The variety of tsunami sources, together with differences in their frequency-magnitude characteristics and geographical distributions, greatly complicates the task of evaluating their global or regional frequency-magnitude distributions. A number of cautionary points need to be borne in mind:

1. Earthquake-generated tsunamis are the most common, and dominate the historical records and compilations such as the NGDC catalogue on which the Risk Atlas (see link to left) is based. However, the upper size limit on earthquake generated tsunamis is well-defined and the largest such events (such as the 1960 Chile and 1964 Alaska tsunamis) are most probably represented in the historical records. Furthermore, the large normal-fault, reverse-fault and especially subduction zone thrust fault earthquakes which produce these tsunamis have a well-defined spatial distribution. Most have occurred around the Pacific Rim and in Indonesia, with smaller concentrations in the Mediterranean region and in the Caribbean, with the result that these regions have experienced most historical tsunamis. However, because of the maximum magnitude cutoff of tsunami earthquakes, this does not mean that there is not a substantial tsunami hazard in other regions due to larger but less frequent tsunamis, potentially underrepresented in the historical record, due to other mechanisms. This is best illustrated by consideration of the tsunami hazard in the North Atlantic region (see below).

2. An important regional variation, even between regions with high levels of seismic activity, results from differences in the fault types.

Most notably, whereas most of the Pacific Rim is a region of high tsunami hazard due to the occurrence just offshore of large subduction zone thrust faults, the strike-slip plate boundary along the coast of California does not produce many major tsunamis because the dominant faults are characterized by strike-slip motion: the only major local earthquake-generated tsunamis are produced by thrust faults in the Coast Ranges west of Los Angeles.

3. There is some evidence for other regional variations in the efficiency of earthquakes as tsunami sources, in the Pacific region in particular. Analysis of 20th Century Pacific Ocean tsunamis by V.K. Gusiakov (unpublished, but see http://omzg.sscc.ru/tsulab/ website for progress on the catalogue concerned) suggests that certain regions are characterized by more efficient tsunamigenic earthquakes. These include Central America, parts of South America including Chile, the Alaska-Aleutian arc, southern Japan and Indonesia. Other regions, in particular oceanic island arcs such as the Mariana arc and the Tonga-Kermadec arc, rarely produce major tsunamis even though they do produce large subduction zone earthquakes. In part this variation reflects the regional variation in the distribution of tsunami earthquakes: Gusiakov has suggested that the variation also reflects the distribution of thick sedimentary sequences in subduction zones, which are implicated in large near-surface coseismic movements with relatively little release of seismic energy (see Causes - Earthquakes for further discussion). Essentially, Gusiakov argues that earthquakes in the regions of more efficient tsunami generation share, to a lesser extent, the characteristics of tsunami earthquakes.

4. The distribution of tsunamis due to stratovolcano lateral collapses partly parallels that of earthquake-generated tsunamis, because this type of volcano is formed by volcanic activity above subduction zones. However, on most continental margin subduction zones (Peru-Chile, Mexico, Cascadia, eastern Alaska, Kamchatka, Sumatra, for example) the volcanoes are well inland and the tsunami hazard due to these is low, although non-zero: mudflows, debris avalanches and pyroclastic flows from such volcanoes may reach inlets and enclosed bodies of water, and generate local tsunamis. In contrast, in regions such as the Aleutians, Kuriles, Bonin Islands, eastern Indonesia and most of the south west Pacific, and the eastern Caribbean, the volcanoes form islands or are submarine, and in these situations lateral collapses and a wide variety of volcanic eruptions can generate tsunamis. Some of these may be very large, although most are likely to be comparable to those which occur in the historic record.

5. In contrast, the distribution of oceanic island lateral collapses follows that of true oceanic island volcanoes, which rather than being located on plate boundaries are mainly found above mantle plumes or "hot spots". Notable groups of such islands include the Hawaiian islands in the Pacific, the Comores and Reunion in the Indian Ocean and the Canaries, Azores, and Cape Verde Islands in the northern and central Atlantic. No historical examples of such collapses exist and the number of tsunami deposits associated with them (even controversially) is very low. Estimates of the frequency of these events are based upon the occurrence and ages of lateral collapse scars on the islands themselves; of debris avalanches offshore; and of giant turbidite deposits associated with the debris avalanches. About 20 collapses occur in every million years in the Hawaiian Islands, and 10 or more per million years in the Canary Islands: other archipelagoes have not yet been studied in sufficient detail to establish rates, but a global average might be one per 20 000 years. Although this frequency is low, the tsunamis produced in these events are likely to be far larger than any in the historic record: furthermore, it is possible that environmental factors and/or random chance mean that the present-day oceanic island collapse hazard is much higher than average. Furthermore, the large number of active and potentially unstable volcanic islands in the Atlantic Ocean may mean that oceanic island lateral collapses form a major component of the tsunami hazard in this important region.

6. Fjord-type tsunamis, produced by rockfalls and landslides from steep fjord walls, are for obvious reasons restricted to a number of specific areas: Chile, Alaska, New Zealand and Norway. Other localized tsunamis due to (for example) delta front sediment landslides, rockfalls into reservoirs and so on are less common but much more widely distributed.

7. Large-scale submarine landslides, although they also occur on the trench slopes above subduction zones, are a particularly important component of the tsunami hazard offshore from the mouths of major rivers (such as the St. Lawrence river, Mississippi, Amazon, Indus and Ganges) and along passive continental margins such as those along most of the Atlantic margins outside the Caribbean. In these regions thick sequences of unstable sediment may accumulate for long periods and then be disturbed by one of the infrequent major earthquakes (as in the case of the 1929 Grand Banks tsunami) or by environmental factors to produce a large or giant tsunami. The result will be a frequency-magnitude distribution of tsunamis in these regions that is very different from that of the Pacific Rim. Sedimentological evidence, largely from deep-sea drilling and coring, indicates that the frequency of the largest of these events in, for example, the North Atlantic, may be of the order of 20 to 100 per million years, with smaller events being more common, but this long - term average may be largely meaningless in view of the potential for environmental controls on these events (see Future Directions in Tsunami research).

8. Plotting of the distribution of meteors and of small meteorites which explode in the upper atmosphere in fireballs recorded by, amongst others, early warning satellites (Tagliaferri et al., 1994) has confirmed the inference from astrophysical arguments that the distribution of impacts over the surface of the Earth should be essentially random. This is in strong contrast to the distribution of impact craters, which is controlled largely by variations in the age of different parts of the continental crust: older regions have more impact craters. As a result, the impact - generated tsunami hazard will vary largely with the size of ocean: about 35% of all impact-generated tsunami will affect the Pacific Ocean since it occupies about 35% of the Earth's surface, for example. However, many of these impacts will be far out to sea and waves from smaller impacts in particular may largely dissipate before they reach major coastal population centers. In contrast, an impact in the Caribbean or Mediterranean is less probable but is more likely to cause catastrophic damage to an immediately-adjacent coastline. The central problems in assessing the tsunami hazard associated with impacts remains, however, the prediction of the efficiency of generation and propagation of these tsunamis (see Hills et al., 1994, for a pessimistic view) and the assessment of the overall probability of impacts of different sizes. This is based upon estimates of cratering rates on the Moon and other planets, estimates of the numbers and sizes of asteroids, and so on (see the book edited by Gehrels (1994) for a recent overview of these problems).

With all these caveats and cautions taken into account, however, it is nonetheless useful to consider frequency- magnitude plots for historical tsunamis in different regions and an order-of-magnitude plot that considers the distribution of events larger than those represented in the historical record, due to different tsunami source mechanisms and based upon the various geological and other data sets discussed above.

This diagram indicates that, whilst at low magnitudes (below tsunami magnitude = 6) the commonest cause of tsunamis is earthquakes, at higher magnitudes different mechanisms are predicted to take over, in succession: submarine landslides, volcano lateral collapses and, at the highest magnitudes, impacts.

An important corollary of this is that whilst the Pacific Rim, Caribbean and Mediterranean are most at risk from the small, frequent earthquake- enerated tsunamis, the pattern changes for the distribution of risk from larger, less frequent tsunamis in the magnitude range of 7 and above. The level of risk from events in the high magnitude range may be as high in the Atlantic Ocean as in the Pacific, particularly bearing in mind the much lower level of preparedness along the Atlantic coasts as compared to the Pacific (see Tsunami Mitigation).