Events
Mon | Tue | Wed | Thu | Fri | Sat | Sun |
---|---|---|---|---|---|---|
1 | ||||||
2 | 3 | 4 | 5 | 6 | 7 | 8 |
9 | 10 | 11 | 12 | 13 | 14 | 15 |
16 | 17 | 18 | 19 | 20 | 21 | 22 |
23 | 24 | 25 | 26 | 27 | 28 | 29 |
30 | 31 |
Legend | |||||||||||
JSG | BEG | UTIG | EPS |
DeFord Lecture: P. David Polly
Start:December 5, 2019 at 3:45 pm
End:
December 5, 2019 at 5:00 pm
Location:
JGB 2.324
View Event
UTIG Seminar: Graduate Student Talks
Start:December 6, 2019 at 10:30 am
End:
December 6, 2019 at 12:00 pm
Location:
PRC ROC Room 1.603
Contact:
Constantino Panagopulos, costa@ig.utexas.edu, 512-574-7376
View Event
In the final UTIG seminar of 2019, three UTIG grad students will present a short overview of their research and recent findings.
Host: Steve Phillips
Speakers (click name for bio):
Janaki Vamaraju
Title: Mini-batch Least-Squares Reverse Time Migration In A Deep Learning Framework
Abstract: Migration techniques have been an integral part of seismic and electromagnetic imaging workflows. They can be used to move reflection points to their correct positions beneath the surface. Least-squares reverse time migration (LSRTM) overcomes shortcomings of conventional migration algorithms by compensating illumination and removing sampling artifacts to increase spatial resolution. However, the computational cost associated with iterative LSRTM is high and can cause slow convergence in complex media. We implement LSRTM in a deep learning framework and adopt strategies from the data science domain to accelerate convergence. As a use case, we solve the problem of pres-stack seismic LSRTM. The proposed hybrid framework leverages the existing physics-based models and machine learning optimizers to achieve more accurate and cheaper solutions. Using a time-domain formulation, we show that mini-batch gradients can reduce the computation cost by using a subset of total shots for each iteration. Mini-batches not only reduce source cross-talk but are also less memory intensive. Combining mini-batch gradients with deep learning optimizers and loss functions can improve the efficiency of LSRTM. Deep learning optimizers such as the adaptive moment estimation are well suited for noisy and sparse data. We compare different optimizers and demonstrate their efficacy in mitigating migration artifacts. Regularized loss functions such as the Huber loss is used in conjunction to accelerate the inversion. We apply these techniques to 2D Marmousi and SEG/EAGE salt models and show improvements over conventional LSRTM baselines. The proposed approach achieves higher spatial resolution in lesser time according to various qualitative and quantitative evaluation metrics.
Xian Wu
Title: Predictability of El Niño Duration Based on the Onset Timing
Abstract: El Niño causes episodic sea surface temperature warming of the equatorial Pacific Ocean and affects global weather patterns through atmospheric teleconnections. El Niño events typically last one year, but about one-third of El Niño events last for a second year, which could prolong and exacerbate their climate impacts. Analysis of observational and model data shows that El Niño events developing in boreal spring-summer usually terminate after peaking in winter, while those developing after summer tend to persist into the second year. To test the predictability of El Niño duration based on the onset timing, perfect model forecasts were conducted with the Community Earth System Model version 1 (CESM1), a model that reproduces the observed dependence of El Niño duration on the onset timing. We select three El Niño events developing in April or September from the CESM1 control simulation. For each event, 30-member ensemble forecasts are initialized with the same oceanic conditions in the onset month but with slightly different atmospheric conditions and integrated for two years. The CESM1 successfully predicts the termination of El Niño after the peak in 95% of the April-initialized forecasts and the continuation of El Niño into the second year in 83% of the September-initialized forecasts. The predictability of El Niño duration arises from the initial oceanic conditions that affect the timing and magnitude of negative feedback within the equatorial Pacific, as well as from the Indian and Atlantic Oceans. The forecast spread of El Niño duration, on the other hand, originates from surface wind variability over the western equatorial Pacific in spring following the peak. These results indicate potential predictability of El Niño events beyond the current operational El Niño forecasts by one year.
Kelly Olsen
Title: Strong sediment inputs at the south-central Chile margin revealed from fault patterns in 2D seismic reflection data and the relationship to megathrust earthquakes
Abstract: South-central Chile is the location of the largest (1960 Mw 9.5) and sixth-largest (2010 Mw 8.8) recorded earthquakes. 2D seismic reflection data collected in 2017 as part of CEVICHE (Crustal Experiment from Valdivia to Illapel to Characterize Huge Earthquake) show that in the region of these ruptures, thick (1.5-3 km) trench sediments are being subducting beneath the frontal wedge at the deformation front. The P-wave seismic velocity (Vp) of the incoming sediment is relatively high and thus the incoming sediment may be stronger than at other margins. We test this hypothesis by using structural analysis of conjugate fault pairs to determine the coefficient of internal friction for the incoming and slope sediments and compare these values to other subduction zones. Based on fault orientations, the average coefficient of internal friction for the incoming trench sediment is ~0.82 (conjugate normal faults dip ~62-70°), while the average frictional coefficient for the slope apron is ~0.55 (conjugate normal faults dip ~53-64°). The coefficient of internal friction for the slope sediment is consistent with values measured in the slope of other subduction zones, such as Costa Rica and Nankai. The coefficients for the incoming sediment sections showed much higher values than both the slope apron in Chile, and incoming sediment sections of other subduction zones, such as Nankai, Costa Rica, Cascadia, and Sunda. Along the Nankai drilling transect, the only samples where sediments have coefficients of friction comparable to those in the Chile trench were recovered from the off-scraped sediment at the toe of the accretionary wedge, which is likely more consolidated than trench sediment in Nankai. These results support the inference from the Vp data that the sediments entering the south-central Chile subduction zone via the trench are inherently stronger than in other subduction settings. The higher strength could allow them to develop higher basal friction and ultimately contribute to the strong interplate locking inferred along this entire segment of the Chile margin. This study highlights the importance of the physical properties of trench sediment for controlling overall development of the subduction interface and may explain why the south-central Chile margin generates such large earthquakes.
Alumni Reception during AGU in San Francisco
Start:December 11, 2019 at 12:00 pm
End:
December 11, 2019 at 2:00 pm
Location:
ThirstyBear Brewing Company, 661 Howard St, San Francisco, CA 94105
Contact:
Kristen Tucek, ktucek@jsg.utexas.edu, 512-471-2223