System Update Log
A plain-language summary of significant changes as we improve Intensity Lab. Minor fixes and behind-the-scenes work aren't listed here.
- Four Japanese earthquakes have joined the validation suite: Chuetsu-oki 2007, Iwate-Miyagi Nairiku 2008, Niigata-ken Chuetsu 2004 and Western Tottori 2000. They were always part of the set the Broadband Platform developers curated, and were missing here for one reason: each needs a crustal velocity model for its region of Japan, and we had only the Californian ones installed. Adding those two models brings the published suite to seventeen earthquakes and means it now covers every event SCEC tabulates in its own published results, so more of our rows can be read beside theirs.
- You can now choose any of the seven simulation methods for a validation earthquake from the command-line sweep, not only Graves & Pitarka. The validation page has always counted method-event pairs, but only the Graves & Pitarka column had ever been run, so every other cell was blank because nothing had run it — not because anything failed. Blank cells still mean not yet run.
- Simulations that run several realizations now have a results page you can read. A realization is one run of the same rupture with a different random seed, and each one produces its own copy of every figure and data file — so a run with many realizations used to fill the page with hundreds of near-identical images. The page now shows one realization in full and names it, with a picker for the others and a table of each realization's seed, status and run time. Every realization is still in the downloadable archive, filed by realization.
- Each site now has a response-spectra panel covering all realizations at once, showing the median and the full range across them. The spread between realizations is the reason to run more than one, and it is the one thing a single result cannot tell you.
- Validation runs can now use more than one realization. They were previously limited to one because every realization would have produced an identical result; that has been fixed, so a validation ensemble now varies as it should.
- Validation runs now compute goodness of fit by default. This was off unless you turned it on, which meant a validation run could finish successfully and produce no comparison against the observed recordings — the one result a validation run exists for. If you ran one before August 17 and the comparison was missing, that is why; re-running it now will produce it.
- Each site on a validation run leads with the comparison against the recording — the observed and simulated waveforms overlaid, and their response spectra — rather than burying them below a plot of the simulation alone. Plots of simulated-only motion now say so.
- My data now shows what each result set is costing you. Every dataset lists its size, how long it has been stored, how much longer it will be kept, and the storage charge accrued so far. Storage is billed by the day against the files that actually exist, so results you delete stop costing you from that moment rather than for a whole pre-paid window.
- You can now delete your own simulation results. The confirmation step tells you what will be removed, what keeping it costs per month, and roughly what re-running the simulation would cost — so the choice between keeping a dataset and regenerating it later is an informed one. Deletion is permanent: download anything you still need first.
- Every stored file now shows its SHA-256 checksum, so you can confirm a download arrived intact with "shasum -a 256 <file>". The same checksum already appears in each run's metadata file.
- Simulation results are now kept for a fixed period rather than indefinitely. Your data page and job list both show a countdown, and anything close to expiring is highlighted. Nothing stored before this change expires without a clear warning period first.
- The validation results page now reports method-event pairs rather than events. Validating ground motion simulation is per method: a goodness-of-fit number describes one simulation method run against one earthquake, and the Broadband Platform offers seven of them. Saying "13 of 13 events" read as completeness while showing one of seven columns, so the page now says how many of the 91 method-event cells have been run.
- All seven simulation methods can now be chosen for a validation earthquake, not just Graves & Pitarka. Each uses the same module chain SCEC ships in its own accepted workflow for that method.
- The page now names SCEC's own published results for this same Broadband Platform release, links the dataset, and explains why our numbers are not directly comparable to theirs: SCEC averages 50 simulations per earthquake where we run one.
- Verification & Validation now has two halves, because the word means two different things here. /validation/events/ is new: it shows how our simulation service scores on the historical earthquakes the Broadband Platform developers curated for exactly this purpose — thirteen US earthquakes across California, each packaged with the ground motion actually recorded at the time. We re-run the whole set for every release of the service and publish what we get, including what each run cost. The earthquakes we processed as they happened have moved to /validation/earthquakes/, unchanged; /validation/ is now a page explaining the difference and linking to both.
- Every one of those validation earthquakes is now something you can run yourself from the simulation dashboard, and each row on the results page links straight into a builder with that event already chosen. The cost column shows what the run charged us, which is also what it would charge you — useful if you are weighing running the Broadband Platform on your own hardware.
- Fixed: simulations of the validation earthquakes other than Northridge finished successfully and were then reported as failed, and none of them produced the goodness-of-fit numbers they were run for.
- The home page now leads with what this site is for: moving past a single magnitude number to site-specific seismic intensities, and the information those provide for deciding what to do after the shaking stops. It also states plainly how we get them — observed intensities at seismic stations, combined with intensities estimated from ground motion simulations at populated places. The Our Vision button is back, and a new About the BBP API page explains why we simulate ground motion rather than only predict it, and what a simulated time series tells you that a single peak value cannot. The headline is also easier to read against the map behind it.
- Corrected the Physics-Based Simulation page, which still described a distance limit on which earthquakes can be simulated and said that most of Nevada, Oregon and northern Mexico fell outside it. That limit was removed some days ago and the page had not caught up. An earthquake qualifies on magnitude — M5.0 and above — and nothing else; one outside a velocity model's home region is simulated against the nearest available model and labelled an approximation rather than being refused. The 200 km figure that remains is a different measurement: how far from the epicenter you may place an individual site.
- The simulation dashboard has been rebuilt to follow the same steps as the Broadband Platform's own command-line program: mode, velocity model, method, source, stations, then options. If you have run BBP from a terminal, the questions and their order should look familiar. An Expert mode switch reveals the advanced settings, matching that program's expert option.
- You can now design an earthquake of your own rather than only starting from a historic one. Every rupture parameter — magnitude, fault dimensions, strike, dip, rake, hypocentre, seed — is editable, and choosing a real earthquake simply fills them in for you. The Simulation Lab now opens with a Create Earthquake Simulation button rather than a list of earthquakes; the list is still there below as a set of worked examples, and Select Historic Earthquake in the builder takes you straight to it.
- The suggested velocity model now follows the earthquake you describe. Move the epicenter and the suggestion updates — a rupture in the Bay Area will suggest the Northern California model rather than keeping a Los Angeles one. If you have picked a model yourself, your choice is left alone.
- Source description (.src) and station list (.stl) contents can be pasted in, and downloaded back out. Blank templates are offered for both. The files you get are the exact files the simulation used, so a run configured here can be taken to a local BBP install and back. If what you paste contains settings this service cannot carry, it names them rather than dropping them quietly.
- Every recording site now needs a Vs30 value, and the builder no longer fills one in silently. Clicking the map looks one up and shows which model it came from; you can also type your own. Vs30 is the average shear-wave velocity in the top 30 m, and it is the only site property that changes the simulated seismogram — it sets how much the soil column amplifies the shaking arriving from below, so softer ground means stronger motion at the surface. A number nobody chose would quietly change every waveform, which is why it is now asked for.
- More of the platform's own settings are now yours to choose: how many realizations to run, the seed strategy, which post-processing products to compute, and per-station filter frequencies. The cost estimate updates to match, so you can see what a choice costs before making it.
- Every completed simulation now comes with a run metadata file describing what produced it — the rupture, the method, the velocity model, the stations, and where each site's Vs30 value came from. A copy travels inside the results archive itself, so the data explains itself wherever it ends up, and a fuller version is shown on the results page and downloadable beside the data. That one also records the archive's checksum, so the two stay matched up after either is renamed or filed away. Until now the downloaded data did not say what had been simulated to make it.
- The home page has been rebuilt around the four things the site offers: intensity reports for recent earthquakes, shaking animations, the simulation dashboard for running your own physics-based ground motion, and the verification work that checks those simulations against real recordings. The list of recent earthquakes now has a page of its own at /events/, and the Events link in the menu goes straight there.
- New Verification & Validation page at /validation/, listing every earthquake we have compared simulated shaking against the seismograms actually recorded, with the bias, the number of stations, and a link to the full report. Reachable from the Learn menu. These comparisons already existed one earthquake at a time; this is the first place you can see them together.
- The simulation dashboard is now reachable from the main menu for everyone who is signed in. You can browse historic earthquakes, build a simulation, place recording sites on a map and see what the computing would cost, all at no charge — running the simulation is what requires an approved account. Signing in now takes you to the dashboard rather than your account settings.
- Your account page has been reorganized around simulation: how much computing you have used and what it cost, links into the dashboard and your saved results, and your API key. Registered sites and notification settings are still there, tucked into sections you can open when you need them. You can also set a display name, when registering or at any time afterwards.
- Simulated seismogram plots now show the part of the record that contains shaking. The simulation always produces just under seven minutes of record regardless of how close the sites are, so most of each plot was empty. The length is chosen per simulation from where the shaking actually ends, and every station in a simulation shares the same time axis so their durations can be compared. Downloads are unchanged and still contain the complete record.
- Fixed: a paragraph of internal notes was showing on the page while a simulation was running.
- Physics-based shaking simulations now use a crustal model matched to the local geology across far more of California — the Los Angeles basin, the San Francisco Bay Area, Central California and Parkfield, the Mojave and Ridgecrest area, and the eastern Sierra and Owens Valley. Earthquakes in these areas get more realistic simulated shaking, and events outside the modeled areas are clearly labeled as approximate.
- The Technical Information page has been split into focused topic pages — intensity calculation, data sources, place/population data, seismic stations, GMPE comparison, physics-based simulation, animations, audio narration, the local language model, and API services — reachable from a new index at /technical/.
- Event pages now carry a "BBP Simulation Comparison" panel showing how closely the physics-based simulation of that earthquake matched the seismograms the network actually recorded. Previously this comparison was only available as a PDF download.
- Each comparison also has a full report page of its own, with the station-by-station summary, the residual at each spectral period, maps and diagnostic plots, and the waveform overlay for every station. The PDF is still downloadable from both places.
- The result is now reported period by period rather than as a single average. A simulation can match the recordings on average while being well off at a particular period, and structures respond at their own period — so any period outside the declared band is flagged, and the limits of the comparison are stated above the numbers rather than in a footnote.
- Fixed: the homepage and the all-events list could come up empty when the site was running against a historical event archive rather than live data. Both now scale their time window to the most recent event available.
- The Technical page now opens with a "Methods at a Glance" summary of the methods behind Intensity Lab's site-specific ground motion: the SCEC Broadband Platform for physics-based simulated shaking, the SCEC UCVM for site soil conditions, ObsPy for retrieving real recorded seismograms, and the GEM OpenQuake engine for ground motion prediction equations.
- Shaking animations can now be driven by physics-based simulated seismograms from the SCEC Broadband Platform, rather than a synthetic signal fitted to a single peak-acceleration value. The Technical page explains the method and its limitations.
- Decay-with-distance animations compute all three sites in a single simulation, so the panels show genuine attenuation with distance instead of differences between separate random ruptures.
- Animation narration now reads a city and its state as one phrase ("Redwood Valley California") instead of pausing awkwardly between them.
- Cloud GPU rendering is more reliable: when GPU capacity is briefly unavailable in one part of the cloud, the system now tries other locations automatically instead of falling back to a much slower render. Observed-waveform animations, which cannot be rendered without a GPU in a reasonable time, now wait for capacity rather than failing.
- Expanded the Technical page's rendering section with the measured speed difference between GPU and CPU rendering.
- Animation narration has a new, much more natural-sounding voice. Times, dates, magnitudes, and station names are now spoken clearly (for example, "magnitude six point four" on "December twentieth" at "ten thirty two in the morning").
- Updated the Technical page with descriptions of the new narration voice and of how animations are rendered on cloud GPUs.
- Shaking animations are now rendered on cloud GPU instances, cutting rendering time from many hours to minutes — animations for new earthquakes appear on the site much sooner after the event.
- Added a Privacy Policy page and this System Update Log so you can see what's changed on the site.
- Our contact email (info@intensitylab.com) is now shown directly on the site instead of only through the feedback form.
- Earthquake shaking animations are now automatically posted to our YouTube channel as they're created, organized into a playlist for each earthquake.
- Fixed an issue that could cause the chat assistant to time out when answering longer questions.
- Launched a new AI-powered chat assistant — ask plain-language questions about earthquakes and seismology and get instant answers.