The impact of WRF PBL parameterizations on the

The impact of WRF PBL parameterizations on the simulation of fire weather parameters over the Northeast U. S. Joseph J. Charney, Brian A. Colle, and Joseph Pollina Fire Summary Introduction Fire danger and fire behavior strongly depend upon local atmospheric conditions, both at the surface and aloft. High-resolution temporal and spatial information about the state and variability of atmospheric winds, temperatures, and moisture, especially within the PBL, provide inputs for fire danger and fire behavior models. Forecasts of these atmospheric quantities are used to anticipate elevated fire danger and the potential for erratic or extreme fire behavior. Since existing observation systems cannot readily diagnose the most relevant smallscale meteorological variations across highly variable land-surface and terrain characteristics, NWP models (such as the WRF model) are vital for both diagnosing and forecasting fire-weather parameters that serve as inputs for fire danger and fire behavior models. WRF PBL parameterizations This study investigates the impact of WRF PBL parameterizations on simulations of fire weather parameters associated with a single fire weather event that occurred in the Northeast U. S. Simulations of important fireweather ingredients such as surface wind speed, mixing ratio, relative humidity, and temperature are presented and compared against observed values at the time of the fire. The results from this study, as part of an ongoing project to enhance ensemble fire-weather forecasting in the Northeast U. S. , will be used to help understand the impact of WRF model physics variability on short-range ensemble forecast modeling systems during periods of elevated fire potential. The simulations presented in this study represent the meteorological conditions associated with the Warren Grove Wildfire in south-central New Jersey. Ignition Date: Tuesday, May 15, 2007 The Warren Grove Wildfire was a 15, 550 acre Wildland Urban Interface fire started by a flare dropped from an F-16 jet at the Warren Grove Target Range of the New Jersey Air National Guard. Initial attack on the 23 -acre fire was abandoned due to live, detonating ordinance at the fire site. The fire was described as “fast-moving, spotting, wind driven, with low humidity” (Warren Grove Major Wildfire Report, New Jersey Forest Fire Service, 2007). Firefighters protected the town of Warren Grove, as well as trailer parks and multiple housing developments from the effects of the fire. It is estimated that 9, 000 homes and structures were directly threatened by the fire. Property losses were limited to 3 homes destroyed and modest fire damage to 25 homes, 4 boats, 2 automobiles, and 1 trailer. Results (4 km simulations) WRF 3. 0 MYJ WRF 3. 1 MYNN WRF 3. 1 QNSE WRF 3. 0 MYJ WRF 3. 1 MYNN We perform four nested (36/12/4 km) simulations of the meteorological conditions associated with a single fire, using the following PBL parameterizations: 1. WRF 3. 0 with the Mellor-Yamada-Janjic (MYJ) Eta operational PBL. A one-dimensional prognostic turbulent kinetic energy scheme with local vertical mixing. 2. WRF 3. 1 with the MYJ PBL. 3. WRF 3. 1 with the Mellor-Yamada Nakanishi and Niino Level 2. 5 PBL (MYNN). Predicts sub-grid TKE terms. 4. WRF 3. 1 with the Quasi-Normal Scale Elimination PBL (QNSE). A TKE-prediction option that uses a new theory for stably stratified regions. Each simulation employs the North American Regional Reanalysis (NARR) for initial and boundary conditions, the Noah Land Surface Model, Eta similarity surface layer physics, the Kain-Fritsch Convective Parameterization, the RRTM longwave radiation scheme, and the Dudhia shortwave radiation scheme. Recent MM 5 PBL Comparisons (away from complex terrain) Zhang and Zheng (2004) investigate the performance of five PBL parameterizations using simulations of summertime weak-gradient flow periods over the central United States. They find that the MYJ parameterization generally underestimates daytime surface temperatures, and produces a colder, shallower mixed layer than other schemes. Daytime surface wind speeds are generally comparable to other schemes, but nocturnal surface winds are too strong. Zhong et al. (2007) assess the MRF and Eta PBL (MYJ) schemes, as well as land surface and radiation schemes, in a coastal environment (Texas) during a 10 -day period in July 2004. Their results indicate that the MYJ scheme exhibits a slight daytime cold bias and strongly over-predicts surface mixing ratios. Simulated surface RH at 1800 UTC shows that WRF 3. 1 is considerably warmer and drier at the surface than WRF 3. 0. The differences among WRF 3. 1 simulations using the MYJ, MYNN, and QNSE are less substantial. Simulated skew-T log-p diagrams at the fire location at 1800 UTC reveal that the low-level thermodynamic structure in WRF 3. 0 is considerably cooler and more moist throughout the mixed layer. WRF 3. 1 MYJ, MYNN, and QESN (not shown) results are nearly identical. Simulated time series at the fire location of surface RH, wind speed, mixing ratio, and temperature from each model run are compared against observed conditions at Mc. Guire AFB (KWRI) on May 15, 2007 from 1200 through 2100 UTC. All WRF 3. 1 runs produce warmer and drier conditions than WRF 3. 0 during the afternoon of the fire. Differences in surface wind speed smaller, but the MYNN parameterization produces the highest wind speeds of the four simulations, more closely approaching the observed wind speeds by 1900 UTC, which corresponds to the time when the wildfire was exhibiting particularly rapid spread. All of the WRF 3. 1 simulations produce lower relative humidities, lower mixing ratios, and higher temperatures than what was observed at the fire location, which contrasts sharply from the biases and errors reported in previous PBL inter-comparison studies. Ongoing and Future Work This study represents a preliminary component of ongoing research taking place under a cooperative agreement between the USDA Forest Service and Stony Brook University. The broader study is investigating the representation of fire weather parameters in a short range ensemble forecasting system. Comparisons are being performed between an established 6 -member WRF 2. 2 ensemble at 12 -km grid spacing and NWS station observations in the Northeast U. S. during the peak fire season (April), as well as for days of elevated fire potential during the past two years. Preliminary results indicate that all of the WRF members exhibit a cool and moist bias during the relatively dry periods that are typical of the Northeast U. S. fire season, with the largest moisture bias occurring in the MYJ PBL member. Han et al. (2007) compare the performance of five PBL parameterizations using three land-surface models in East Asia. They find that surface temperatures are consistently under-predicted while surface mixing ratio is over-predicted in most scenarios. Additionally, they find that the Noah LSM considerably enhances the quality of surface mixing ratio simulations. from Zhong et al (2007) The results presented here, as well as preliminary WRF 3. 1 simulations at Stony Brook University, indicate that WRF 3. 1 simulations in the Northeast U. S. fire cases are substantially warmer and drier than those produced by previous versions of the WRF model. Additional simulations and comparisons employing other PBL schemes, and other combinations of available land surface models and surface physics schemes, will be performed to document the performance of WRF 3. 1 and assess the potential impact of the WRF model on forecasts of fire-weather parameters in the Northeast U. S.