DICast® : A Dynamic Integrated ForeCast System NCAR/RAL 10/10/2014 The n  n  n  n  n  n  n  ® DICast System An automated point weather forecast system Provides timely, tuned, worldwide forecasts Designed to emulate the human forecast process Applicable to a variety of forecast problems Uses state-of-the-art scientific and engineering principles Requires only modest computing systems and common data sources Custom data sources can add skill Applications of n  n  n  n  n  ® DICast Lay forecasts for the public Road Weather Forecasts Agricultural Soil Forecasts Wind Turbine Forecasts Solar Power Forecasts DICast® Output and Operations n  Can produce a variety of tuned forecast variables •  Daily Max/Min Temp •  Probability of Precip •  Precip Amount and Type •  Temp & Dew point •  Wind u-, v-, speed Cloudiness Probability of Thunder Probability of Fog Visibility More… DICast® Output and Operations n  Can be set up in different temporal configurations •  Long Term •  Short Term •  Near Term Extent 0-16 days 0-4 days 0-24 hrs Resolution 3 or 6 hours 1 or 3 hours 1 hour Update Freq 3 hour 1 hour 1 hour DICast® Output and Operations n  Can produce tuned (with observations) or interpolated forecasts for thousands of locations 72565000;72565;KDEN;39.83;-104.66;1655;4;DENVER, DENVER INTERNATIONAL AIRPORT;CO;UNITED STATES 72494000;72494;KSFO;37.62;-122.36;3;4;SAN FRANCISCO, SAN FRANCISCO INTERNATIONAL AIRPORT;CA;UNITED STATES 72793000;72793;KSEA;47.44;-122.31;130;4;SEATTLE, SEATTLE-TACOMA INTERNATIONAL AIRPORT;WA;UNITED STATES 72503000;72503;KLGA;40.78;-73.88;6;4;NEW YORK, LA GUARDIA AIRPORT;NY;UNITED STATES 72530000;72530;KORD;41.98;-87.92;203;4;CHICAGO, CHICAGO-O'HARE INTERNATIONAL AIRPORT;IL;UNITED STATES 72202000;72202;KMIA;25.79;-80.32;3;4;MIAMI, MIAMI INTERNATIONAL AIRPORT;FL;UNITED STATES 72658000;72658;KMSP;44.88;-93.23;256;4;MINNEAPOLIS, MINNEAPOLIS-ST. PAUL INTERNATIONAL AIRPORT;MN;UNITED STATES 03772000;03772;EGLL;51.48;-0.45;24;6;LONDON / HEATHROW AIRPORT;XX;UNITED KINGDOM 07149000;07149;LFPO;48.73;2.40;89;6;PARIS-ORLY;XX;FRANCE 47662000;47662;RJTD;35.68;139.77;5;2;TOKYO;XX;JAPAN 94767000;94767;YSSY;-33.95;151.18;6;5;SYDNEY AIRPORT;XX;AUSTRALIA … Basic DICast® System Diagram Forecast Module A Integrator Forecast Module B Data Ingest Forecast Module C Forecast Module D . . . Forecast Module N Post Processing Forecast Products Dynamic MOS Forecast Module A Integrator Forecast Module B Data Ingest Forecast Module C Forecast Module D . . . Forecast Module N Post Processing Forecast Products Dynamic MOS •  Linear regression-based statistical method •  Similar to NWS MOS, but regressions built dynamically •  Can be applied to any NWP forecast model fairly easily •  Uses “default equations” if statistical model fails. 80 Good Regression 80 70 Max Temp Max Temp 70 Bad Regression 60 50 40 60 50 40 530 540 550 560 Thickness 570 580 530 540 550 560 Thickness 570 580 DMOS Default Equations •  Default equations are substituted whenever the statistical methods fail to produce a suitable result. •  Default equations are combinations of one or more of the regressors. •  Several regressors were designed specifically as defaults •  Example: Surface Temp: Model’s Vertical Temperature Profile Model’s Station Elevation Actual Station Elevation DICAST Estimated Surface Temperature Regression Extrapolation 80 Regression Max Temp 70 60 MaxT = 35 + .13 * CAPE 50 40 50 100 150 200 CAPE 250 Application 0-250 Max Temp CAPE Range CAPE = 2500 J 80 70 60 50 40 MaxT = 35 + .13*2500 0 500 1000 1500 CAPE 2000 2500 = 360 F Applied CAPE: 2500 Forecast Integrator Forecast Module A Integrator Forecast Module B Data Ingest Forecast Module C Forecast Module D . . . Forecast Module N Post Processing Forecast Products Forecast Integrator Objectives To combine forecasts from a set of models: •  Discovers the “best” combination of forecast modules for a given forecast time and location. •  Computationally •  Can simple and robust. easily adapt to the addition of new modules or removal of obsolete modules. DICast® Forecast Integrator •  Integrated forecasts (F) are bias-corrected, confidence-weighted sums of the module inputs (fi): F = ( Σ ci wi fi ) / ( Σ ci wi ) + Bias •  Confidences (ci) are determined by the forecast modules themselves. •  Weights (wi) are adjusted daily in the direction of steepest descent of the error (difference between verification, V, and the forecast) in weight space Δwi = S * (∂/∂wi) {(V - F)2} Forecast Integrator Forecast error as function of W1 & W2 1 W2 W2(i) Integration Step 0 0 W1 1 W1(i) Post Processing Forecast Module A Integrator Forecast Module B Data Ingest Forecast Module C Forecast Module D . . . Forecast Module N Post Processing Forecast Products Post Processing •  Quality Control •  Range Checks •  Minimal Inter-variable comparisons •  Temporal Interpolation •  Variable Derivation •  Forward Error Correction Deg F •  Spatial Interpolation 36 34 32 30 28 26 24 Raw Corrected 0 3 6 9 12 15 18 Lead Time (hours) Forecast Products Forecast Module A Integrator Forecast Module B Data Ingest Forecast Module C Forecast Module D . . . Forecast Module N Post Processing Forecast Products Forecast products n  n  n  Output Data Formats: u  netCDF u  ASCII - CSV Data can plug into other systems Decision Support ® DICast Advantages DICast® forecasts: •  Outperform every constituent forecast module •  Outperform human beyond 12 hours •  Based on sponsor feedback •  Are totally automated •  Are more cost effective than human generated forecasts DICast® is scalable •  Additional sites, NWP models or new forecast variables can be easily integrated Short Range Predictions" (0-96 hours)" Temperature Short Range Predictions" (0-96 hours)" Dew Point Temperature Short Range Predictions" (0-96 hours)" Wind Speed Medium Range Predictions" (0-10 days)" Temperature Medium Range Predictions" (0-10 days)" Dew Point Temperature