170 likes | 269 Views
Verification of NAEFS Land-Surface Forecasts in Warm Season JAS 2006. Wanru Wu, Kingtse Mo and Muthuvel Chelliah Climate Prediction Center/NCEP/NOAA CDPW 2007, Tallahassee Florida. Motivation & Objective.
E N D
Verification of NAEFS Land-Surface Forecasts in Warm Season JAS 2006 Wanru Wu, Kingtse Mo and Muthuvel Chelliah Climate Prediction Center/NCEP/NOAA CDPW 2007, Tallahassee Florida
Motivation & Objective Provide experimental objective drought monitoring and outlook over the United States and Mexico in support of the National Integrated Drought Information System (NIDIS) Explore the possibility to apply the North American Ensemble Forecast System (NAEFS) for short-range drought outlook operationally
Data NAEFS: an ensemble forecast system that combines state-of-the-art weather forecast tools developed at the U.S. National Weather Service and at the Meteorological Service of Canada to provide numerical weather prediction products in both countries for a forecast period of 1-14 days (see details at http://www.emc.noaa.gov/gmb/ens/NAEFS.html). Data: archived since 23 May 2006. Means and spreads for week1 (1-7 days) and week2 (8-14 days) forecasts were calculated from 60 ensemble members, with 1x1 degree spatial resolutions and 6-hour temporal intervals. Forecast Verification: focused on the warm season JAS 2006. Data from the North American Regional Reanalysis (NARR) were used for the verification. Note that the NAEFS DO NOT initialize the land-surface conditions.
Mean Errors Averaged over JAS 2006 Errors are large ( > 1 std), especially for SM Error patterns are similar for Week1 & week2 forecasts • Errors are consistent: • Positive P error positive SM error positive E error • Cooler Ts is also consistent with more P
Fcst (corrected) = Fcst - X1(Fcst) + X2(RR) where X1(Fcst) = ave(Fcst, t=t-T-t0, t=t-t0) X2(RR) = ave(RR, t=t-T-t0, t=t-t0) T = the training period t0 = 7 & 14 days for week1 & week2 forecast correction, respectively RMSE averaged over 70o-125oW 20o-50oN RMSE vs. Training Period RMSE is not sensitive to T > ~ 2 weeks
Anomaly Correlation over 70o-125oW 20o-50oN Black– no correction Red– after correction Week1 Week2 SM and E errors are mostly systematic T2m and P are less so
Errors Averaged over 31o-35oN for Week1 Forecasts P: complicated SM & E: mostly corrected T2m: largely corrected
Errors Averaged over 31o-35oN for Week2 Forecasts Similar as week1
P T2m Anomalies Averaged in JAS with Respect to RR Climatology Forecasts Verification Patterns improved after the correction Error-Corrected Forecasts
Errors in the Initial Conditions Averaged for July 2006 Forecast Error IC Error The NAEFS do not initialize the land-surface conditions. Errors are mostly inherited from the initial conditions.
(a) SM IC - RR (b) T2m IC - RR Errors in IC of May 23, 2006
(a) Ensemble Anomaly (b) Spread The Verification Reliability RR, Noah, Mosaic & VIC 4-Model Ensemble Anomaly & Spread JAS 2006 NAEFS Forecast Mean Errors JAS 2006 Errors >> Model Spreads
Made on 09Oct2007 Made on 02Oct2007 NAEFS Forecasts 00Z10Oct2007– 00Z17Oct2007 http://www.cpc.ncep.noaa.gov/products/Drought Experimental Operational Forecasts – Precip (mm) CPC real time precipitation analysis 7-day accumulation 12Z10Oct2007–12Z17Oct2007
http://www.cpc.ncep.noaa.gov/products/Drought Experimental Operational Forecasts – Temp (oC) Made on 09Oct2007 Made on 02Oct2007 NAEFS Forecasts 10 Oct 2007–16 Oct 2007 CPC 7-day temperature analyses 7-day mean anomaly ending 16 Oct 2007
http://www.cpc.ncep.noaa.gov/products/Drought Experimental Operational Forecasts – SM (mm) Made on 09Oct2007 Made on 02Oct2007 NAEFS Forecasts 10 Oct 2007–16 Oct 2007 RR 7-day mean anomaly CPC Leaky Bucket Model 7-day mean anomaly 10 Oct 2007–16 Oct 2007
Summary • Large systematic errors exist in the verified land-surface quantities, especially for soil moisture and evaporation. • The NAEFS do not initialize the land-surface conditions for forecasting, a large portion of the mean errors is inherited from the initial conditions. • Forecasts are significantly improved after the error correction. • The systematic error correction is not sensitive to the training period.
Future Plans Add more hydro-meteorological variables: Snow depth, Snow water equivalent, & Runoff. Use ensemble NLDAS to correct systematic errors. Leave no county behind - downscale NAEF forecasts forcing NLDAS models to produce ensemble forecast products at 1/8 degree. 4. Use CFS to extend forecasts beyond 2 weeks.