Variable Resolution Operational Implementation for 2018 Field Season US Hydro 2019 Variable Resolution Operational Implementation for 2018 Field Season NOAA is constantly striving to create value for the end user and customer of our products. Create up-to-date, accurate products (Quality) Create optimal number of in-demand products (Quantity) Create products quickly (Efficiency) Create high quality, useful products quickly while reducing time sinks and user interaction that can lead to waste and blunders, respectively (Value) VR Grid data at highest, or user defined, resolution in one surface 3 Components Resolution map, comprised of tiles of different resolutions, Surface nodes, which are equally spaced within each resolution map tile, Surface mesh, which describes the relationships between nodes Input HDCS data Generate resolution map Nodes located Node value computed and populated (CUBE, uncertainty, etc) Resolution map and surface nodes stored as CSAR Surface mesh created “on-demand” 3 Types Depth Range: user defined table Super grid tiles split based on user-defined criteria Each tile assigned a resolution based on lookup table Calder Rice (Density): CARIS adaptation of CHRT’s density-based resolution estimator Each tile assigned a resolution based on target density CARIS (Density): CARIS density-based resolution estimator Not currently being used by NOAA; further testing John Doroba, NOAA Hydrographic Systems and Technology Branch March 2019
Achieved Efficiency Gains Reduce Surface Quantity Remove Inconsistent Hypotheses R1 R2 R3 R4 R5 Reduce Designated Soundings Remove Orphan Fliers Surface Quantity Each task performed on one surface Designated Soundings Highest resolution surface empirically supported by data density Inconsistent Hypotheses One surface, one hypothesis Orphan Fliers No depth gates clipping flier in another surface
Side Effects Adapt VR to QC Tools Holiday Finder Grid QA & Flier Finder Adapt SR tools for VR CSAR/BAG VR forced NOAA to automate = easier VR forced us to adapt tools to VR. Difficult to implement, Processing and analysis is easier
Anticipated Trade Off Computing/Add To VR is Inefficient on Daily Basis It takes longer to create a VR surface, but only one surface to create, review, recompute, and finalize. Use SR for daily processing and management. One VR surface makes life easier on the back end. Gains on back end (Complete surface generation/analysis)
Unexpected Hurdles VR Rendering was inhibitively slow VR rendering was so slow. It couldn’t be used for manual cleaning or analysis. Zoomed In (1x) – Zoomed Out (1x) – Panned (1/4 Map View to Right) 1:90,000 1:30,000 1:15,000 scale Summed each action at each scale for each test Tested -Core Configuration -Cache Size -Rendering Method/Level of Detail -NOS Baseline
NOAA VR Utilization NOAA utilized VR for 48% of processing pipeline Used more on back end (Deliverables and Survey Review), than front end (Night Processing and Sheet Management) Platforms in dynamic survey areas leveraged VR more than those in homogenous areas
VR Operational Implementation Front End Inefficiency (Processing/Management) Surface Rendering Too long! Surface Processing Too long! Use SR for daily surface Back End Efficiency (Deliverables/Survey Review) One Surface One surface analyzed No turning surfaces on/off No orphan flier or inconsistent hypotheses Surface Rendering -Panning and Zooming were slow Surface Processing Times -VR takes longer to compute than SR -Computing / Add To is inefficient short term One surface is easier to QC, analyze, and review
Change is Hard Ongoing challenges Efficient Cleaning/Review Let VR help you Let CUBE do it’s job Examine “other” layers Utilize QC Tools CUBE does what it does, trust it Use other layers like standard deviation or uncertainty to find anomalies Let QC tools QC the data VR makes this easier with one surface that is empirically supported by data density
Improvements Software Improvements Multiple threads process Load Tile Grid Separate thread Rendering Grid ~70% increase in rendering speed CARIS modified rendering method and we saw a huge increase. The 3 variables that have the most impact are: VR rendering operations process Rendering method / Level of detail Rendering method has greater impact than level of detail Cache size
Anticipate the Future VR utilization will increase with increased rendering speeds VR would be utilized more with decreased processing times Infrastructure is critical, but don’t neglect software VR will gain front end efficiency during 2019 season due to rendering modifications VR will be used even more if processing times improve