slipping or flooding. We determined the type of disturbance that had initiated each cohort by inspecting each stand for evidence of erosion or sedimentation (landslide and debris deposits, buried soils, partially buried trees, dissected ground surfaces, silt deposits, abrupt topographical changes and trees established on rocks). We also recorded the microsite (log, rock or forest floor) on which each cored tree had established. We mapped the extent and boundaries of all such features in the study sites and compared these with the area of each post-disturbance cohort. Tree establishment on fallen logs or cohorts growing on stable surfaces with well developed soil profiles indicated establishment in response to one or more tree-fall events.

In single-cohort stands, we calculated the area affected by a disturbance event as the area of the stand. In multiple-cohort stands, we estimated the area affected by each disturbance event as the area spanned by trees in the associated cohort. We used the number and timing of cohort-initiating disturbance events, the area affected by each disturbance event, and the type of event (tree fall or erosion/sedimentation event) to reconstruct the history of major disturbance in the study sites. However, whereas the most recent disturbance events will be preserved in their original extent, the record of older disturbances will have been erased to some degree. Where a recent cohort overlapped an older cohort, such that remnant patches of the older cohort surrounded the recent cohort, we reconstructed the extent of the older cohort by assuming that, prior to the more recent event, the older cohort occupied the whole area between the remnant patches.

Disturbance events recorded in tree growth patterns

Trees that survive a disturbance event may record the impact of that event in their radial growth patterns, expressed as synchronous growth suppressions (caused by damage to trees) or releases (caused by the death of competitors) evident in tree rings (e.g. Veblen et al. 1994; Jacoby et al. 1997). Temperature extremes can also cause abnormal growth in trees, but synchronous releases and suppressions coinciding with widespread initiation of new forest cohorts in the area would link any growth anomalies to disturbance events.

We examined the history of tree growth disturbance in all trees of L. bidwillii in the Welcome Flat study site that could be dated to at least 1690 ad (33 trees in total). Many of these trees occurred in multiple-cohort stands where they had survived the most recent disturbance events and were therefore likely to record the impact of those events in their growth rings. We measured ring widths in cores from each tree to the nearest 0.01 mm using a Henson bench, and examined the ring-width series for growth anomalies. We defined a major growth release or suppression as a > 150% increase or decrease in mean ring width when consecutive 5-year means were compared, and identified the starting date of all such growth anomalies by ring counting from the outermost ring (cf Veblen et al. 1992). The dates of growth anomalies were combined for all trees and grouped into 10-year age classes. The trees were not cross-dated, and errors in absolute age of about ± 0–5 years are therefore possible due to locally absent or false rings (Norton 1986).


Type of disturbance events

Most of the study area (1207 ha or 86%, Table 2) comprised surfaces that had been disturbed by erosion or sedimentation (abbreviated to ES) within the last 650 years. Part of the study area (307 ha or 22% ) was unforested because it was very recently or constantly disturbed by ES events (present riverbed, stream margins and exposed bedrock slopes). Of the forested area, 711 ha (51% of the study area) comprised single-cohort forest stands. Of these, all but one occurred on recent surfaces formed by a single catastrophic ES event, with the trees established on bare ground or large rocks exposed or deposited by the disturbance. One single-cohort stand (1 ha, 0.1% of the study area) had been initiated by tree fall. Here, at least 20 trees had been uprooted and a young cohort of conifers had subsequently established. Every tree that colonized in response to this disturbance had established on the trunk of a fallen tree, at heights of 0.4–1.8 m above the forest floor.

Table 2.  Summary of the cohort-initiating disturbance events recorded in forest stands in the Karangarua study area, south Westland, New Zealand, catalogued by disturbance type and forest structure
Disturbance type and forest structureNo. of disturbance eventsPresent spatial extent
  1. ES, erosion or sedimentation.

(1) Stands disturbed by ES since 1350 ad
(i) Single-cohort forest stands
Debris avalanches 1 95 7
Landslides 6 100 7
Debris fans 2 17112
Terraces 7 27820
Sedimentation surfaces12 66 5
(ii) Multiple-cohort forest stands:
Debris fans12 14810
Sedimentation surfaces10 42 3
(iii) Unforested, frequently disturbed surfaces:
Present riverbed 17012
Exposed bedrock slopes 70 5
Other disturbed sites, lacking forest trees > 10 cm d.b.h. 67 5
Total for stands disturbed by ES within the last 650 years:50120786
(2) Stands disturbed by tree fall since 1350 ad
Single-cohort, on old debris fan 1  1 0.1
(3) Stands undisturbed by large events since 1350 ad
Old terraces 74 5
Old debris fans 53 4
Stable hill-slopes 77 5
Total for undisturbed stands: 20414

Multiple-cohort stands, covering 190 ha (13% of the study area), were present where more than one ES event had initiated tree establishment in the last 650 years and at least some trees had survived the most recent event. In all stands there was clear evidence that ES events had initiated the cohorts. This evidence included the younger cohort of trees having established in erosion channels between the older cohort of trees, older trees having large root plates exposed by erosion, and older trees being partially buried with younger trees established on the new higher surface. Of the 205 trees we aged in multiple-cohort stands, only four had established on fallen logs.

Stands for which there was no evidence of recent disturbance in their age structure covered 204 ha (14% of the study area). Most of these were on old terraces and debris fans that must have been formed by older ES events. Only 77 ha (5% of the study area) comprised stable hill slopes that had been unaffected by erosion or sedimentation within at least the last 650 years.

Timing of disturbance events

Forest cohort evidence

A total of 51 cohorts of trees (corresponding to the disturbance events in Table 2) had been initiated by disturbance events at the two study sites within the last 650 years. The estimated dates of the disturbance events that initiated these cohorts are shown in Fig. 3(a). Cohort-initiating disturbances have occurred episodically since 1350 ad with four distinct episodes of high disturbance during c. 1817–70 (12 events), 1695–1740 (17 events), 1614–67 (13 events), and 1435–66 ad (5 events) (Fig. 3a). These four episodes of cohort establishment are evident as pulses in the age distribution of all trees aged in the catchment (Fig. 3b).

Figure 3.

(a) Frequency distribution of the estimated dates of 51 cohort-initiating disturbance events in the Karangarua catchment, south Westland, New Zealand. Dates were obtained from the age of the oldest colonizing tree, plus an addition of 28 years for estimated colonization and tree growth to coring height. (b) Age frequency distribution of all cored trees (n = 831) in the Karangarua catchment.

Combined, these four episodes occupy 37% of the last 650 years, but during this time 92% of the cohorts were initiated and 90% of the sampled trees established. Although the estimated dates of disturbance events in each episode span 31–53 years, over half the events in each episode fall within a 20-year period. This suggests a strong synchrony in disturbances within each episode.

Cohort-initiating disturbances at times outside these episodes were remarkably limited: only four cohorts were preserved, all at the Welcome Flat study site (Fig. 3a; these dated from about 1350, 1580–90 and 1770–90 ad). However, areas that have been disturbed within the last c. 50 years, and that are too young to support forest trees > 10 cm diameter at 1.4 m, were also present (Table 2).

Because the trees in all but one cohort had established on surfaces formed by ES events, the four episodes of cohort initiation reflect episodes of catchment-wide, major natural disturbance during which many ES events occurred. We term these episodes, in chronological order from earliest to latest, the Ruera, Sparkling, McTaggart and Junction episodes (the names derive from locations where large surfaces associated with each episode had formed).

Tree-growth evidence

The release and suppression results for the 33 L. bidwillii at Welcome Flat are shown in Fig. 4. Over the last 450 years, four time-periods had at least 10% of trees showing a release (1610–20, 1710–20, 1820–40, 1960–70 ad), and five periods had at least 10% of trees showing a suppression (1610–20, 1710–20, 1750–60, 1820–30, 1900–20 ad). There were six decades in which > 15% of trees were affected by either release or suppression: 1610–20, 1710–20, 1770–80, 1820–30, 1900–20 and 1960–70, but only three (1610–20, 1710–20 and 1820–30) showed both as would be expected following major disturbance (see Methods). Responses seen in other decades could be explained by climatic extremes without invoking major disturbance.

Figure 4.

A combined frequency distribution of growth releases and suppressions in 33 Libocedrus bidwillii from Welcome Flat, south Westland, New Zealand. (a) Proportion of trees showing a release or suppression; the white bars represent releases, and the grey bars suppressions. A suppression/release was defined as a greater than 150% difference in mean growth rate between consecutive 5-year means. Decades marked with arrows contain > 10% released and > 10% suppressed trees. The numbers at the base record the variable sample size in each time period. The three decades of unusually high releases and suppressions coincide with the three most recent episodes of catchment-wide landscape disturbance, shown in (b), upper panel.

Each of the three decades of major release and suppression coincides with an episode of catchment-wide disturbance (Fig. 4). Given this correspondence, the three decades of unusual tree growth most likely record the impacts of the McTaggart, Sparkling and Ruera episodes of disturbance. Furthermore, each period of unusual tree growth appears to be confined to a single decade, which suggests that the numerous cohort-initiating disturbances associated with each episode were synchronous or within a few years of each other. The tree-growth disruptions therefore allow us to refine the dates of the three most recent catchment-wide disturbance episodes to 1610–1620 ad (McTaggart), 1710–1720 ad (Sparkling), and 1820–1830 ad (Ruera).

Comparison with the record of regional forest disturbance

Three region-wide episodes of landscape disturbance caused by large Alpine fault earthquakes have been previously identified in central Westland, with the earthquake dates estimated at 1717 ad, 1630 ad ± 25 years and 1460 ad ± 25 years (Wells et al. 1999), suggesting (Fig. 5) that cohort initiation for the Sparkling, McTaggart and Junction episodes in the Karangarua catchment was triggered following ES disturbance as a result of the last three Alpine fault earthquakes. In contrast, there is no evidence of a regional disturbance event that may have caused the Ruera episode (1820–30 ad).

Figure 5.

A comparison of the history of forest disturbance events in (a) sampled forest stands scattered over the Westland region of New Zealand (excluding Karangarua catchment; n = 59) (adapted from Wells et al. 1999), and (b) the Karangarua catchment study area (n = 51). The three episodes of disturbance initiated by Alpine fault earthquakes are present in both (a) and (b), and are indicated by the dashed lines.

Further evidence comes from a flight of successively older terraces that has formed above the modern river. The age of the oldest tree on the four terraces closest to the river dates their formation as coincident with the Ruera, Sparkling, McTaggart and Junction episodes (Fig. 6). Such preserved flights of terraces in Westland valleys upstream of the Alpine fault are generally accepted as being the result of episodic uplift by earthquakes along the fault (Adams 1980a; Tonkin & Basher 1990).

Figure 6.

The terrace sequence in the Karangarua River, south Westland, New Zealand. Successive terraces were abandoned at times matching the four episodes of catchment-wide disturbance. Oldest tree ages include a colonization delay adjustment of 28 years.

The youngest terrace (established about 1860 ad) differs from the others in being within the modern flood range, and is therefore strictly part of the active flood plain. Trees dating from the Ruera episode are patchily distributed as seral colonizing trees, probably reflecting partial abandonment of the flood plain at this time. The record of this terrace, however, is unlikely to be preserved in the long term, because the surface has not been colonized by long-lived conifer trees and because any future major aggradation event is likely to impact the surface, destroying the present vegetation. Hence, the terrace associated with the Ruera episode does not have the features associated with a large Alpine fault earthquake, nor does it coincide with a known earthquake date. However, sharp earthquakes were recorded about 200 km south of the Karangarua, in northern Fiordland, during 1826, where they were felt by sealers and caused considerable land slipping and forest toppling. Accounts suggest that part of the Fiordland coast was elevated at this time (McNabb 1907) and severe landscape disturbance was observed around the Cascade River, 120 km south of the Karangarua. The timing leads us to consider these earthquakes to be the most likely trigger of the Ruera disturbance episode (1820–30 ad).

Spatial extent of disturbance in the episodes

The types of disturbance and their spatial extents varied between disturbance episodes in the Karangarua (Tables 3 and 4). The current preserved extent of disturbances in each episode ranged from 6–35% of the study area. The current extent of disturbance for the Ruera episode (10%) provides a good representation of its original extent because no areas of the landscape have been devastated since this time. We estimate the original extent of disturbance to be 40% and 49% during the Sparkling and McTaggart episodes, respectively, but it was impossible to estimate the original extent of the Junction episode because over 80% of the landscape has been catastrophically disturbed since then.

The types of disturbance caused by Alpine fault earthquakes also differed from those caused by the more distant Fiordland earthquakes or other forms of disturbance. Alpine fault earthquakes impacted on a greater range of land surfaces, probably reflecting the greater intensity and severity of these events. This may explain why impacts during the Ruera episode were restricted almost entirely to sites adjacent to watercourses, whereas disturbance from the Alpine fault earthquakes impacted on all land forms.

Spatial dependence in the occurrence of disturbance events

Sites that are particularly prone to ES can be repeatedly disturbed during separate episodes of earthquake-induced disturbance (Veblen & Ashton 1978), and spatial dependence in the occurrence of disturbances over time is thus possible. In our study, several land surfaces were impacted by disturbances during more than one episode (Table 5). Unfortunately, many surfaces had been so severely disturbed during the most recent two episodes that it was not possible to determine if these surfaces had also been disturbed during previous episodes. Consequently, we were unable to determine if the frequency of disturbance was predictable from topographic or substrate variables.

Table 5.  Occurrence of disturbance events on land surfaces in the Karangarua catchment, south Westland, New Zealand, during the last 650 years. The numbers in the table give the number of land surfaces that have 0, 1, 2, 3, or 4 separate disturbances preserved
Land-surface typeNumber of disturbances recorded on the land surface
Debris fans (n = 11)3341 
Sedimentation surfaces (n = 15)1941 
Landslide surfaces (n = 3) 21  
Terraces (n = 11)47   
Stable hill-slopes (n = 4)4    


The present landscape and forest pattern in the Karangarua catchment has been determined by a history of large, severe and relatively infrequent episodes of earthquake-triggered disturbance. In the last 650 years, these episodes have occurred often enough (every 100–200 years), relative to the longevity of the major canopy trees, that they exert a dominant influence on the present forest structure. Consequently, over 50% of the study area comprises simple-structured, even-aged forests composed of first-generation trees established after major erosion or sedimentation events. Because of periodic massive disturbance to the landscape, only 14% of the study area comprises cohorts of trees old enough for gaps caused by canopy tree senescence to have resulted in a more complex, all-aged structure.

Earthquake disturbance

All four disturbance episodes identified in this study were most probably triggered by earthquakes, but the episodes caused by Alpine fault earthquakes stand out as exceptionally large-scale and severe. If we estimate conservatively that Alpine fault earthquakes recur every 250 years on average, and that each event disturbs 35–49% of the land surface (Table 4), then these earthquakes alone would have disturbed an average of 14–20% of the land surface in the catchment per century. If we include the Ruera episode, then overall earthquake-induced disturbance rates are even higher.

Table 4.  Summary of the extent and key features of disturbance impacts in the Karangarua study area, south Westland, New Zealand, during the four disturbance episodes
EpisodeArea disturbedKey features of disturbance
Ruera10%Side-stream sedimentation dominant
  No multiple-cohort stands initiated
  No debris fan disturbance
  Partial terrace abandonment; preservation unlikely
  Cause: Fiordland earthquakes
Sparkling35–40%All land forms affected by disturbance
  New terrace formation preserved
  Large-scale debris fan disturbance
  Cause: Alpine fault earthquake
McTaggart32–49%All land forms affected by disturbance
  New terrace formation preserved
  Large-scale debris fan disturbance
  Cause: Alpine fault earthquake
Junction> 6%Not possible to reconstruct original extent or impacts
  New terrace formation preserved
  Large-scale debris fan disturbance
  Cause: Alpine fault earthquake

Furthermore, this estimate of earthquake disturbance rate is likely to be low because the study sites we selected are not representative of the catchment in which we worked. Most of the catchment east of the Alpine fault comprises steep hillsides prone to catastrophic erosion events. Similar areas that we sampled in the steeper sections of our study sites were forested with single-cohort stands that had established on landslide surfaces after the most recent Alpine fault earthquake (Sparkling episode). The steep and highly erosion-prone nature of much of the catchment outside the study area suggests that the proportion of land surface disturbed during each episode in our study area most likely underestimates the total area disturbed in the catchment east of the Alpine fault. In addition, large areas of lowland forest west of the Alpine fault occur on old, stable moraines that have probably been free from major erosion for at least 1000 years. Tree fall, which can be caused by seismic shaking, will be the major type of disturbance in these areas (Stewart et al. 1998) and the disturbance pattern and forest structure may consequently differ.

In contrast to the high rate of earthquake disturbance, storms have disturbed 2–3% of the land surface in the study area since the last Alpine fault earthquake. Even if we assume that the entire area disturbed during the Ruera episode had been previously impacted by storms, then storms would have disturbed just 5–6% of the landscape per century since the last Alpine fault earthquake.

The rate of earthquake disturbance in the Karangarua study area (at least 14–20% of the landscape disturbed per century) is high compared with rates reported from other tectonically active landscapes. In Panama, earthquake-induced erosion was estimated to disturb 2% of the forest per century, while in Papua New Guinea 8–16% of the forest was denuded by earthquake-induced erosion compared with 3% from normal weathering processes (Garwood et al. 1979). A landscape history similar to that described in this study has also been documented in south-central Chile, based on the impacts of the 1960 earthquake and the estimated recurrence interval of large earthquakes (one per century). Although no long-term quantitative reconstructions have been made, earthquakes in this region exert an important influence on forest and landscape dynamics over a large area (Veblen & Ashton 1978; Kitzberger et al. 1995).

Infrequent, massive disturbance can play a central role in structuring forest ecosystems in other regions prone to such events. As in this study, large events typically create a mosaic of forest stands of varying ages that may persist in the landscape for several centuries. Nevertheless, earthquakes that trigger erosion and sedimentation events may differ from other large-scale disturbances in important ways. Sites that are prone to ES are likely to be repeatedly disturbed during separate episodes of disturbance (Veblen & Ashton 1978) and the damage caused by massive, earthquake-triggered flooding and land-slipping will be largely independent of vegetation structure and stage of development. Thus, areas are likely to be repeatedly disturbed, and damage will be largely independent of the time since the last major disturbance. This contrasts with several other large disturbances, such as fires or hurricanes, where the pattern of damage is often influenced by the structure of the vegetation, which is in turn a function of the time since the last disturbance (Hemstrom & Franklin 1982).

A further significant feature of earthquake-triggered disturbance is that the impacts can be synchronous and severe at a truly regional scale. Data for Westland suggest that the three Alpine fault earthquakes that caused major disturbance in the Karangarua catchment had a profound impact on forests over most of the region. Cohorts of trees dating from these events are found along at least 200 km of the Alpine fault, on all major land forms, and arising from both ES and tree-fall events (Wells et al. 1998, 1999). Consequently a disturbance regime similar to that we have documented in the Karangarua catchment, where periodic massive disturbance has synchronized forest regeneration and maintained much of the area in a relatively young stage of forest development, may apply over much of Westland.

Disturbance legacies and regional forest patterns

The persistent legacy of these large, earthquake-induced disturbances can also explain two features of Westland forests that have long puzzled scientists. The first of these is the region-wide over-abundance of mature conifer trees and the lack of small to intermediate sized trees (Cockayne 1928; Hutchinson 1932; Holloway 1957; see Fig. 3(b)). Previous explanations for this ‘regeneration gap’ included climatic cooling, an increased frequency and severity of droughts, and decreased precipitation (McKelvey 1953; Nicholls 1956; Grant 1963; Wardle 1963, 1978). Our results suggest that the region-wide over-abundance of mature trees and lack of small to intermediate sized trees reflects the long time (c. 280 years) since the last Alpine fault earthquake, which triggered formation of the extensive disturbed surfaces required for successful conifer regeneration.

A second feature of Westland forests are the extensive, even-sized M. umbellata/W. racemosa stands that are prominent along the steep front ranges and middle reaches of valleys over at least 250 km (Stewart & Veblen 1982a; Rose et al. 1992). It has been suggested that many of these stands originated following a large Alpine fault earthquake (Holloway 1957; Wardle 1980; Stewart & Veblen 1982a) and our study supports this idea. The three M. umbellata/W. racemosa stands that we sampled in the Karangarua catchment all dated from the last Alpine fault earthquake (Sparkling episode). Wardle (1980) also cored trees in two stands about 25 km north of the Karangarua, and dated their establishment at about 1730–40 ad, also during the Sparkling episode. Hence, many of the mature, even-sized M. umbellata/W. racemosa stands along the Alpine fault on sites prone to instability were probably initiated synchronously during the last Alpine fault earthquake. This could help to explain the widespread dieback of these stands at mid elevations in Westland (Chavasse 1955; Stewart & Veblen 1982b). Stand dieback in M. umbellata/W. racemosa forest has traditionally been attributed directly to browsing by the introduced possum (Trichosurus vulpecula) (e.g. Batchelor 1983). However, trees growing in a regionally extensive, even-aged cohort that developed after major region-wide disturbance would senesce at about the same time, thus becoming susceptible to widespread synchronous dieback from a triggering factor such as possum browse (Veblen & Stewart 1982b). Determining the location and extent of even-aged cohorts of M. umbellata/W. racemosa that date from past earthquakes may help to explain variation in the magnitude of canopy dieback that has been observed in these stands throughout Westland (Stewart & Veblen 1982b; Stewart & Rose 1988; Rose et al. 1992).


We thank Lincoln University for funding this project (Doctoral Scholarship, and New Development Fund). We are grateful for field assistance given by Steve Urlich, Louise Cullen, Bruno Pfister, Mark Yetton, and Jim and Catherine McKie, and for comments by Chris Peterson and two referees.