July 22, 2019 - by Sarah Waldrip
Fang-Li Qiao, Chair Professor of Physical Oceanography at the First Institute of Oceanography, China, has spent more than 30 years studying the complexities of the upper ocean and developing numerical ocean models based on those findings. Professor Eleni Chatzi, Chair of Structural Mechanics at the Institute of Structural Engineering at ETH Zurich, has similarly taken a hands-on approach to incorporating simulation tools into her work. Though their areas of expertise greatly differ and their institutions are literally oceans apart, it is their shared interest and experience with high-performance computing (HPC) that brought them before an international audience on the campus of ETH Zurich.
Before the dialogue could truly begin, Fang-Li took a few moments to share his research with the audience. He primarily focused on the recent revelation that non-breaking surface waves can generate strong turbulence in the upper ocean, which resulted in the development of new wave-tide-circulation coupled ocean models. “The wave-induced mixing in the upper ocean plays a dominant role. Unfortunately, the problem is that we did not consider this before in the ocean model,” said Fang-Li. Equipped with this new knowledge, he and fellow researchers are trying to produce more accurate predictive models that can be useful in the short-term for improving ship safety, typhoon predictions, and tourism protection; in the medium-term for disaster prevention and reduction, like floods and droughts, for the benefit of agriculture; and in the long-term as we monitor and prepare for the effects of climate change on global ecosystems and the vital fishing industry.
“For me, coming from the field of dynamics, and also for those from other domains, it’s very nice to see the similarities and the shared challenges in computation so very well-described,” said Eleni Chatzi at the conclusion of the short presentation. Beginning the dialogue in earnest, Chatzi went on to ask Fang-Li to elaborate further on what “state of the art” in numerical ocean modeling meant to him, to which he replied that coupling different processes into single models has been key to his latest research progress—specifically the ability to couple wave, tidal and circulation models together. However, “If we couple so many processes, the computing cost will be much higher. We require more and more resources, computers that are more and more powerful,” Fang-Li said. Integrating the newly quantified processes into the model comes at a cost.
Chatzi continued, “I noticed in your recent paper from 2019, you mentioned as well that numerical and parameterization improvements will continue to define the state of the art in numerical ocean modeling—in concert with integration and core analysis of observations. In this respect, what is your view regarding the developments in the near future?”
Fang-Li’s answer was one that many researchers who use HPC across many fields can relate to: models will continue to improve as physical processes are better understood and quantified; higher resolution models supported by powerful HPC systems will be sought; and new technologies, such as data assimilation and artificial intelligence, will be implemented as they become viable—which, in the case of ocean modeling, can mean incorporating data to simulate more accurate initial conditions. “With these new technologies, we can identify the difference between models and observations very quickly.”
“Models do come with errors. This is true across fields,” Chatzi added. “Even during your presentation, you discussed biases like the erroneous temperature model in the North Atlantic leading to propagations of errors in the resulting models. … How can we increase our confidence in what the model predictions are?”
“Confidence is about reducing uncertainty,” replied Fang-Li. He explained that this progress depends on acquiring the tools needed to quantify those uncertainties. When the tiniest changes in initial conditions can produce totally different climate outcomes, this is a daunting task. “One solution is to make an ensemble, which means to run the model many times and take an average. We identify some runs that are out of scope and should not be considered, but then take an average and call that our result. Averaging model results helps, but that is of course very high cost.”
Chatzi responded, “This is a good point. We should maybe not be presenting these results if they come with uncertainties as deterministic, but rather in their full probabilistic form.”
On the topic of model validation, Fang-Li concluded, “We as modelers always try to be honest, but our models are not always correct. Model and reality are simply different. So we need to compare the data from different sensors with the model results and try to identify the bias. By identifying bias, we improve our physical understanding and we improve the model.”
Chatzi continued, “We do something similar—we call it model updating. I agree that model validation is very important.” With that thought in mind, Chatzi transitioned to the final question regarding the real-world application of these ocean models. Specifically, researchers like Fang-Li have been using these models to tackle extreme events, such as the nuclear radiation spread following the Fukushima catastrophe in 2011, and also the rescue of lives at sea after the capsizing of passenger boats. “Can you elaborate on these extreme events and the recovery?”
Fang-Li recalled that in the days following the Fukushima disaster, the radiation forecasts he saw were only able to predict a few days at a time, “but what about one month? What about in one year?” he asked. So he returned to his research team with the idea to use the ocean models to predict the spread of radiation much further into the future. What they found was that after one month, the whole northern hemisphere would be affected by the radiation. After 18 months, they had models of the surface down to 800 meters. Later, they compared real data to the model results, indicating that the coupled model had performed very well. They even developed a model to show surface concentration in the Pacific Ocean over the course of 30 years, up to the year 2041.
A more recent example of the ocean model’s real-world application was a response to two cruise ships capsizing on July 5, 2018, near Phuket, Thailand. In an effort to support search and rescue efforts, Fang-Li and his team quickly developed models to predict where possible survivors might be drifting over the course of a three-day period. The predictions proved to be highly accurate, as all the located survivors were found in the areas that the model had indicated.
“I find two very important points here,” Chatzi responded. “One is that fact these applications really practically assist governments and recovery, and also that one has to operate with different means to do short-term forecasting, something very fast, and then long-term and how this evolves on a greater scale.” She went on to describe Fang-Li’s work regarding less extreme, but nonetheless impactful events such as the algae blooms affecting global water systems and coastal cities. Again, the models were able to offer reliable predictions for algae flows near particular cities along the coast of China, which were of great interest to local governments concerned about the possible impacts.
After the conclusion of the Interdisciplinary Dialogue session, PASC19 attendees also had the opportunity to learn more during a mini-symposium entitled “Development of the Global High-Resolution Wave-Tide-Circulation Coupled Ocean Model and Its Operational Application”, which was led by one of Fang-Li’s colleagues, Bin Xiao, also from the First Institute of Oceanography, China.