Construction of earthquake location uncertainty maps for the Netherlands

A large part of the Netherlands is covered by a thick blanket of unconsolidated sediments. In the underlying more consolidated part of the upper crust, various anthropogenic activities are taking place. Oil and gas are produced, salt is extracted through solution mining, liquids and gasses are stored and geothermal heat is exploited. All these activities can alter the stress field around existing faults and therewith potentially induce seismicity. Over the years, an extensive seismic network has been installed to monitor seismic activity. This network has a dense distribution of sensors in regions with past subsurface activity and is more sparse in other regions. In this report, the network capabilities are assessed for the September 2021 network configuration, to find out whether this network suffices to monitor current and future subsurface operations. Also, results are shown for the September 2022 network configuration. New tools are developed to assess location thresholds and expected location uncertainties. The main seismic network-assessment parameter is magnitude of completeness. This parameter describes the spatial variation of the minimum magnitude for which almost every earthquake can be located. The location of an earthquake can be determined if the signal is detected on at least three sensors. As soon as an event can be located, it is important to know with which precision. Knowing the depth uncertainty is important, e.g., to discriminate induced events from tectonic events. Knowing the epicentral uncertainty is important, e.g., to assign the event to one of multiple nearby mining operations. By knowing the expected location uncertainties pertaining to the current network, decisions can be made on the need of additional sensors. In this report, maps are developed that show the expected location uncertainty over the Netherlands for upper-crustal seismicity. An important prerequisite is a travel-time uncertainty model, which is derived from a large collection of records from induced events. Uncertainty in travel time is mapped to uncertainty in location using a Bayesian framework. Location is considered using differential P-wave arrival times over the network (P-delays), and also by using the travel-time di erence between first P and S waves at each receiver (P-S delays). It is shown that both data attributes yield complimentary information on the location problem and that a joint inversion with both P- and P-S delays yields smallest location uncertainty. For this specific combination of data attributes, the location-uncertainty maps are computed for local magnitudes of 0.5 until 4.0, with steps of 0.5. The maps have been computed over a grid of 1x1 km. For each scenario event, the location probability density function has been determined, both as function of depth as as function of the horizontal coordinates. The assumption is made that these probability density functions can be approximated well as being (multivariate) normal distributions, such that they can be described with standard deviations. These standard deviations are shown on the maps. From the standard deviations the size of, for example, the 68% and 95% con dence zones can be computed. The maps have been computed for a source depth of 3 km. Besides, a sensitivity analysis has been performed to evaluate how the location uncertainty varies as function of di erent parameters. When the source is placed at depths larger than 3 km, the epicentral uncertainty increases, and also the depth uncertainty somewhat increases. Signal-to-noise levels play an important role in the determination at which sensors a potential earthquake is likely detected. The signal is modeled through deriving P-wave ground-motion prediction equations that describe the expected P-wave amplitudes as function of magnitude and distance from the source. The noise level is based on receiver-specific recordings of background seismic noise in the frequency band that is used for detecting induced events (5 to 40 Hz). As the location-specific noise level, the 90th percentile of the recorded noise distributions is used, meaning a noise level that is exceeded 10% of the time. The magnitude-of-completeness and location-uncertainty maps show large variations over the Netherlands. The main underlying factor is the large variation in sensor density. Another important factor is the position of the sensor, i.e., on the Earth's surface or 200m below the surface in a borehole. On average, in the detection frequency band, noise levels are a factor of 26 lower at 200m depth than at the Earth's surface. On the other hand, signal levels are a factor of 4 lower at 200m depth. Combining the two effects, this results in a signal-to-noise gain of 6.5 at the 200m depth sensors. Moreover, the local setting is important. A regional network in South-Holland performs worse than a similar network in Twente because of much higher seismic noise conditions near Rotterdam and the Hague than in the north of Twente. The sensors with the best detection capabilities are located on hard-rock conditions in the southernmost part of the Netherlands. The report contains magnitude-of-completeness (MoC) maps for 2021 and 2022. The underlying values can be obtained at KNMI (2023a). The coming years, updates will continue to appear at the KNMI Data Platform. Previous maps were made with assumed average noise conditions over the seismic network. The new maps have been made with recorded station-specific noise conditions, which yields a more accurate representation of the MoC that can regionally be achieved. On the new 2021 and 2022 maps, MoC is smaller or equal to 2.0 in the Netherlands. In areas with dense sensor distributions (Groningen, parts of Friesland, Drenthe, Twente, south of Limburg and region around Alkmaar) the MoC is lower than 0.5. The network in South-Holland does not reach MoC=0.5 due to high local noise conditions. In the middle of the country, there are not many seismic sensors and the MoC has values between 1.5 and 2.0. For the computation of the MoC, the 90th percentile of the noise distribution has been used. As a consequence, in about 10% of the cases the stated MoC can (just) not be reached. Location uncertainty reduces with event magnitude as a better spatial distribution of receivers detects the event for higher magnitudes. Within the dense regional networks, the epicentral uncertainty is within a few hundred meters. The uncertainty grows to a few kilometers at the edges of the regions where location is just possible. Depth uncertainty behaves quite dfferently from epicentral uncertainty. In general, the depth is less well constrained than the epicenter. For constraining the depth, it is important to have receivers nearby the epicenter, so that the hyperbolic part of the travel-time curve is sampled well. For constraining the epicenter it is more important to have a good azimuthal distribution of receivers. For accurate depth estimation, local velocity information needs to be available. For estimating epicenter, one can mostly rely on background or event-derived travel-time models. The computed location uncertainty is less well represented for source-receiver con gurations for which the azimuthal gap is larger than 250 degrees. The azimuthal gap is the largest angle between two receivers as seen from the epicenter. When the azimuthal gap is larger than 250 degrees, the location probability density function cannot be approximated well as being a multivariate normal distribution. The largest uncertainties occur when the receivers are nearly on one line, which corresponds to an azimuthal gap of nearly 360o. In this case, location can be ambiguous and the polarisation of P-waves is needed as an additional constraint to remove this ambiguity. In the appendix, the location-uncertainty maps can be found for the Netherlands seismic network state of 2021 and 2022. Also the underlying database, with the values at each grid point, is publicly available (KNMI, 2023b). This database is to be consulted if the precise values at a certain location are of interest. The coming years, updates will continue to appear at the KNMI Data Platform. The developed tools also provide a means to study design options for future extensions of the network. A scenario layout can be checked for its capability in terms of the minimum magnitude of locatable earthquakes and their associated uncertainty in location.