Scientists have described the prediction of earthquakes as an impossible task. The best one can do is determine the possibility, and thus forecast, of when earthquakes may occur. Nevertheless, while exact prediction is not (currently) possible, advancements have been made.
Where progress has been made is the use of complex simulation and modeling techniques to better forecast the occurrences of earthquakes. For instance, machine learning, using Gradient Boosted Regression Trees, that incorporates training data, has been used to better determine spatiotemporally complex loading histories within subduction zones. The model employed earthquakes ranging in different magnitude (Mw 6.2–8.3) at a given place and showed that estimation on the timing and place of occurrence could be better forecasted than other techniques. This has been deployed to laboratory earthquakes, which are more controlled and likely more predictable than natural earthquakes.
Increasingly, forecasting itself is seen as a potential pitfall in better determining when earthquakes may occur. Most previous techniques used time, measured in days or months, rather than natural time, which looks at events and the interval in which they occur. Nowcasting tries to utilise how the Earth’s more natural temporal cycles and patterns of earthquakes may relate. Machine learning and statistical procedures have been used to assess the likelihood earthquakes would occur in given regions with some success by simply looking at cycles of activity.
Although overall, full-scale prediction is difficult for earthquakes, aftershocks are proving potentially more predictable or at least forecasts for them have demonstrated better predictive results. In one recent study, a machine learning neural network model, that uses deep learning techniques and patterns of past aftershocks, was not only far better able to forecast the timing ofaftershocks but showed that 98% of stress-related variation in faults helped to demonstrate the results seen. Overall, this improves post-earthquake recovery better prepare for aftershock dangers in a given area.
Other improvements have been also applied to early warning systems. Signals along the ocean can sometimes be deceptive as to the occurrence of earthquakes, sending false alarms regarding tsunamis and tectonic events. Using a random forest technique, Pwaves emitted by past earthquakes can be used to train the machine learning algorithm to better recognize the energy wave and differentiate it with other waves that may trigger a false alarm. Effectively, this better distinguishes what an earthquake-related wave is, where in the study the accuracy was over 99% as to what waves constituted those related to earthquakes.
While natural earthquakes are often the main concern, another recent focus area has been on induced earthquakes that occur in fracking or wells that have pressurised water added. Small faults or the underlying sedimentology and lithology can greatly affect the likelihood of an earthquake. A logistic regression machine learning model was used to show how hydrologic and geologic features combine to create such earthquakes and the likelihood of such earthquakes in induced areas.
Earthquakes are a major concern in increasingly populated regions. While the quest for predicting them has long been a goal of science, only recently have forecasting tools improved earthquake forecasting. Some of these advancements have included not only the main earthquake event but also aftershocks and artificially induced earthquakes that can threaten population zones. Overall, there is no easy solution for earthquake prediction, but machine learning in particular has made forecasting far better.
 For more on using machine learning to forecast earthquakes, see: Corbi, F., Sandri, L., Bedford, J., Funiciello, F., Brizzi, S., Rosenau, M., & Lallemand, S. (2019). Machine Learning Can Predict the Timing and Size of Analog Earthquakes. Geophysical Research Letters. https://doi.org/10.1029/2018GL081251.
 For more on nowcasting, see: Williams, C. A., Peng, Z., Zhang, Y., Fukuyama, E., Goebel, T., & Yoder, M. R. (2019). Earthquakes and multi-hazards around the Pacific Rim. Vol. II.
 For more on aftershock forecasting, see: DeVries, P. M. R., Viégas, F., Wattenberg, M., & Meade, B. J. (2018). Deep learning of aftershock patterns following large earthquakes. Nature, 560(7720), 632–634. https://doi.org/10.1038/s41586-018-0438-y.
 For more on detecting earthquake waves in early warning systems, see: Li, Z., Meier, M.-A., Hauksson, E., Zhan, Z., & Andrews, J. (2018). Machine Learning Seismic Wave Discrimination: Application to Earthquake Early Warning. Geophysical Research Letters,45(10), 4773–4779. https://doi.org/10.1029/2018GL077870
 For more on inducted earthquake prediction, see: Pawley, S., Schultz, R., Playter, T., Corlett, H., Shipman, T., Lyster, S., & Hauck, T. (2018). The Geological Susceptibility of Induced Earthquakes in the Duvernay Play. Geophysical Research Letters, 45(4), 1786–1793. https://doi.org/10.1002/2017GL076100