Assimilation of Surface Meteorological Observations in COAMPS® NAVDAS Dan Tyndall¹, Pat Pauley², Nancy Baker², and Clark Amerault² ¹National Research Council, Monterey, CA; ²Naval Research Laboratory, Monterey, CA

Introduction Case Study: 0000 UTC 01 Dec 2010 Results: 15 Day Study Period

Mesoscale surface observations are vital sources of data in meteorological applications such as The forecasts from 0000 UTC 01 Dec 2010 offer an excellent depiction of the impacts from the assimilation of surface observations. The Figure 5 shows root-mean-square (RMS) error and bias of 2-m air , 2-m dewpoint , short-term , power management, transportation safety, during this time was characterized by a ridge over the western United States and a strong low system just north of Lake Superior. The low pressure temperature, 10-m and speed over the 15 day study period, verified against wildfire management, dispersion modeling, and defense applications (Dabberdt et al. 2005; system produced strong recorded at many surface stations in and the eastern Dakotas (see the markers, which represent observation METAR observations (observations mostly from the NWS network category). As seen in Figure Horel and Colman 2005). Often, these observations are utilized in the creation of diagnostic wind speeds, depicted in the upper left and middle panel of Figure 3). While the lowest level observations do capture the high winds in the 5, the reduction in RMS error during the first 12 hours of the forecast caused by the mesonet surface analyses, which are typically used as end products to help diagnose current conditions region, the relatively few are unable to significantly adjust the 10-m wind analysis (Figure 3, upper left panel). Assimilation of the mesonet observation assimilation is statistically significant at the 99% confidence level. The lackof or verify previous forecasts (Tyndall and Horel 2013). However, some numerical weather surface winds causes the analysis wind speeds in the region to more closely match observations. The difference between the experimental and control forecast improvement after 12 hours is likely due to the increasing dominance of large scale prediction models are assimilating these observations to improve their initial conditions and 10-m wind analyses is depicted in the upper right panel of Figure 3; the mesonet observations increase analysis winds by 8 m/s in some areas. weather patterns on the small scale information captured by surface observations with time. forecasts (Benjamin et al. 2010). Mesonet observations also produced significant differences in the 2-m temperature analysis when compared to the control analysis (Figure 3, lower The reduction in wind direction RMS error at forecast hours beyond the analysis time is most

panels). Temperature observations reduced in the experimental analysis by 2-3 K over northern , and almost 1 K over much of the likely the result of COAMPS failing to maintain the strong winds in the analysis throughout the The NRL Atmospheric Variational System (NAVDAS) for the Coupled Midwest. The surface observations produced additional differences in the relative analyses of ±25% over much of the CONUS domain (not model integration. The lack of improvement in RMS error with mesonet Ocean-Atmosphere Mesoscale Prediction System (COAMPS®) assimilates very little surface shown). assimilation is mostly likely the result of high biased wind speed observations passing the data as part of its operational cycle—only wind measurements from coastal, ship, and buoy quality control—a situation which was not expected (generally, mesonet wind observations are observations are assimilated (temperature and humidity observations from these stations are Unfortunately, increased wind speeds in the experimental wind analysis are not maintained by the forecast model (see Figure 4). Within the first hour of low biased because many observations are measured at heights lower than 10-m). automatically rejected). The research presented here tests the assimilation of mesoscale the forecast, COAMPS reduces wind speeds in the region by more than half of the observed values. Forecasted wind speeds in the region continue to slow surface observations across the CONUS domain and their impact on COAMPS forecasts. through the next 5 hours of the forecast (but at a much slower rate). This erroneous behavior was not seen in the temperature or relative humidity Figure 6 shows innovation statistics from mesonet observations. Generally, mesonet forecasts for this time period (not shown). observations were relatively unbiased, except for pseudo-relative humidity (which showed a slight negative bias), and wind speeds (not shown), which may explain the lack of improvement Mesonet Assimilation and Experiment Methodology 2010120100 − 10 m Wind Speed Ctl. Obs. and Anl. − N2 2010120100 − 10 m Wind Speed Exp. Obs. and Anl. − N2 2010120100 − 10 m Wind Speed Anl. Diff. (Exp.−Ctl.) − N2 in wind speed RMS error as noted above. Surface observations are generally more difficult to assimilate than radiosonde or o o o observations due to representativeness issues between the model terrain elevation and the 48 N 48 N 48 N real terrain elevation. Figure 1 shows the elevation difference between the reported mesonet o o o station elevations and the 15-km model terrain used in this experiment. While large elevation 42 N 42 N 42 N differences are not surprising in the mountainous terrain of the western United States, there o o o 36 N 36 N 36 N are still some large terrain differences in the Midwest (± 50 m). To account for the terrain representativeness error, the base observation error (set individually for each network type) 30 oN 30 oN 30 oN for each mesonet observation is multiplied by the error multiplier depicted in Figure 2. 24 oN 24 oN 24 oN

120 oW 108 oW 96 oW 84 oW 72 oW 120 oW 108 oW 96 oW 84 oW 72 oW 120 oW 108 oW 96 oW 84 oW 72 oW

0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 −8 −6 −4 −2 0 2 4 6 8 Wind Speed Obs. and Anl. (m/s) Wind Speed Obs. and Anl. (m/s) Wind Speed Anl. Diff. (m/s) 48°N 14 2010120100 − 2 m Air Temp. Ctl. Obs. and Anl. − N2 2010120100 − 2 m Air Temp. Exp. Obs. and Anl. − N2 2010120100 − 2 m Air Temp. Anl. Diff. (Exp.−Ctl.) − N2

12

42°N 10 48 oN 48 oN 48 oN

8

o o o 36°N 6 42 N 42 N 42 N

4 Observation Error Multiplier Observation 36 oN 36 oN 36 oN 30°N 2

0 30 oN 30 oN 30 oN 0 100 200 300 400 500 24°N Terrain Difference (m)

120°W 108°W 96°W 84°W 72°W Figure 2. Observation error multiplier function 24 oN 24 oN 24 oN for terrain difference. 120 oW 108 oW 96 oW 84 oW 72 oW 120 oW 108 oW 96 oW 84 oW 72 oW 120 oW 108 oW 96 oW 84 oW 72 oW −500 −400 −300 −200 −100 0 100 200 300 400 500 Figure 1. Elevation difference (m) between reported mesonet station elevation and 15-km 250 260 270 280 290 300 310 250 260 270 280 290 300 310 −4 −3 −2 −1 0 1 2 3 4 Figure 5. Forecast RMS error and bias for 2-m temperature, 2-m dewpoint, and 10-m wind speed and direction over the 15 day resolution model terrain used in this research. Air Temp. Obs. and Anl. (K) Air Temp. Obs. and Anl. (K) Air Temp. Anl. Diff. (K) study period. Statistically significant differences are denoted by the larger markers/markers with outlines. Assimilated observations near the surface plotted

Surface Land/Surface Coastal Buoy/Ship/Drifter Lowest RAOB Mesonet In addition to terrain representativeness, differences between mesonet networks can also Figure 3. Control, experimental, and control-experimental analysis differences of 10-m wind speed (top panels) and 2-m air temperature (bottom panels) analyses from 0000 UTC 01 Dec 2010. Markers in the analysis panels denote Air Temperature Innovations − Nest 2 Pseudo−relative Humidity Innovations − Nest 2 lead to more difficulty in assimilating these surface observations compared to radiosonde near surface observations assimilated by each forecast. 35000 6000 assimilation. Network operators must balance sensor cost against the number of stations they 2010120100F000 2010120100F001 2010120100F002 2010120100F003 o o o o 30000 must deploy. Higher grade sensors generally are more accurate; however, the operator may 50 N 50 N 50 N 50 N 5000 25000 not be able to deploy as many stations due to the increased cost. Additionally, some operators 4000 o o o o design networks that try to measure local area atmospheric extremes instead of the local area 48 N 48 N 48 N 48 N 20000 3000

averages that would be forecasted by a numerical prediction model. For example, stations Count Count within the Remote Automated (RAWS) network are often located on 15000 46 oN 46 oN 46 oN 46 oN 2000 southern facing exposures for wildfire condition monitoring. Additionally, many networks that 10000 support agricultural operations utilize 3-m towers for wind observations, which can lead to 1000 5000 low-biased wind observations when compared to a 10-m wind field from a numerical model 44 oN 44 oN 44 oN 44 oN 0 0 (Tyndall and Horel 2013). Following Tyndall and Horel, were grouped into network −15 −10 −5 0 5 10 15 −60 −40 −20 0 20 40 60 Air Temperature Innovations (C) Pseudo−relative Humidity Innovations (%) categories based on their observational purpose. Networks with potential biased observations o o o o 42 N o o o o o o o 42 N o o o o o o o 42 N o o o o o o o 42 N o o o o o o o u Wind Component Innovations − Nest 2 v Wind Component Innovations − Nest 2 from siting or other reasons were set with higher observation errors (see Table 1).These 100 W 98 W 96 W 94 W 92 W 90 W 88 W 100 W 98 W 96 W 94 W 92 W 90 W 88 W 100 W 98 W 96 W 94 W 92 W 90 W 88 W 100 W 98 W 96 W 94 W 92 W 90 W 88 W 30000 30000 observation errors were multiplied by the terrain difference multiplier (Figure 2) to compute 2010120100F004 2010120100F005 2010120100F006 50 oN 50 oN 50 oN 25000 25000 the station’s observation error. The most trusted networks had their observation errors set to the radiosonde observation errors (at the lowest level near the surface). Because NAVDAS 20000 20000 o o o Surface Land/Surface Coastal computes analyses on pressure surfaces, only observations that also reported station pressure 48 N 48 N 48 N Buoy/Ship/Drifter 15000 15000 Count were assimilated. Mesonet observations within a ±30 minute time wind about the analysis Lowest RAOB Count Mesonet time were used. 10000 10000 46 oN 46 oN 46 oN

Network Purpose/Type 5000 5000 Abbreviation o o o 0 5 10 15 20 AG Agricultural networks 3 20 3 44 N 44 N 44 N Wind Speed Obs. and Fcst. (m/s) 0 0 −40 −30 −20 −10 0 10 20 30 40 −40 −30 −20 −10 0 10 20 30 40 AQ Air quality monitoring 3 20 3 u Wind Component Innovations (m/s) v Wind Component Innovations (m/s) EXT Offshore, Canadian, and Mexican networks 3 20 3 o o o Figure 6. Mesonet observation innovations over 15 day study period for air temperature, pseudo-relative humidity, and u and v wind 42 N o o o o o o o 42 N o o o o o o o 42 N o o o o o o o FED+ Federal networks, W. Texas mesonet 2 10 2 100 W 98 W 96 W 94 W 92 W 90 W 88 W 100 W 98 W 96 W 94 W 92 W 90 W 88 W 100 W 98 W 96 W 94 W 92 W 90 W 88 W components. HYDRO Hydrological networks 4 25 4 Figure 4. First 6 hours of 10-m wind speed forecasts and the analysis from 0000 UTC 01 Dec 2010 over Minnesota. Markers in the analysis panel denote near surface assimilated wind speed observations, while markers in the LOCAL Commercial, state, and local mesonets 3 20 3 forecast panels show all mesonet wind speed observations recorded within ±30 minutes of the verification time (some of these observations in the forecast panels may not pass quality control procedures, as they were not assimilated; they are shown for reference). NWS NWS/FAA/synoptic stations 2 10 2 PUBLIC Primarily citizen weather observing program (CWOP) 3 20 4 Future Work and Acknowledgements

RAWS Fire weather/remote area monitoring 4 25 4 References Future work for this research will include studying the impacts of expanding the assimilation TRANS Road and rail weather monitoring 3 20 3 Table 1. Observation errors as a function of parameter and network category. Benjamin, S., B. D. Jamison, W. R. Moninger, S. R. Sahm, B. E. Schwartz, and T. W. Schlatter, 2010: Relative short-range forecast impact from aircraft, profiler, time window for mesonet observations, attempting to remove biases from mesonet observations, as well as evaluating the benefits of mesonet assimilation on COAMPS forecasts Impacts from mesonet observation assimilation were assessed using a data withholding radiosonde, VAD, GPS-PW, METAR, and mesonet observations via the RUC hourly assimilation cycle. Mon. Wea. Rev., 138, 1319-1343. by utilizing the COAMPS adjoint to determine observation impacts by mesonet observation and methodology on forecasts between 0000 UTC 01 Dec 2010 and 1800 UTC 15 Dec 2010. The Dabberdt, W. F., and Coauthors, 2005: Multifunctional mesoscale observing networks. Bull. Amer. Meteor. Soc., 86, 961-982. mesonet network. control and experimental forecasts (which assimilated the mesonet data) were both initialized Horel, J. D. and B. Colman, 2005: Real-time and retrospective mesoscale objective analyses. Bull. Amer. Meteor. Soc., 86, 1477-1480. by COAMPS forecasts made between 27-30 Nov 2010 (this warm up period did not use any We gratefully acknowledge support from the Office of Naval Research and the Naval Research mesonet observation data). Tyndall, D. P., J. D. Horel, 2013: Impacts of Mesonet Observations on Meteorological Surface Analyses. Wea. Forecasting, 28, 254-269. Laboratory under program element 062435N.