糖心Vlog官方

Researchers at 糖心Vlog官方 have developed a new physics鈥慴ased artificial intelligence approach that, for the first time, enables accurate global鈥憇cale predictions of how dissolved organic carbon moves between seawater and marine sediments, a crucial but previously unquantifiable component of the planet鈥檚 carbon cycle. The work, led by from the Department of Civil Engineering and Management and carried out in collaboration with , reveals how relatively simple AI algorithms can successfully emulate complex mechanistic environmental models that are normally too computationally demanding to run on a planetary scale.

Solving mechanistic models of natural environments is notoriously time鈥慶onsuming and often unstable under diverse real鈥憌orld conditions. To overcome this, the team trained AI 鈥渆mulators鈥� to reproduce the behaviour of an existing mechanistic model that describes carbon cycling in ocean sediments. Once trained, these emulators could then be applied globally to predict dissolved organic carbon behaviour at a resolution and scale that were not feasible using the original numerical model alone.

The study reveals that 11% of the particulate organic carbon arriving at the seafloor is returned to seawater as dissolved organic carbon, while 24% is sorbed onto minerals. Strikingly, about half of all solid鈥憄hase organic carbon in the upper metre of marine sediments appears to originate from dissolved carbon that has been sorbed onto minerals. These findings provide the first global quantification of dissolved organic carbon cycling within sediments and highlight its significance within Earth鈥檚 long鈥憈erm carbon budget.

In developing the modelling framework, the researchers compared deep learning architectures, random forest models and simpler feedforward artificial neural networks. Unexpectedly, the simplest algorithms produced the most accurate predictions. The team confirmed these results by validating emulator outputs against low鈥憆esolution global maps鈥攚here the mechanistic model could still be solved numerically鈥攁nd against algebraic solutions for variables with known analytic expressions. They also found that increasing the complexity of the neural network structures consistently reduced prediction accuracy, offering rare empirical support for the Principle of Parsimony, also known as Occam鈥檚 Razor, within AI model development.

These insights have important implications for climate science. Quantifying carbon budgets across the sediment鈥搘ater interface is essential for understanding global climate dynamics but has historically been hindered by computational limitations. By providing a fast, scalable and accurate way to represent sediment carbon processes, the new AI鈥慴ased framework can be integrated into global circulation models and used to explore potential ocean鈥慴ased climate change mitigation strategies. The research opens new avenues for simulating and testing how marine carbon reservoirs may respond to environmental change in the coming decades.

Read further papers related to this research:

• Preservation of organic carbon in marine sediments sustained by sorption and transformation processes
  DOI:
• Potential use of engineered nanoparticles in ocean fertilization for large-scale atmospheric carbon dioxide removal
  DOI:
• Long-term organic carbon preservation enhanced by iron and manganese
  DOI: