What is EEW?
An earthquake warning system(wp), often abbreviated to EEW, EEWS or EWS, is, strictly speaking, any mechanism that can detect the occurrence of an earthquake and some of its characteristics quickly and reliably enough to inform populations that will be affected by the event before the shaking actually reaches them, or before they begin realizing it.
Who has EEW?
Japan(wp) implemented the first nationwide system, which is still considered the state of the art. When an earthquake occurs in Japan, a large network of sensors quickly reacts and sends shaking data to a server that computes predictions, and then decides which areas will likely be affected enough to warrant a warning to the population. These warnings are then sent via a multitude of channels: TV, radio, internet, with many dedicated devices available as well as functionality built into everyday devices (such as TVs and radio receivers turning on automatically when an earthquake warning is received).
Today, other regions like Taiwan and South Korea have implemented similar networks. New Zealand and Chile have networks which react very fast, but are not always able to react as fast as an ideal EEW system would, yet, and do not disseminate this information as quickly or thoroughly as they could. The US West Coast(wp) has an experimental system that is not yet usable by the public at large. Mexico(wp), Romania, Italy and Pakistan have implemented partial networks covering some more sensitive areas of their territory.
There are also privately-run networks, which mainly cater to users running smartphone apps; sometimes, these networks deploy their own seismographs or strong-motion detectors, while sometimes they exploit the users' own smartphones (which usually contain accelerometers) as a unreliable but very large network of "crowdsourced" motion detectors.
Why is EEW hard?
EEW doesn't have to be hard, if implemented in a simple way. The smartphone applications are sometimes written by single individuals, with moderate knowlege of the subject matter, and they have proven to be effective to some degree.
When it comes to government-run systems, various hurdles arise. There is an expectation that these network will be highly reliable, provide few false positives (which are expensive in many ways), and have costs that governments will consider justifiable.
Almost every country on the planet has some kind of seismological network. One may wonder "why don't they just use those sensors and process their data really fast, to provide EEW?". There are multiple answers to this:
- Existing networks are old. Even though most data in these networks are exchanged over the internet (or dedicated links) today, they often still rely on protocols that were designed for non-realtime, defered communication: seismographs would periodically upload (perhaps via a slow dial-up link) the last few minutes of the trace they recorded, and then sub-networks would periodically link up to bigger national, and then international, networks to update the situation with their data.
- Things are only checked periodically. Specifically, in this "hierarchical" architecture, most communication protocols are still based on polling: in a fast system, you would have strong motion sensors(wp) pushing a signal "EARTHQUAKE! Intensity 3, this sensor is in Tokyo" to a central system that would immediately process the information from all these detectors. Instead, in "traditional" systems, the network of, say, Central Italy will ask its seismograph "Please give me the traces for the last 5 minutes"; then, if it detects an event from the traces, it will stores it; then after some time, the wider Italian network will ask all its sub-networks "Please send me a list of all recent events". Then, the Euro-Mediterranean network will periodically check with the various national networks. Delays add up fast.
- Calculations are slow. What an EEW system wants to do is to know, for every place in its coverage area, where there will be shaking, when, and how strong it may be. This is something that the EEW system trivially knows for the sensors that have already triggered, but that's just the input: it's already too late for the places where those sensors are. The system needs to extrapolate it quickly to yet-to-be-reached areas. Typically, this is done by first regressing (which means answering the questions: where was the hypocenter of the shaking? when did it start? how strong was it there, i.e. what was the magnitude?), and then by knowing these basic parameters of the earthquake, predicting what will happen at other locations. Both the regression and the prediction are slow tasks: the regression, in its most basic form, is a simple triangulation(wp), but this does not work in practice to provide accurate data, and instead, data from many stations are used, but this means that the solution has to be found numerically, nor algebraically: this software is an example of something that refines a numerical solution over time by improving it at each iteration, then there are things that do it by machine learning methods including deep learning, and the older algorithms can be complicated too; now, once you've done the regression, you cannot just say "okay, then places n km away from the epicenter, if the magnitude was m, will get Shindo/Mercalli x shaking", because that depends on a number of things, like how the seismic waves propagated, the types of terrain, the types of buildings... so, you have to model all that, using complicated pieces of software that contain a lot of mapping data like USGS's ShakeMap or perhaps something more "light" like CARAVAN, but in any case, unlike these systems which call themselves "near real-time" (i.e. "minutes"), this needs again to be done very quickly, in a matter of seconds.
- Decisions are hard. The numerical regression systems have a margin of error due to both the fact they have received data from only a limited number of stations, and the fact that numerical solutions take time to converge to a more "certain" result. These systems are expected to be highly reliable, and a false positive could be at least a PR disaster (while with "amateur" smartphone apps, it is more likely to be tolerated). So the algorithms need to incorporate a robust, tested decision system that only authorizes the dissemination of a warning when the chance of being right is high enough, while still pushing this to a matter of a few seconds, or else the system will be increasingly useless. Complete warning systems that can make decisions based on existing sensor networks include Virtual Seismologist, PRESTo and E-larmS.
- You need a way to send the warning. An alert can be sent to a computer using a relatievly simple program like EEWD, developed for VS and PRESTo, but this is usually not enough for alerting an actual population. Many countries have networks of sirens, but many don't, or only do in bigger cities. TV stations and radios need to be equipped with reliable systems to receive the warnings and very quickly present them to their audience, and this still won't help if people don't have their TV or radio on... which is why TVs and radios in Japan can be turned on remotely and automatically by any J-Alert(wp) warning, but most appliances sold in other countries don't have this ability. Smartphones need to have a functioning connection, and push notifications are slow when they need to be sent to a large number of users. Cell Broadcast(wp) can solve this issue, but the GSM networks in many countries don't have functional cell broadcast set up, at least not in a manner that allows sending timely alerts as in WEA/CMAS(wp). And then, in a "proper" system like Japan's, virtually everything is equipped with an EEW receiver that can take automated actions: trains, elevators, hospital equipment, railways, construction sites... these are all expensive things to replace or upgrade.
- You need education. If suddenly Italy sent an early earthquake warning, the first reaction of most people would be to run down the stairs (most people in Italy live in apartment buildings), which is the worst thing they could do, as stairwells are among the weakest structures in a building. The public needs to be educated on what to do: turn off the gas, open doorways to avoid remaining trapped... and they need to do this quickly, so you need periodic drills to make it become almost a reflex, because during a real earthquake, people's mind goes black if they are unprepared. This education, and periodic simulation, are another thing that is expensive, to the point that the USGS, at some point, announced ShakeAlert wouldn't be deployed to the general public yet, simply because their expected funding for education of the public had been withdrawn. Later on, funding was restored.