Data complexities

Historical (observed) records of events are both short and incomplete as they depend on the presence of observers and the development of written records. Over the last few centuries there has been an increase in the frequency of observed eruptions. This is an apparent result of the global population increase and improvements in communication technologies. However, there are also noticeable drops in records of eruption onsets around the time of both world wars (Simkin, 1994), when documenting volcanic events was not of paramount importance. Subsequent chapters in this thesis introduce a catalog of eruptive events from Mt Taranaki (New Zealand). Only one historically observed eruption is included in the Mt Taranaki cat- alog, an account from the diary entries of an early settler (Turner et al., 2008a). Some of the more detailed historical records (such as Etna, Kilauea/Mauna Loa, Vesuvius, and Col- ima) contain eruption durations and estimates of eruption size/volume. However, extracting detailed information of that kind from geological records is difficult. Prehistorical eruption durations cannot be determined from geological records. Although there are methods of esti- mating eruptive volumes, which are discussed in Section 2.4, they contain large uncertainties. Additionally, tephra from individual eruptions that occur in close succession will often be merged if there is no observable organic material to separate them.

Eruption records become more incomplete looking further back in time. This is not only a result of the decline in historical accounts of eruption activity, but is also a consequence of working with geological catalogs. Each eruption adds a new layer of tephra to the landscape. New soils form after a prolonged period of time. Therefore, series of eruptions are recognized in geological records as a sequence of layers interspersed with organic material. Older tephra layers become more deeply buried over time. Consequently, more is known about the younger eruptions as they correspond to the most accessible tephra layers. The oldest tephras are often unexposed or unrecognizable so the onset of the first eruption is rarely known.

Observations of eruptions also tend to be biased towards the largest events. In historical records, this tendency is likely because the larger events caused the most damage and were worth reporting. Whereas, in a geological setting, larger eruptions leave thicker deposits with greater chances of being preserved. In addition, the expense involved in establishing records of events means that often the longer and more detailed records belong to volcanoes from wealthier countries. The more money, resources, advanced technologies, and trained specialists, the more complete the record of events is likely to be.

Available records of eruptions often represent only partial listings of the actual eruptions that occurred (Wang and Bebbington, 2012). If catalog incompleteness is not properly detected, it may result in false patterns subsequently leading to wrong interpretations. Since incomplete- ness is primarily due to short historical records, it is a difficult function to quantify (Mulargia et al., 1987). Wang and Bebbington (2012) discussed the sensitivity of hazard estimates with respect to missing observations. They adopted a hidden Markov model (HMM) framework to reflect missing onsets, and to estimate record completeness. However, the HMM developed by Wang and Bebbington (2012) requires a steady state, such that the ‘true’ or ‘underlying’ rate of eruptions does not exhibit trends over time.

Eruption behavior patterns can have a tendency to change which causes past records to become misleading. Events may occur that are unusual for a specific volcano. Changes in the size, shape, and composition of a volcano over time, may change hazard estimates (Decker, 1986). Even volcanoes with well documented records of many historical eruptions can show a wide variation in repose times, and large variations in eruption characteristics (Decker, 1986). This problem was considered by Wickman (1966), who first illustrated that while some volcanoes show a random pattern in the timing of their historical eruptions, others show patterns of increasing or decreasing probability of eruption depending on the time window considered. Wadge (1982) suggested that while a volcano can experience periods of steady state behavior, where phases of quiescence and activity do not change for a few hundred years, over longer periods of time the assumption of steady state behavior can be unsustainable. Identifying and modeling departures from stationary behavior can further complicate methods of forecasting eruption occurrences.

Many eruptions do not leave a long-lasting deposit. Erosion can result in the complete

disappearance of some tephras in different environments. Unexposed locations, such as lakes

and swamps, provide ideal sites for observing very fine ash only millimeters thick. Sediment

cores extracted fromunexposed locations have a higher tephra preservation rate thanexposed

surfaces (such as cliff faces, outcrops, road cuttings and drains). However, the coring process for extracting sediment columns can disturb tephra layers and embedded deposits can be altered by twisting and compaction. Figure 2.3(a) shows an example of a small portion of a sediment core extracted from Lake Umutekai containing tephra sourced from Mt Taranaki. Figure 2.3(b) shows tephra beds exposed on the flanks of Mt Taranaki.

Exposed locations are vulnerable to rapid erosion and reforestation. Therefore, they only preserve thicker tephras from much larger eruptions. In addition, human impact on previously

Figure 2.3: Comparison between records from (a) unexposed and (b) exposed locations. Photos courtesy of Shane Cronin.

(a) Sample portion of a sediment core extracted from Lake

Umutekai, Taranaki, New Zealand.

(b)Tephra section from the Kaupokonui Valley on the flanks

of Mt Taranaki, at approximately 1400 m elevation.

unmodified landscapes can change the configuration of layers causing tephra to be reworked or displaced. Tephra thickness and grain-size are also affected by variations in topography. Different sized catchment areas in the case of unexposed sites, and diversity of surrounding landscapes, can result in over- or under-thickening of tephra layers.

Larger eruptions are expected to leave thicker deposits, thus tephra thicknesses are important for understanding the size of previous eruptions. Most models for estimating eruption size (or volume) assume thickness trends are dictated by eruptive and atmospheric conditions, rather than depositional processes (Engwell et al., 2013). Understanding and quantifying the uncertainty in thickness measurements is difficult. Little attention has been paid to variations in tephra preservation from site to site. Throughout this thesis particular attention is paid to these site-specific effects for over- or under-thickening of tephra.

Multiple sites are needed to build the most accurate composite tephra record. Not all tephra producing eruptions leave a deposit in the same place, so it is preferable to sample from

multiple locations in different dispersal directions, both proximal and distal to the vent. In practice, difficulties in finding suitable sites for tephra preservation and the associated costs, means establishing a comprehensive record of events is challenging. Merging records obtained from multiple sites is complex. Common events and site-specific gaps must be recognized to avoid over or under inflating the true frequency of events. Establishing age estimates for the tephras is useful. However, the error associated with age estimates allows possible overlap among multiple tephras deposited at different locations. Separating tephras on the basis of geochemistry can help, but it is not a fail safe solution. Some volcanoes exhibit cyclic variations in deposit chemistry, where tephra fingerprints are ‘non-unique’ (Turner et al., 2011a). Distal records may also contain tephra erupted from multiple volcanoes. This poses the additional task of identifying the source of each tephra.

It is important to recognize that available data is only a sample of the eruptions that have oc- curred. Many eruptions may have gone unnoticed or unrecorded. Even if numerous eruption records are available, merging them can cause eruption frequencies to be over-represented. Alternatively, if only one record is used the true frequency may be underestimated (Turner et al., 2009).