Annual Reprt fr Blue Waters Allcatin: Snia Lasher-Trapp, Oct 2016 Prject Infrmatin Simulating Cumulus Entrainment: A Reslutin Prblem, r Cnceptual? Snia Lasher-Trapp, UIUC slasher@illinis.edu Executive summary (150 wrds) Deep cumulus cluds prduce the majrity f the earth s precipitatin, but their strength, lngevity and ability t precipitate are affected greatly by entrainment. Entrainment is the prcess by which cluds intrduce dry air frm utside the clud inward, decreasing their buyancy and their water cntent (i.e., dilutin). Fr decades it has been argued that numerical simulatins underpredict the dilutin f cumulus cluds due t inadequate mdel reslutin limiting entrainment. Using 3D clud simulatins n Blue Waters at increasingly higher reslutin, and ur new entrainment and dilutin quantificatin algrithms, we have nt nly quantified the increase in simulated entrainment that was expected, but als have cnceptually clarified a limit at which further grid refinement is unprductive in diluting the clud cre. This limit appears near a grid spacing f 10 m, althugh simulatins f cumuli in ther envirnments are yet needed t establish the generality f this result. Descriptin f research activities and results Key Challenges: The riginal mdel we intended t use fr this prject was determined t be insufficient t run at the high reslutins with many prcessrs n the BW system. Thus, we had t mve t a cmpletely new mdel with a mre mdern architecture and better cmmunity supprt, CM1, which ther BW users have run successfully. In rder t quantify the entrainment and the resulting clud dilutin, cluds n the scale f several kilmeters must be simulated with grid spacing dwn t the 10 m scale. Sme f these runs required mre than 10,000 prcessrs. An issue related t the time step in the CM1 mdel was discvered that delayed prgress in prducing simulatins at the smaller grid spacings fr ver tw mnths, but was finally islated and crrected. (The issue has nw been dcumented with the BW team and the CM1 lead authr t assist ther users in the future.) These simulatins must utput the results frequently (every 4-6 secnds) in rder fr ur entrainment and dilutin algrithms (run utside f the CM1 mdel) t capture their evlutin accurately, requiring significant amunts f strage n lcal disks with direct access needed by these analysis prgrams. The algrithms had t be adapted frm the frmer mdel t run n the utput f the CM1 mdel. 1
As mdel reslutin increased, the quantificatin algrithms culd nt handle the full variable utput arrays frm the mdel, and thus they had t be adapted t perfrm upn smaller subsets f the mdel dmain. Why it Matters: Large-scale weather and climate mdels have inadequate spatial and tempral reslutin t mdel cumulus entrainment, and thus rely n parameterizatins that assume a particular underlying cnceptual mdel f entrainment. Our wrk is helping t determine what reslutin finer-scale mdels must use in rder t represent the prcess faithfully, s that their results can frm the basis fr imprved parameterizatins used in larger scale mdels. In additin, we have determined that dilutin, i.e., the decrease in clud buyancy and water inside the cre f the clud, cnverges faster than the entrainment values themselves, as the mdel reslutin is increased. This is gd news, as in general the dilutin effects are mre imprtant t the larger-scale mdels themselves than the exact amunt f entrainment. Why Blue Waters: Our Blue Waters allcatin was essential fr testing the reslutin-dependency f the entrainment prcess, and in particular fr determining the sizes f the eddies that were mst critical t represent in simulatins t cnverge upn a slutin f the clud dilutin. The Blue Waters supercmputer allwed us t push the spatial scale limit much farther than in the past, with its huge number f ndes, its high speed, and its large strage capability fr high-reslutin mdel utput and analysis. The hardware needed t run these kinds f simulatins quickly supersedes the limits f mst cmputers. Accmplishments: We ve run clud simulatins with grid spacing f 200, 100, 50, 30, 15, and 10 m t understand hw and why the values f entrainment d nt, but the results f entrainment (i.e. dilutin) eventually d, cnverge. This is an imprtant result with far-reaching implicatins acrss the atmspheric science cmmunity that is cncerned with cumulus parameterizatin at larger scales, as well as thse members f the cmmunity cncerned with the details f the amunt f liquid water inside the cluds available fr prducing precipitatin. This wrk was presented at the Internatinal Cnference n Cluds and Precipitatin in July 2016 and fueled significant enthusiasm and discussin. 2
100 m 50 m Reslving ~ 400-600 m eddies Reslving ~ 200-300 m eddies 20 m Reslving ~ 80-100 m eddies Figure 1: Snapshts f cluds simulated at different grid spacings (nted at tp) and the sizes f turbulent eddies that are well-reslved in each. Time Series f Entrainment Prfiles 100 m 50 m 20 m Entrainment int cre*, nrmalized by cre surface area-- units kg s-1 m-2 Figure 2: Vertical prfiles f entrainment (hrizntal mass fluxes) int the cre f the cluds, as a functin f time (different curves in each plt) and grid spacing used fr the simulatin (nted at tp f each plt). Entrainment values cntinue t increase as the simulatins used a grid spacing as small as 10 m (nt shwn). 3
100 m 50 m 20 m % f clud vlume % f clud vlume % f clud vlume Figure 3: Vertical prfiles f clud cre dilutin, i.e., the percent f the clud vlume at each height that cntains 80% r mre f its theretical maximum liquid water amunt in the absence f entrainment, versus time (different curves) and fr different simulatin grid spacing (nted at tp f each plt). Dilutin values eventually cnverged as the simulatins used a grid spacing as small as 10 m (nt shwn). List f publicatins and presentatins assciated with this wrk Presentatins: Simulating Cumulus Entrainment: A Reslutin Prblem, r Cnceptual? BW Users Cnference, Oregn, June 2016 (presented sme preliminary results) High-Reslutin Simulatins f Cumulus Entrainment. Internatinal Cnference n Cluds and Precipitatin, Manchester, England, July 2016 Publicatins: Lasher-Trapp, S., G., D. M. Mser, and D. Len, 2016: High-Reslutin Simulatins f Cumulus Entrainment and Dilutin, in prep. Other prducts: A new prpsal t the NSF t extend this wrk t the effects f entrainment n severe strm generatin and precipitatin is als in preparatin. These simulatins are als being used in Prf. Lasher-Trapp s curses n Cluds and Precipitatin. 4
Plan fr next year I am requesting a similar allcatin t that received last year: 250,000 nde-hurs n Blue Waters fr ur new prject: Untangling Entrainment Effects n Hail and Rain in Deep Cnvective Cluds Executive Summary: Entrainment describes hw cluds bring dry air frm utside the clud inward, by their wn mtins. Its effects can limit strm develpment, lngevity, and varius interdependent micrphysical prcesses that ultimately gvern the precipitatin prduced. Our understanding f the cnnectins between entrainment and precipitatin has been limited by inadequate mdel reslutin in past studies. Our previus wrk utilizing Blue Waters shwed that the resulting effects f entrainment didn t cnverge until the grid spacing apprached 10 m fr a cumulus clud. We will investigate if thse results can be scaled fr carser mdel reslutin used when simulating severe thunderstrms. New high-reslutin 3D simulatins f cnvective strms run n Blue Waters, alng with ur entrainment quantificatin algrithms, will be used t untangle the intricate web f cnnectins between entrainment and its effects upn excessive rainfall and hailfall, with the ultimate gal f imprving frecasts f these events. Scaling the runs cnducted n BW ver the past year t thse planned fr this prject (t be run using the same CM1 mdel, with slightly lwer reslutin but ver a much wider dmain, and with mre sphisticated micrphysics which means mre equatins t be marched frward in the mdel) the few, highest reslutin runs will require nearly 20,000 nde-hurs t run. But these highest reslutin runs will be limited in number, less than ten. The majrity f the runs will require n the rder f hundreds f nde-hurs t a few thusand. This allcatin will be split amng 3 graduate student prjects and my wn, t study the details f hw entrainment f dry air int the clud alters the micrphysical prcesses within it, ultimately affecting rain and/r hail at the grund. The default strage qutas shuld be sufficient. I will scale the anticipated usage schedule accrding t the timeframe in which sme students anticipate finishing their degrees at UIUC, and thers ramp up in ur research grup. Expressing this per quarter as a percent f the requested allcatin, we anticipate the fllwing usage: Q1: 10%, Q2: 25%, Q3: 35%, Q4: 30%. 5