Comprehensive California-wide large-scale noise cross correlations

Julian Schmitt, Tim Clements, Lise Retailleau, Aurelien Mordret, Greg Beroza, Marine A. Denolle

Ambient noise cross-correlations have proven use in crustal tomography, structure, and source monitoring. We combine all available seismic data in the state of California since 2000 until 2020, which accumulates over 2000 asynchronous 3-component stations combining the Northern California Earthquake Data Center, the Southern California Seismic Network, and temporary arrays available at IRIS and nodal data from the BASIN experiment (Clayton et al, 2019). The total original data size is on the order of 200TB. We create three libraries of ambient noise cross-correlations. The first is a low-frequency, 1Hz sampling rate, accounting for all station pairs that are spaced by at least 20 km and cover the entire state. The second, intermediate frequency correlation functions of 20 Hz sampling rate, links all stations within 300 km. The third is all of the station autocorrelations sampled at 100 Hz or below depending on the instrument. Here, we frame a methodology to compute large scale cross correlations by leveraging cloud computing, GPU processing, and performant scripting in the Julia language, presented in a companion abstract (Clements et al.).

We keep daily-stacks of the correlation functions to perform monitoring of shallow structure, such as the impact of groundwater on seismic velocities, or deeper structure over the course of 20 years. The full stacks of the correlation functions can be used for crustal tomography, either with travel-time information or with the full waveforms. The full stacks are also used for ground motion prediction in Los Angeles for future earthquakes in another companion abstract (Denolle et al.).

Illuminating Snow Droughts: The Future of Snowpack in the Western United States

Julian Schmitt, Kai-Chih Tseng, Mimi Hughes, Nathaniel Johnson

Sustained snowpack in the Western United States (WUS) is crucial for meeting summer hydrological demands, reducing the intensity and frequency of wildfires, and supporting snow-tourism economies. By developing a nonparametric drought classification scheme for monthly snowpack, we can characterize regions by snow drought severity. Using the Seamless System for Prediction and EArth System Research Large Ensemble (SPEAR-MED LE, hereafter SPEAR)1 , we find that the incidence of severe drought (SD) increases ~25-50% from the early to late historical period (1930-1970 vs 1971-2011) across large WUS watershed regions. An observationally based dataset2 validates these findings. All watershed regions experience observed changes in SD occurrence that fall within the SPEAR ensemble spread, and 3 of the 5 regions experience observed increases in SD occurrence similar to the ensemble mean change. The second half of the historical period also experiences 60-70% increases in anomalously hot winter days, with nights also becoming significantly warmer. The climatology predicted by SPEAR under RCP 4.5 and 8.5 shows even more dramatic increases in SD occurrence. Under RCP 8.5, we find a quintupling in SD frequency by 2050, while RCP 4.5 sees the same increase by 2070. By examining the temperature and precipitation climatologies alongside the SD changes, we determine the increase in SD is driven by rapid warming after 2000, while average precipitation increases slightly. Temperature continues to increase dramatically after 2000, with the average anomalies surpassing 2 standard deviations above the mean by 2090 under RCP 8.5, and 1.5 standard deviations under RCP 4.5. We close by examining implications for the WUS’s environment and economy.