Main Page

From geohpgc
Jump to navigation Jump to search

Welcome Message

Hello! This is the homepage of the High Performance Geo-Computing (HPGC) research group in Tsinghua University. Here, we accumulate all kinds of information that may interest you, and the members of the group, :).

The name describes the major focus of the group, which is to motivate the interaction between HPC and geoscience. The goal is to bridge the gap between the grand challenges in earth science, and the emerging computing methods and hardware platforms.

In one way, better supercomputing technologies would lead to better solutions or even scientific breakthroughs; and in the other way, science improvements would also guide the development of future supercomputing technologies. Moreover, the interaction between these two, I believe, would lead to successful synergy of methods, data sets, perspectives, and ideas.

We currently have 1 professor, 2 postdocs, and 9 PhD students.


Group Members

We are Hiring!!!

We are constantly looking for new talents to join the group as students or postdocs. Especially, we are currently looking for 2-3 new postdocs to join the group.

2 postdocs would be working in our recently-started Tsinghua University (Department of Earth System Science)- Xi'an Institute of Surveying and Mapping Joint Research Center for New Smart Surveying and Mapping.

1 postdoc would be working in the domain of AI-based solutions for climate modeling.

Postdocs in the group would receive the standard benefits provided by Tsinghua University, as well as a competitive additional package from the group.

Contact: haohuan@tsinghua.edu.cn, please send your CV, research statement, and 3 major publications.

Recent Research Highlights

AI for Remote Sensing

Exploring the AI-based methods to analyze and understand the remote sensing images has been one of our group's major focus. In China, we say, "take the history as a mirror, we see the rise and the fall (以史为镜,可以知兴替)". The satellite images serve as a perfect mirror of the Earth, covering a long history (starting from the 1960s) and providing a rich set of information. Combining the richness of the data and the intelligence of the method, we expect to identify more knowledge about the changing patterns and rules, derive a more clear understanding of the past, and to make a more accurate prediction into the future.

Our efforts include: (1) transferring existing land cover maps to a higher resolution (e.g.: 10m to 3m) through learning with noisy labels, thus minimizing the manual efforts involved; (2) generalized deep learning networks to detect specific objects (e.g.: buildings, roads, trees, etc.) in remote sensing images; (3) super-resolution and fusion methods to integrate historical remote sensing data sets with different spatial resolutions and spectrums. See more in [Remote Sensing]

Example results for improving the land cover map from 10m to 3m resolution.
Example results for improving land cover maps from 10m to 3m.



Building-extraction.png
Building extraction results of satellite images.



High-resolution.png
Super-resolution results.



Extreme-Scale Simulation on Supercomputers

Our group were fortunate to be a part of the Tsinghua team that operate and manage the Sunway TaihuLight Supercomputer in Wuxi. Starting from the middle of 2015, many of the students in the group have had the opportunities to explore their research ideas on the fastest computer in the world at that time.

Our work has enabled or supported a number of important scientific applications to run at a scale of millions of or tens of millions of cores. The quantum circuit simulation (2021, 40 million cores with a sustained performance of 4.4 Eflops), non-linear earthquake simulation (2017, 10 million cores with a sustained performance of 18.9 Pflops), and atmospheric dynamic solver (2016, 10 million cores with a sustained performance of 7.95 Pflops) won the Gordon Bell Prizes of those years.


Simulation of Tangshan Earthquake, 2017, Gordon Bell Prize






HPC+AI for Geophysics

HPC solutions for geophysics exploration used to be the main research focus of the group in the early days. We have collaborated with a number of oil and gas companies (Statoil and Saudi Aramco) and service companies (BGP and Schlumberger) to explore the potential of heterogeneous computing architectures in migration and imaging applications.

The recent trend jumps to AI-based methods to process seismic data, such as analysis of seismic ambient noise, separation of P- and S-wave, and PSF resolution enhancement. See more in [Geophysics]

Crustal velocity variations can be estimated by analyzing seismic ambient noise. We obtained the crustal velocity changes in Sichuan, China, using the seismic ambient noise data in 2013. The changes in Sichuan present an obvious correlation with the surface displacement observed by GNSS. We demonstrate the consistent variation in crustal and surface with different observations for the first time.