in Cloudy Atmospheres Alexandra L. Jones Blue Waters Graduate - PowerPoint PPT Presentation

High Accuracy 3D Radiative Transfer in Cloudy Atmospheres Alexandra L. Jones Blue Waters Graduate Fellow Advisor: Larry Di Girolamo Department of Atmospheric Sciences, University of Illinois, Urbana-Champaign Blue Waters Symposium May 12, 2015 aljones4@Illinois.edu

Importance of Radiation 2 qld.gov.au MISR, 2007 Adapted from graphic by NOAA Why is radiation in a cloudy atmosphere important?

Current Representation of Radiation 3 W ×Ñ I ( r , W ) = - s e ( r ) I ( r , W ) + s a ( r ) B ( T ( r )) + s s ( r ) p ( r , W , W ) I ( r , ¢ W ) d ¢ ¢ W ò 4 p 4 p PP 3D ICA What’s the problem?

Goals 4  Produce Highly Accurate Benchmark Simulations  Quantify bias in our approximations  Improve simpler/faster parameterizations  3D Broadband Monte Carlo Community Model  Faster science progress  Starting point: I3RC What’s the problem?

Model Development: Spectral Integration 5 Spectral Integration Solar, thermal broadband Domain for I3RC each λ Solar Only β , ω , Ρ , x, Source y, z, A,T, λ Function How do we do better? The I3RC

Comparing to Analytical Solution 6 𝐺 ↓𝑡𝑔𝑑 −𝑐 𝑏∆𝑨 2 𝑓 −𝑏λ 𝑔 ∆𝑨 𝑏λ 𝑔 ∆𝑨 + 1 = − 𝑓 −𝑏λ 𝑗 ∆𝑨 𝑏λ 𝑗 ∆𝑨 + 1  TOA input flux linear function of wavelength  𝐺 ↓𝑈𝑃𝐵 λ = 𝑐 λ  Absorption coefficient linear function of wavelength  𝛾 𝑏 λ = 𝑏 λ How do we do better? The I3RC

Model Development: the Real Atmosphere 7 ARTS HITRAN Model S Spectral Integration CIRC T, P , z, Absorption VMRS Tables σ Solar, thermal broadband SSP I3RC Table Solar Only Creation Tool Source Function Domain SSP Table T, P , σ , ω , Ρ , x, y, z A, λ How do we do better? The I3RC

Checking Transmittance How do we do better? The I3RC

Real Atmosphere: Clear-sky Absorption 9 LBLRTM Flux I3RC Flux How do we do better? The I3RC

Reducing MPI-communication Time FORMER  Master assigns lesser of #perBatch or total remaining photons in bin to each worker  Workers trace those 5 2 photons. Then ask for 8 1 more work to do  Repeat until no photons 4 remaining How does it perform?

Reducing MPI-communication Time CURRENT  Master assigns #perBatch photons even if spanning multiple bins  Workers trace those photons. Then ask for more 50 50 work to do 50  Repeat until no photons 50 remaining 50 92% improvement in total compute time How does it perform?

Acknowledgements and Questions 12 This research is part of the Blue Waters  sustained-petascale computing project, which is supported by the National Science Foundation (awards OCI- 0725070, ACI-1238993 and ACI-14- 44747) and the state of Illinois. Blue Waters is a joint effort of the University of Illinois at Urbana-Champaign and its National Center for Supercomputing Applications. This work has also been funded by the  NASA Earth and Space Science graduate fellowship program. **List of references can be found on my poster** aljones4@illinois.edu