All the catalogues that need to be combined for the current study have their own version of both origin time and location of each earthquake. and several later publications propose that two records that differ by less than 16 s and 100 km refer to the same earthquake. We have discovered that this is too strict and use windows of 16 s and 320 km. Increasing the window to 25 s and 1000 km gave identical event pairings. Furthermore, the AMB-SIG catalogue only provides times to the nearest whole minute, so for that a 90 s time window is used. For each earthquake, the ISC time, all available locations (ISC, AMB-SIG, IMO, other local sources), and all available magnitude values of different types (Mw, Ms, mb) and from different catalogues/contributors are entered into a data file. This file is then used for further processing as described below. It contains some smaller earthquakes that are absent from the final catalogue, as explained at the beginning of Sect. 4 below. The counts of events according to period, region, location source, and magnitude source, in Sects. 3.1, 3.2, and 4.1, are however all made using the catalogue, instead of this data file, as we deem that information to be more relevant for the reader.

In the period 1900–1925 there are 22 earthquakes in the ICEL region listed in our final catalogue. All of these are in the AMB-SIG catalogue, and 4 are also in the ISC catalogue, originally coming from . The authors have viewed all these earthquakes on a map and checked newspapers articles for contemporary accounts of them (using the web service timarit.is mentioned in Sect. ), as well as scientific publications, in particular the report of . The result of this scrutiny is to use the AMB-SIG location for 14 earthquakes and the aforementioned report for 1 event and to adjust the location of 6 events using the methodology described at the beginning of this section. In the new catalogue these location sources have been specified as Amb-Sig, Report, and New, respectively. Finally, for the 22 January 1910 earthquake, we use the location provided by , 20 km offshore North Iceland. This source is marked as [1] in the catalogue, with details in an accompanying reference list.

In the period 1926–1954 there are 87 earthquakes in the catalogue, and their locations have been scrutinised in the same way. Sometimes we can take into account that an origin time is within a known earthquake series. For this period additional data sources are the IMO catalogue (Sect. ), as well as the reports of which often provide direct epicentres. This results in using 37 AMB-SIG locations, 9 IMO locations (marked IMetO in the new catalogue), 33 locations from the reports, 4 computed as averages of the most believable reported locations (marked Average), and 4 relocations (marked New).

In the period 1955–1990 there are 346 earthquakes in the catalogue. Having multiple local seismometers offers the possibility of computing locations from local measurements. Such locations have found their way into several of our sources, but the quality is variable. There are several journal articles stemming from this period providing locations for 39 earthquakes, and our choice is to trust these. The relevant articles are listed in the reference list in the readme file accompanying the catalogue and are specified as [2], [3], etc., in the catalogue itself. Some of the articles are also cited in Sect.  above. Available locations for the remaining 307 earthquakes were viewed on a map, with up to four locations per earthquake: from AMB-SIG; the IMO; the ISC; and one of the earthquake reports, newsletters, or bulletins. It transpired that none of these sources could be used as an overall first choice, but instead we had to select the most believable one in each case or sometimes take an average or relocate. The result was to use AMB-SIG for 64 cases, the IMO catalogue for 94, the ISC for 17, reports for 3, locations from the Skjálftabréf (Earthquake letter) (marked Letter) for 73, averages for 15, and relocations for 40.

For the period 1991–2019 the catalogue contains 826 earthquakes in the ICEL region. With the introduction of the SIL system described in Sect. , the quality of the local epicentre information vastly improved after 1990. We have viewed maps of these locations together with ISC and USGS locations, along with a background layer showing microearthquake activity. From this comparison it was evident that the errors in the teleseismic locations are in many cases of tens of kilometres (see Sect. ). The SIL locations are however accurate to a few kilometres inside the station network, and they are judged to be more accurate than the teleseismic locations in the region 63–67∘ N, 13–25∘ W. Outside this region ISC, AMB-SIG, and USGS locations are used for 118 events, 2 events, and 1 event, respectively. Inside the region SIL locations are used for 703 events and ISC locations are used for 17 (these events are missing in the SIL-catalogue), and finally there are 2 events located using published results as detailed in Sect. 5.2.

To obtain some indication of the uncertainty in event locations in the international catalogues we have looked at the variability between different catalogues, which can be considered a proxy for the precision of the locations. For 293 events in both the AMB-SIG and the ISC catalogues (period 1910–1996), the median location difference is 10 km, the 90th percentile is 31 km, and the maximum is 311 km. The difference does not seem to decrease markedly with time or with earthquake magnitude. A similar comparison between the ISC and the USGS catalogues in the whole NMAR region (4186 events, 1973–2019) gave a median of 9 km, 90th percentile of 25 km, and maximum of 284. Comparison of the ISC and SIL in the region defined in Sect.  (860 events, 1991–2019) gave a median of 4.4 km, 90th percentile of 23 km, and maximum of 150 km.

In the New Manual of Seismological Observatory Practice, recommend the use of general orthogonal regression to convert between magnitude types when uncertainties in the types differ significantly, as when estimating Mw from Ms or mb. They also recommend using a nonlinear relationship. An implementation of such a procedure is given by , which is based on , and also by , and we have chosen to follow this procedure. A proxy Mw value is computed from Ms using 5Mwproxy=exp⁡(a+bMs)+c, where Ms is the best estimate of Sect.  and a, b, and c are parameters determined by χ2 regression using MATLAB's optimisation toolbox and the formulae in Appendix B of (note that the two terms in curly braces in Eq. (B2) in the Appendix should be squared).

and also recommend weighing data points in magnitude ranges with low data frequency higher (histogram equalisation). We use a moderately weighted regression of this type: an earthquake with moment and surface-wave magnitudes Mw and Ms receives a weight of Mw+Ms-2. The effect is that the largest earthquakes weigh about twice as much as the smallest ones.

There is freedom in the regression to fix one of the uncertainties, σ(Ms) or σ(Mw), and it is also possible to fix their ratio. If the ratio is taken as 2.0, as in , the NMAR data give σ(Ms)=0.176 and σ(Mw)=0.0881.

Exactly the same method could be used to compute Mw from best estimates of mb. However the NMAR dataset contains far fewer large earthquakes than the one used by , so when this is attempted, the relationship turns out to be very slightly concave rather than convex (logarithmic rather than exponential). The nonlinearity is so slight that it can be ignored with a linear model. For earthquakes larger than about mb=5.75, an Ms value is almost always available and, as explained below, preferred. Thus a model valid for mb<5.75 is constructed and used: 6Mwproxy=a+bmb. Earthquakes in the Bárðarbunga caldera (Fig. ) exhibit a different relationship between Mw and mb than the rest of the dataset: for the same Mw, their mb is ∼ 0.15 higher. Therefore a separate model is used for these earthquakes. The relationship between Mw and Ms is also slightly different in the caldera than elsewhere, and for consistency separate models are also used in this case. The ratio used by , σ(mb)/σ(Mw)=2.5, gives σ(mb)=0.225 and σ(Mw)=0.0900.

As one might expect, the deviation in the Ms model is considerably lower than in the mb model (Fig. ). Thus Ms is used to compute a proxy Mw when it is available, for 4217 events in the NMAR region, of which 933 are in the ICEL region. In the absence of an Ms value, the mb relation must be used, for 2954 events in NMAR, of which 379 are in ICEL. Ms is available for almost all large earthquakes, the ones that are important for hazard assessment. Only three mb>5 values are used to compute proxy Mw in the ICEL region, and therefore the regression only uses data with mb<5.5 (Fig. ).

To use a somewhat round number and to have a single Mw uncertainty, the current work uses σ(Mw)=0.09 for all the models, mb and Ms, in and outside Bárðarbunga (Fig. , Table ). These uncertainty values are in good agreement with the results quoted in Sect. , although perhaps somewhat lower, which might reflect that our data are more recent and that there is continuous improvement in the quality of the global catalogues.

To study possible change in the Ms–Mw relationship or in the accuracy of the moment tensor Mw values, separate modelling was tested for a few subperiods. A slight, somewhat erratic, improvement in the accuracy was observed, but there was no significant change in the relationship. Thus it was decided to use a single model for the whole period.

Following , the variance of Mwproxy for an earthquake obtained with Ms regression may be estimated with 7σproxy2=(f′(Ms)σMS)2+σ(Mw)2=b2exp⁡(a+bMs)2σMS2+σ(Mw)2, where σMS2 is the variance estimate for the earthquake, obtained as described in Sect. , σ(Mw) is 0.09 as in Sect. , f is the model function given in Eq. (), and a and b are the regression parameters (Table ). The values of σproxy computed with Eq. () are in the range 0.11–0.24, and their rms average is 0.15, indicating that only a few earthquakes have uncertainty at the high end of the range. A similar procedure is used in the mb regression case, and the uncertainties given by the analogue of Eq. () are in the range 0.26–0.53 (rms avg 0.27). For the caldera models, the uncertainty ranges are 0.10–0.18 (rms avg 0.11) for Ms and 0.28–0.39 (rms avg 0.28) for mb. Note that because f′(Ms)<1, σproxy will be smaller than σMS.

The SIL system described in Sect.  provides two types of local magnitudes, denoted Ml and Mlw . To assess the uncertainty in these values, χ2 regression has been applied, with modelled (non-proxy) Mw magnitudes on the y axis and Ml and Mlw on the x axis with σ(Mw)=0.09, as in Sect.  (with caldera earthquakes excluded). The resulting estimates are σ(Ml)=0.47 and σ(Mlw)=0.57, far higher than the corresponding values 0.18 for Ms and 0.23 for mb. Restricting the comparison to earthquakes onshore (24 events) gave an improved σ(Ml) of 0.22 but a worse σ(Mlw) of 0.75. In all cases there is a considerable negative bias of 0.6–1.4 magnitudes, more offshore (outside the SIL network) than onshore. One explanation for the large spread and bias of the local magnitudes is that the SIL system's analysis is optimised towards robust magnitude estimation of earthquakes smaller than those of this comparison. Figure  shows the spread of the data, evidently in line with these estimates. There would be no meaning in showing the regression curves because of the high uncertainties.

The new catalogue is available in the repository Mendeley Data, as the ICEL-NMAR Earthquake Catalogue . There are three files, icel-nmar-v3.txt with the actual earthquake data, supporting-info-v3.txt with meta information, and origin-time-v3.txt. For each earthquake icel-nmar-v3.txt provides the region (ICEL or NMAR), origin time, location, Mw, Mw uncertainty estimated with Eqs. () or () as appropriate, and information on how the Mw value is computed or what its source is. When available, similar information for Ms and mb are given, and finally information on the origin time and location sources is given. For completeness, all available origin times are provided in origin-time-v3.txt. All events smaller than Mw 4 were excluded and the uncertainty was not computed for Mw<4.5 because the regression accuracy is reduced at the lower magnitudes. Hypocentral depth is not provided in the catalogue. The brittle part of the Icelandic crust in most areas is less than 12 km thick, and earthquakes of any significance will rupture the whole thickness .

To investigate the magnitude of completeness of the new harmonised catalogue for the whole NMAR region, two methods were used. Firstly, histograms with 10- to 30-year bins of the earthquake count with magnitudes exceeding different thresholds were created (Fig. ), and secondly Gutenberg–Richter models were constructed for a few selected periods and minimum magnitudes. The histograms show that the catalogue appears to be complete for Mw≥6 for the whole period, for Mw≥5.5 since 1915, for Mw≥5 since 1970, and for Mw≥4.5 since 2000. Gutenberg–Richter modelling with simple declustering indicates a magnitude of completeness of 5.5 for the whole period and 4.5 for the period after 1970 (data not shown). For the ICEL region similar histograms indicated a completeness magnitude of 5.5 for the whole period, 5 from 1915, and 4.5 from 1965.

It is interesting to compare the number of large events during the 20th century with lists of historical earthquakes in earlier centuries. Table  shows earthquakes with estimated magnitude ≥6 in Iceland or within 20 km offshore during 1700–1899, in total 17 events. In the new catalogue there are 8 earthquakes with Mw≥6 in the 20th century in the same region and 4 more in the first 2 decades of the 21st century.

In the final catalogue there are a few periods with disproportionately many earthquakes connected to tectonic activity (SISZ 2000 and 2008) and volcanic activity (Krafla region 1975–1976, Hengill 1994–1999, Bárðarbunga 2014–2015).

In the wake of large earthquakes it is possible that other events are triggered by their propagating waves. These secondary events can be missing from the international catalogues because their signal is lost in the coda of the primary event at teleseismic distances. Examples of this are two events on the Reykjanes Peninsula triggered by the Mw 6.52 South Iceland event on 17 June 2000 15:40:41 UTC, occurring 26 and 30 s later and 65 and 80 km farther west, respectively. The size of the first one was estimated to be Ml 5.5 , and that of the second one was estimated to be Mw 5.79 . Our estimated Mw for the first event is 5.5, and both Mw values have been added to the new catalogue with uncertainties of 0.4 and 0.2, respectively. These are the only events that do not come from one of the four teleseismic catalogues of Sect. 2.1.

Version 7.0 of the ISC-GEM catalogue was released in 2020. In the NMAR region it contains far fewer events than our new catalogue (168, with Mw in the range of 5.42–7.00), and no local information is used to relocate them. Non-proxy Mw magnitudes in ISC-GEM and the current catalogue are identical, but in general the proxy values differ, both because ISC-GEM uses a different regression model and because the underlying Ms and mb data may differ. The difference in the more important Ms regression curves is slight. Comparing Fig.  and the corresponding figure in , for Ms=5 the ISC-GEM curve is 0.06 higher, for Ms=6 it is 0.02 lower, and for Ms=7 it is 0.05 lower.

There are 119 earthquakes with proxy Mw common to the catalogues; of these 30 are in the ICEL region. Their ISC-GEM magnitudes are on average 0.06 lower than the ones presented here. The largest absolute difference is 0.47, and for 85 events the difference is less than 0.2. For the ICEL region, the mean difference is 0.02 and the largest absolute one is 0.26, and there are 24 events which differ by less than 0.2 magnitudes.

A few events which differ the most were investigated, and it transpired that the explanation was usually a combined effect of the regression curve difference and the underlying data difference.

The question arises of how representative the seismic activity of the catalogue period is for any period of 120 years. The answer depends on the length of the typical earthquake cycle. If the cycle is significantly longer than 120 years, our sample may underestimate the seismicity greatly, e.g. if the period does not contain a characteristic maximum magnitude earthquake. Studies of South Iceland earthquakes indicate that we may be near this critical duration of the cycle. The study of gave an average time between major earthquake sequences of about 80 years, ranging between 45 and 112 years. concluded that the South Iceland Seismic Zone (SISZ) had a total release of accumulated strain in about 140 years. estimated the accumulated strain by plate movements since the 1896–1912 earthquakes and compared this to the released seismic moment during the earthquakes of 2000 and 2008. They found that only about half of the strain had been released by these events.

For comparison with our catalogue, we estimate the potential seismic moment release in the two fracture zones, the SISZ and the TFZ, by a simplified geometric model of two transform faults parallel to the relative plate motion. The simplification is justified by the arguments of , who showed that the seismic moment of many closely spaced, short transverse faults (bookshelf faults) is equivalent to that released by a single transform fault. We also assume that almost all the seismic moment is released by the transform zones and not by the divergent segments of the plate boundary or the magmatically induced seismicity. The length of the transform zones is taken as 180 and 150 km for the South Iceland and North Iceland zones, respectively, i.e. the offset of the ridge axes. The width of the fault is taken to be the thickness of the seismogenic part of the crust, about 10 km, below which the slip is assumed to be aseismic. The spreading rate is 19 mm/yr, and the shear modulus is 20×109 Pa . The moment rate will then be 8(20×109)×(19×10-3)×(330×103)×(10×103)=1.25×1018Nm/yr. This result can be compared with the total seismic moment released in Iceland during the catalogue period, which may be estimated using the catalogue data and the completeness information of Sect. . Such computation for all earthquakes ≥Mw 4 in the area shown in Fig. 6, excluding the Reykjanes Ridge and Bárðarbunga, gives a total of 1.61×1020 Nm. Adding a simple correction for smaller events assuming the Gutenberg–Richter law with b=1 raises the estimate to 1.64⋅1020 Nm, corresponding to an annual rate of 1.37×1018 Nm/yr. This agrees quite (even surprisingly) well with the result of Eq. ().

We have constructed a new catalogue of earthquakes in Iceland and, as a byproduct, for the northern Mid-Atlantic Ridge. A general criterion for entry into the catalogue is that an earthquake has been instrumentally recorded by agencies outside Iceland. Locations of events in the ICEL region (Fig. ) have been reassessed, and proxy Mw values for earthquakes without modelled moment magnitudes have been computed. The resulting moment magnitudes range from 4 to 7.08. For the ICEL region the catalogue is reasonably complete for Mw≥5.5 for the whole period. There are 36 earthquakes of this size onshore or less than 20 km offshore, i.e. 2.8 per decade, and of these 10 have Mw≥6, i.e. 0.8 per decade.

To our knowledge, the map in Fig.  is the first earthquake map of Iceland which is not substantially confounded by misplaced events. The locations of the two large TFZ events marked with a star in Fig.  (the easternmost 1910 and the westernmost 1963) are still uncertain and controversial. Neither of them appears to have occurred on the best known structures, the Húsavík-Flatey Fault or the Grímsey Oblique Rift. suggest that the 1963 event originated on a NNE-striking fault offshore Skagafjörður, based on the distribution of recent earthquakes and the focal mechanism solutions of and . They furthermore suggest that the 1910 event originated on the eastern margin of the Grímsey Shoal. We adopt these locations in our catalogue. Distribution of epicentres and recent bathymetric data support these suggestions .

The largest events occur in the two seismic zones, where the plate boundaries are parallel to the plate movements (Figs.  and ). The distance from these events to the Reykjavík capital area, where 63 % of the population live, is some tens of kilometres, and the same holds for Akureyri in North Iceland, with 5 % of the population. However there are several towns and villages within the zones. An important future task is to carry out a detailed analysis of the seismic hazard both in these urban areas and elsewhere in Iceland. The new catalogue should prove to be an essential resource for such seismic hazard mapping.