Oen Preditive Skill Assessments in the South Atlnti: Crowd-Souring of Student-Bsed Disovery *United Sttes Nvl Ademy Annpolis, MD, USA Rhel Stelli, Tois Shofield, Ktherine Todoroff, Angel Crndng*, Alyson Eng*, In Lowry*, Hrrison Mther*, Antonio Rmos**, Sestin Swrt***, Mrelo Dottori****, Nilsen Strndskov, Josh Kohut, Osr Shofield nd Sott Glenn Rutgers University Costl Oen Oservtion L New Brunswik, NJ, USA **Universidd de Ls Plms de Grn Cnri Ls Plms, Grn Cnri ***Southern Oen Cron & Climte Oservtory, Counil for Sientifi nd Industril Reserh, Stellenosh, South Afri nd Deprtment of Oenogrphy, Mrine Reserh Institute, University of Cpe Town, Rondeosh, South Afri ****University of São Pulo, Oenogrphi Institute São Pulo, Brzil Astrt Autonomous Underwter Gliders hve over dede long history of suessful regionl deployments serving sientifi, soietl nd seurity needs in pplition res rnging from pole to pole nd inluding the full rnge of wter depths from shllow ostl ses to the deep oen. Glider deployments overing the sin sle re muh fewer, ut re growing pity s demonstrted y the Woods Hole to Bermud line tht rosses the Gulf Strem, the Atlnti Crossing line tht follows the Gulf Strem, nd the sin irling flights now eing onduted s prt of the Chllenger Glider Mission. The next step in the evolution of the glol Chllenger mission is to enle n ensemle of modelers from different institutions nd genies to prtiipte in meningful wy. This proess with e formlized in 2014 y leverging the dt mngement tools of the U.S. Integrted Oen Oserving System (IOOS) nd the edution tools of the U.S. Ntionl Siene Foundtion s (NSF) Oen Oserving Inititive (OOI). The Edution Visuliztion (EV) tools developed y the OOI s Edution nd Puli Enggement (EPE) Implementing Orgniztion (IO) re urrently eing onfigured through the yer OOI net to disply rel time OOI glider dt with intuitive intertive rowser-sed tools, reduing the rriers for student prtiiption in se explortion nd disovery. Through U.S. IOOS, forest oen dt will e hrvested from the ephemerl oen snpshots produed y n ensemle of oen models long the sme glider trks s Chllenger. The prllel oserved nd forest dtsets, oth evolving in rel time, will e essile through the sme OOI EV tools, enling student prtiiption in rowd-soured oen preditive skill experiment. The result will stisfy one of the importnt gols of the Chllenger mission y enling students to ssess of the qulity of the ensemle of ville glol sle oen models. Student reserh tem projets tht use the new model dt omprison pilities will e onduted during the summer of 2014. Students will ompre n ensemle of the glol oen models long the high veloity trnsport pthwys y gliders on sin-sle missions, suh s one tht trverses the northern side of the South Atlnti gyre long the Brzilin shelfrek. The lsting impt of the Chllenger mission will e glol fleet ville to respond to events, n ssessment of the oen models long the fstest oen trnsport pthwys, nd the estlishment of network of gliderports for glol response. Keywords Oen Foresting; Autonomous Underwter Gliders; Chllenger Glider Mission. I. INTRODUCTION Autonomous Underwter Gliders hve een utilized y the sientifi nd defense ommunities for multitude of different
missions tht over ll res of the gloe from the surfe of the oen to 1000 meters elow. Reently, long sle endurne missions hve strted to eome fous. The Chllenger Glider Mission is one suh mission whose im is to use gliders to explore five oen sins overing 128,000 kilometers. Amerin RTOFS oen model nd the Europen MyOen oen model nd expnd upon the work done lst yer in the Doson et l. [5] rtile. This flight ompletes the exmintion of the northern setion of the South Atlnti gyre. While the oen forest models urrently ssimilte dt from n Argo network of over 3000 drifters, ssimilting glider dt tht rosses frontl fetures my e enefiil to inresing the forest ury. Glider dt n help to inrese smpling resolution in res not overed extensively y Argo drifters. Figure 1: The projeted pths of the Chllenger Glider Mission. Two of the gliders tht re involved in this Chllenger Glider Mission re Silo nd RU29. Silo hs not ontriuted signifintly to the mission sine the Doson et l. [5] rtile where dt from Silo nd RU29 were used to ompre ginst the then urrent oen models. As suh, n extension on the nlysis of the Silo dt is not needed. RU29, however, hs stopped in the Asension Islnds nd flown ll the wy to São Pulo, Brzil sine the pulition of the Doson et l. [5] rtile. Figure 3: A photogrph of RU29, n utonomous underwter glider tht suessfully rossed the Atlnti Oen in 2014.Behind RU29, the Oenogrphi Reserh Vessle Alph Delphini, from the University of São Pulo This study will report the results of student investigtions ompring temperture, slinity, nd depth-verged urrents etween RU29 nd the model forests. Preliminry student oservtions long RU29 s flight indite tht the MyOen model, while different from RU29 on wider re, ws loser to RU29 t points of disrepny. The RTOFS model, however, hd fewer res of disrepny, ut the disrepnies tht did exist were generlly of greter mgnitude. II. METHODS The two oen foresting models used in this pper re the MyOen model nd the RTOFS model. The MyOen model is the result of ollortion etween Europen ountries suh s the United Kingdom, Frne, Germny, nd Denmrk. The RTOFS model is reted y the Ntionl Center for Environmentl Predition, sugroup of the Ntionl Oeni nd Atmospheri Administrtion sed in the United Sttes. Figure 2: A mp of the history of trks overed y Rutgers Costl Oen Oservtion L s Chllenger gliders. Bsin sle missions, in ollortion with Teledyne We Reserh nd Universidd de Ls Plms de Grn Cnri. The dt olleted from the flight of RU29 from the Asension Islnds to São Pulo, Brzil n ontriute signifintly to our understnding of the ury of oth the Figure 4: Exmple of the pth plnning tools tht n e mde using dt from the oen foresting models RTOFS (left) nd MyOen (right) y Universidd de Ls Plms de Gr Cnri.
The sensor used y the G2 tegory of gliders is SeBird pumped Condutivity, Temperture nd Depth (CTD) sensor. Temperture nd Slinity dt is reovered from the glider nd therml inerti from the ondutivity sensor is orreted for using the proess from Kerfoot et l. [3]. This dt is ompred to setion of the RTOFS model nd the MyOen model simultions long the pth tken y the glider (Figure 5). Figure 6: The differene in temperture etween the RTOFS model with RU29 () nd the MyOen model with RU29 (). Figure 5: The portion of RU29 s trk tht ws used for omprison to the models, fousing on the green, white, nd red setions. This trk represents n est to west pth long the northern side of the South Atlnti Gyre. The omprisons mde in this pper re result of n nlysis of temperture, slinity, nd depth-verged urrents mde y the models nd RU29. MyOen nd RTOFS dt ws olleted from respetive Internet soures, while the glider dt ws olleted y RU29 (Figure 3). A series of MATLAB sripts filitted dt proessing nd figure retion. By ompring the in-situ glider dt to dt otined y the Argo Flots (Figures 7 & 9), we were le to onfirm the qulity of the glider temperture nd slinity oservtions of the onditions of the wter olumn. The most striking oservtion is of the temperture profiles, where the deep dt from RU29, Argo, nd RTOFS re similr, ut MyOen deep tempertures re out 1 C higher. The temperture nd slinity dt olleted y RU29 ws then ompred to RTOFS nd MyOen (Figures 6 & 8). The figures were reted y oserving the differene etween glider dt nd model predited dt. There ws generl 1 C differene etween the MyOen nd RU29 temperture dt (Figure 6), wheres the disrepnies etween the RTOFS nd RU29 dt were less widespred ut more signifint (Figure 6). The most vrition etween the two models nd RU29 exist within the first 300 meters of the wter olumn, s shown in figures 6 nd 8. Hene, the reminder of this pper will fous on nlyses of depths ove 300 meters. Figure 7: Temperture depth profile omprison etween RU29, RTOFS, MyOen, nd Argo Flot dt t smple point. Figure 8: The differene in slinity etween the RTOFS model with RU29 () nd the MyOen model with RU29 ().
Figure 9: Slinity depth profile omprison etween RU29, RTOFS, MyOen, nd Argo Flot dt t smple point. III. RESULTS RU29 s mission from Asension Islnd to São Pulo, Brzil lsted from Novemer 15, 2013 to My 18, 2014. It trveled 4,420 km rossing the Atlnti on the northern prt of the South Atlnti gyre. An nlysis of temperture, slinity nd urrents ws onduted on three res of interest: 1,262 km leg off the ost of Asension Islnd, 265km leg off the ost of Brzil, nd 179km leg on the Brzilin ontinentl shelf (Figure 5). A. Turtle Trks The first re of interest ourred from Novemer 15, 2013 to Deemer 30, 2013. This setion will e referred to s Turtle Trks nd is denoted y the green setion of the trk in Figure 5. It is worth noting euse the migrtion pth of Green Se Turtles etween Asension Islnd nd Brzil runs long this pth. An nlysis of temperture nd slinity ws onduted. There ws not muh disprity in temperture etween the two models nd RU29 (Figure 10). However differene etween RU29 nd MyOen exists t the surfe nd etween 250 nd 300 meters (Figures 10 &10). The slinities of RTOFS nd RU29 hve lrger differenes thn the slinities of RU29 nd MyOen. Most notiele is the onsistent susurfe slinity pek ner 100 m visile in RU29 nd MyOen, ut missing from RTOFS. Neither model urtely depits the depth verged urrent vetors mesured y RU29 long this re of the trk. Figure 11: Temperture omprison etween RTOFS (), RU29 (), nd MyOen () for the Turtle Trks. Figure 12: Current Vetor omprison etween RTOFS (), RU29 (), nd MyOen () for the Turtle Trks. B. Deep Eddy The seond re of interest ourred from April 26, 2014 to My 6, 2014. This setion will e referred to s the Deep Eddy nd is denoted y the white setion of the trk in Figure 5. It is worth noting euse of the strong presene of n eddy. This is n re of strong meso-sle tivity due to the Brzil Current[6] nd the dt olleted with the glider n provide vertil nd horizontl in-situ detils tht re not possile with other methods There is evidene of deep-wter lokwise spinning eddy in the urrent plot of RU29 (Figure 15), whih is lso seen in its temperture nd slinity plots (Figure 13 & 14). This eddy ppers in the RTOFS urrent plot s well, with the sme lokwise diretion (Figure 15). Although this eddy is present in the sme spot s RU29 in oth RTOFS temperture nd slinity plots, it mnifests less grdully thn the eddy shown in RU29 s temperture nd slinity plots (Figures 13, 13, 14 & 14). MyOen does not reognize the eddy struture t ll (Figures 13, 14, & 15). Figure 10: Temperture omprison etween RTOFS (), RU29 (), nd MyOen () for the Turtle Trks.
Figure 13: Temperture omprison etween RTOFS (), RU29 (), nd MyOen () for the Deep Eddy. of the trk in Figure 5. It is worth noting euse glol models often hve diffiulty resolving fetures in shllow wters nd the Co Frio Eddy is well known eddy. There is evidene of shllow-wter ounter-lokwise spinning eddy in the urrent plot of RU29 (Figure 18), whih is lso seen in its temperture nd slinity plots (Figure 16 & 17). This eddy ppers in the MyOen urrent plot s well, with the sme ounter-lokwise diretion (Figure 18). The eddy is present in the sme spot s RU29 in the temperture nd urrents plots ut the MyOen model elongtes the eddy (Figure 16, 16, 18, & 18). The MyOen slinity plot lso shows the eddy elongted, however, the eddy is shifted towrds the ost. (Figure 17 &17). RTOFS does not reognize the eddy struture t ll (Figures 13, 14, & 15). On the ontinentl shelf, possily euse MyOen elongtes the Co Frio eddy, the temperture nd slinity it reports for the ontinentl shelf re too wrm nd slty (Figures 16, 16, 17, & 17). While MyOen is inorret, it is not rdilly different from RU29, espeilly losest to the ontinent. RTOFS, however, reports notiely wrmer nd sltier wter in the very shllow wters (Figures 16 & 17). Figure 14: Slinity omprison etween RTOFS (), RU29 (), nd MyOen () for the Deep Eddy. Figure 16: Temperture omprison etween RTOFS (), RU29 (), nd MyOen () for the Continentl Shelf nd Co Frio Eddy. Figure 15: Current Vetor omprison etween RTOFS (), RU29 (), nd MyOen () for the Deep Eddy. C. Continentl Shelf nd Co Frio Eddy The third re of interest ourred from My 7, 2014 to My 18, 2014. This setion will e referred to s Continentl Shelf nd Co Frio Eddy nd is denoted y the red setion Figure 17: Slinity omprison etween RTOFS (), RU29 (), nd MyOen () for the Continentl Shelf nd Co Frio Eddy.
Figure 18: Current Vetor omprison etween RTOFS (), RU29 (), nd MyOen () for the Continentl Shelf nd Co Frio Eddy. IV. DISCUSSION Two foresting models, MyOen nd RTOFS hve een studied in omprison to the glider RU29. The findings presented hve led to the onlusion tht these two models hve inonsistenies with in-situ glider dt. Neither model n e onsidered superior s in ertin res, suh s the Deep Eddy, the MyOen model ws lking fetures in the oen s struture wheres the RTOFS model did not predit the existene of the Co Frio Eddy. Both the RTOFS model nd the MyOen model ontin some importnt fetures tht orrespond with the glider dt yet oth lso do not lwys reflet the struture pprent in the in-situ glider dt. Over the length of the RU29 Asension Islnd to Brzil flight, the MyOen model hs firly onsistent 1 C to 2 C differene in temperture from RU29, however there were not mny ples in either slinity or temperture tht the MyOen model differed notly from RU29. In omprison, the RTOFS model ws firly onsistent with the RU29 dt t lower depths ut hd very lrge disrepnies in mgnitude t ertin lotions. Oen forest models hve eome n indisputle tool for sientists nd students, with the ility to resolve eddy strutures nd inorporte new dt. They re esily essile to reserhers round the world. Suh improvements to the models ome from dt tken from stellite nd Argo flots tht mesure se-surfe temperture, se surfe height, nd temperture/slinity profiles. In order to further improve the models, more dt must e integrted from other soures. This will help dvne forest modeling in prediting fetures suh s hurrine intensity. Autonomous Underwter Gliders re le to ontriute to the oen forest modeling euse of their ility to trverse previously under-smpled regions of the oen. It is ruil to the future of oen modeling tht glider-olleted dt e inorported espeilly where res of disgreement etween the models our. V. ACKNOWLEDGEMENTS The Chllenger Glider Mission is growing glol prtnership of glider opertors, sientists, edutors nd students oordinted y Rutgers University nd endorsed y the Chllenger Soiety. The results presented in this pper were initited in Rutgers reserh ourse nd ompleted y three undergrdute summer interns (Rhel Stelli, Tois Shofield, nd Ktherine Todoroff) supported y Teledyne We Reserh nd four U.S. Nvl Ademy Interns (Angel Crndng, Alyson Eng, In Lowry, nd Hrrison Mther). Glider pth plnning nd model/dt omprison tivities re led y Nilsen Strndskov (Rutgers) nd Antonio Rmos (Universidd de Ls Plm de Grn Cnri) with yer infrstruture support provided y John Kerfoot (Rutgers). Glider opertors inlude Chip Hldemn, Dve Argon, Tin Hskin (Rutgers) nd Ben Alsup (TWR). The RU29 deployment in South Afri ws enled y the University of Cpe Town, The Nnsen- Tutu Center, nd the South Afrin Mrine Engineering & Rootis Centre t the Counil for Sientifi nd Industril Reserh (CSIR). The reovery nd redeployment on Asension Islnd ws filitted y the British Royl Nvy, the U.S. Nvy, nd Asension Islnd Fishing Chrters (AIFC). The reovery nd redeployment of RU29 in São Pulo, Brzil ws enled y University of São Pulo. RU29 mission support is provided y the U.S. Integrted Oen Oserving System, the U.S. Offie of Nvl Reserh Glol, nd the G. Unger Vetlessen Foundtion. Tehnology support provided y Teledyne We Reserh, Iridium nd CLS Ameri. Argo dt ws olleted nd mde freely ville y the Interntionl Argo Progrm nd the ntionl progrms tht ontriute to it. (http://www.rgo.usd.edu, http://rgo.jommops.org). The Argo Progrm is prt of the Glol Oen Oserving System. VI. REFERENCES [1] Glenn, S., + 47 o-authors, Proess-driven improvement to hurrine intensity nd storm surge forests in the Mid- Atlnti Bight: Lessons lerned from Hurrines Irene nd Sndy. Oens 13 MTS/IEEE Bergen. [2] Glenn, S., Kohut, J., MDonnell, J., Seidel, D., Argon, D., Hskins, T., Hndel, E., Hldemn, C., Heifetz, I., Kerfoot, J., Lemus, E., Litenwlder, S., Ojnen, L., Rorty, H., Atlnti Crossing Students, Jones, C., We, D., Shofield, O. 2011. The Trns-Atlnti Sloum glider expeditions: A tlyst for undergrdute prtiiption in oen siene nd tehnology. Mrine Tehnology Soiety 45: 75-90. [3] Kerfoot, J., S. Glenn, H. Arngo, C. Hldemn, C. Jones nd D. Pinkel, 2010. Corretion nd omprison of pumped nd non-pumped CTS sensors on the Sloum Glider, EOS Trnstions, AGU, 91(26), Oen Si. Meet. Suppl., Astrt IT35B-03. [4] Shofield, O., Kohut, J., Glenn, S., Morell, J., Cpell, J., Corredor, J., Orutt, J., Arrott, M., Krueger, I., Meisinger, M., Peh, C., Vernon, F., Chve, A., Cho, Y., Chien, S., Thompson, D., Brown, W., Oliver, M., Boiourt, W. 2010. A regionl Sloum glider network in the Mid-Atlnti ostl wters leverges rod ommunity enggement. Mrine Tehnology Soiety 44(6): 64-74. [5] Doson, C., Mrt, J., Strndskov, N., Kohut, J., Shofield, O., Glenn, S., Jones, C., We, D., Brrer, C., Rmos, A. 2013. The hllenger glider mission: glol oen preditive skill experiment. Oens 13 MTS/IEEE Sn Diego: 1-8. [6] Cldo, L., Gngopdhyy A., d Silveir I. C. A. 2006. A prmetri model for the Brzil urrent menders nd eddies off southestern Brzil, Geophysil reserh letters, 33(12)