Ensemble Experiments in an AI-NWP Coupled Framework: A Typhoon Case

The well‐recognized constraint of nonlinear and non‐Gaussian distribution of rainfall observation limits its assimilation in the high‐dimensional numerical weather prediction (NWP) model. In this study, rainfall observed from Indian Meteorological Department (IMD) rain gauges over Indian landmass is assimilated in the Weather Research and Forecasting (WRF) model using particle filter. In the framework of imperfect weather models, particles (or ensembles) for rainfall predictions are created with various combinations of model physics (viz., cumulus parameterization, microphysics and planetary boundary layer schemes). The multiple hypotheses are used to determine the weights for different particles, and this is the step where IMD rainfall data are used for assimilation. Further, a resampling step is performed to generate new particles from high weight particles using stochastic kinetic‐energy backscatter scheme (SKEBS) method in which dynamical variables are perturbed into the model physics. Results, based on rainfall verification scores, suggest that the assimilation of the rainfall using particle filter could improve the prediction of rainfall over CNT runs (unweighted particles; without assimilation). Moreover, surface and vertical profile of temperature, water vapor mixing ratio (WVMR), and wind speed are also improved in 24‐hr forecasts.

Numerical weather prediction (NWP) models such as the Weather Research and Forecasting (WRF) model are increasingly used over the Indian region to forecast extreme rainfall events. However, studies which explore the application of high-resolution rainfall simulations obtained from the WRF model in urban hydrology are limited. In this paper, the utility of a model coupling framework to predict urban floods is explored through the case study of Bangalore city in India. This framework is used to simulate multiple extreme events that occurred over the city for the monsoons of years 2020 and 2021. To address the uncertainty from the WRF model, a 12-member convection permitting ensemble is used. Model configurations using Kain Fritsch and WSM6 parameterization schemes could simulate the spatial and temporal pattern of the selected event. The city is easily flooded with rainfall events above a threshold of 60 mm/day and to capture the response of the urban catchment, the Personal Computer Storm Water Management Model (PCSWMM) is used in this study. Flood forecasts are created using the outputs from the WRF ensemble and the Global Forecasting System (GFS). The high temporal and spatial resolution of the rainfall forecasts (<4 km at 15-min intervals), has proved critical in reproducing the urban flood event. The flood forecasts created using the WRF ensemble indicate that flooding and water levels are comparable to the observed whereas the GFS underestimates these to a large extent. Thus, the coupled WRF–PCSWMM modelling framework is found effective in forecasting flood events over an Indian city.

Assimilation of accurate external satellite-based water vapor data has been demonstrated to be able to improve the Numerical Weather Prediction (NWP) model’s forecasting performance. In this study, two types of near-infrared Precipitable Water Vapor (PWV) data from the Ocean and Land Colour Instrument (OLCI) of the Sentinel-3A and -3B satellites are assimilated into the Weather Research and Forecasting (WRF) model to enhance the weather forecasting performance over the South China area. One type is the raw PWV products obtained directly from the Sentinle-3A and -3B satellites; the other type is the Sentinel-3A and -3B PWV calibrated by PWV from Global Navigation Satellite System (GNSS). The Back Propagation Neural Network (BPNN) model is used in the PWV calibration. Correspondingly, two WRF assimilation schemes are adopted to assimilate the Sentinel-3 raw PWV and Sentinel-3 calibrated PWV. Additionally, the WRF without data assimilation scheme is used as the background run to investigate the impact of data assimilation. Three WRF experimental schemes are performed for two periods, i.e., February 2020 (dry period) and July 2020 (wet period). The forecasting performance of these three WRF schemes are evaluated by observations from GNSS, radiosonde, and surface meteorological stations. The comparison results with GNSS PWV show that assimilation of Sentienl-3 calibrated PWV can improve PWV forecasting accuracy by 5.6% for February period for the first 6 h after data assimilation. Evaluated by radiosonde profiles, the assimilation of Sentinel-3 both raw and calibrated PWV can improve WRF humidity and temperature forecasting accuracy over the July period. The accuracy improvement can be up to 9.2% (at 2 km to 3 km of altitude) for water vapor mixing ratio and 6.0% for temperature (at 4 km to 5 km of altitude). The WRF rainfall evaluation result has been validated by rainfall observations from meteorological stations and it shows that the assimilation of Sentinel-3 PWV can improve the rainfall forecasting performance over the July period. The rainfall forecasting success rates of three schemes of assimilating no data, Sentinel-3 raw PWV, and Sentinel-3 calibrated PWV are 69.4%, 69.6%, and 71.1%, respectively, over the July 2020 in the South China region.

<p>The moisture recycling is defined as the fraction of precipitation over a delimited region that comes from the evaporation over that region. Its importance lies in the fact that it is an approximated measurement of a regional feedback between the atmosphere and the surface. Thus, this study estimates the spatio-temporal distribution of moisture recycling over the Iberian Peninsula (IP), and focuses on the impact of the use of 3DVAR data assimilation during the modeling stage.</p><p>For that purpose, two different simulations were run using the Weather and Research Forecasting (WRF) model with a horizontal resolution of 15 km over the IP. The first simulation (WRF N) was nested inside ERA-Interim as usual in numerical downscaling exercises, with information passed to the domain through the boundaries. The second run (WRF D) presents the same configuration as WRF N, but it also includes 3DVAR data assimilation step every six hours (at 00, 06, 12 and 18 UTC). Sea surface temperature was updated daily, and observations in PREPBUFR format included in the NCEP ADP Global Upper Air and Surface Weather Observations dataset were used for the data assimilation step. Only those inside a 120-minute window centered at the analysis times were assimilated. Both simulations cover the period 2010-2014, but the experiment WRF D was extended later until 2018.</p><p>The lowest values of moisture recycling (around 3 %) are obtained from November to February, while the most remarkable values are observed in spring (around 16 %) in both simulations. The moisture recycling is confined to the southeastern corner of the IP during winter. However, during spring and summer, a gradient of higher values towards the northeastern corner of the IP are observed in both simulations. The differences between simulations, associated to the dryness of the soil in the model, are highlighted during summer and autumn. WRF D presents a lower bias and produces more reliable results because of a better representation of the atmospheric moisture.</p><p>A Cross-Correlation Function (CCF) based analysis was performed for each combination of moisture recycling, accumulated precipitation and mean soil moisture over the IP. For the common period (2010-2014), the results show that the WRF D experiment extends the lifespan of moisture over the IP. The CCF analysis for soil moisture against precipitation also shows an unphysical negative lag (-1 month) for WRF N, whilst for WRF D both variables are simultaneous. For the extended WRF D simulation (2010-2018), it was found that the delay between precipitation and moisture recycling over the IP is five months.</p>

The orthogonal conditional nonlinear optimal perturbations (O-CNOPs) approach for measuring initial uncertainties is applied to the Weather Research and Forecasting (WRF) Model to provide skillful forecasts of tropical cyclone (TC) tracks. The hindcasts for 10 TCs selected from 2005 to 2020 show that the ensembles generated by the O-CNOPs have a greater probability of capturing the true TC tracks, and the corresponding ensemble forecasts significantly outperform the forecasts made by the singular vectors, bred vectors, and random perturbations in terms of both deterministic and probabilistic skills. In particular, for two unusual TCs, Megi (2010) and Tembin (2012), the ensembles generated by the O-CNOPs successfully reproduce the sharp northward-turning track in the former and the counterclockwise loop track in the latter, while the ensembles generated by the other methods fail to do so. Moreover, additional attempts are performed on the real-time forecasts of TCs In-Fa (2021) and Hinnamnor (2022), and it is shown that O-CNOPs are very useful for improving the accuracy of real-time TC track forecasts. Therefore, O-CNOPs, together with the WRF Model, could provide a new platform for the ensemble forecasting of TC tracks with much higher skill.

Remote sensing precipitation or precipitation from numerical weather prediction (NWP) is considered to be the best substitute for in situ ground observations for flood simulations in transboundary, data-scarce catchments. This research was aimed to evaluate the possibility of using a combination of a satellite precipitation product and NWP precipitation for better flood forecasting in the transboundary Chenab River Basin (CRB) in Pakistan. The gauge-calibrated satellite precipitation product, i.e., Global Satellite Mapping of Precipitation (GSMaP_Gauge), was selected to calibrate the Integrated Flood Analysis System (IFAS) model for the 2016 flood event in the Chenab River at the Marala Barrage gauging site in Pakistan. Precipitation from the Global Forecast System (GFS) NWP, with nine different lead times up to 4 days, was used in the calibrated IFAS model to predict the flood hydrograph in the Chenab River. The hydrologic simulations, with global GFS forecasts, were unable to predict the flood peak for all lead times. Then, the Weather Research and Forecasting (WRF) model was used to downscale the precipitation forecasts with one-way and two-way nesting approaches. In the WRF model, the CRB was centered in two domains of 25 km and 5 km resolutions. The downscaled precipitation forecasts were subsequently supplied to the IFAS model, and the predicted simulations were compared to obtain the optimal flood peak simulation in the Chenab River. It was found in this study that the simulated hydrographs, at different lead times, from the precipitation of two-way WRF nesting exhibited superior performance with the highest R2 and Nash–Sutcliffe efficiency (NSE) and the lowest percent bias (PBIAS) compared with one-way nesting. Moreover, it was concluded that the combination of GFS forecast and two-way WRF nesting can provide high-quality precipitation prediction to simulate flood hydrographs with a remarkable lead time of 96 h when applying coupled hydrometeorological flow simulation.

Understanding the features of weather system around and over the Arabian Gulf is a prerequisite for setting up of a numerical weather forecasting system. The reanalysis, ERA-Interim data with a spatial resolution of ∼80 km has been utilized for analyzing the wind climatology for a period of 35 year. It is found that in the Arabian Gulf, throughout the year, the predominant wind is from north-west and the average wind speed is high in the summer between June and August. The high wind speed systems in summer are known as summer Shamals and have a typical duration of one week. Often, the high speed and long duration of the Summer Shamals makes it a good carrier of sand/dust. The winter season is dominated by south-east passage of frontal systems. The presence of land-sea contrast can cause mesoscale wind systems such as land-sea breeze. Qatar being a narrow peninsular region with width less than 100 km, the setting up of these land-sea breeze systems can be from either of the sides. This can lead to the interaction of two sea breezes from opposite sides, which may lead to the development of a convergence zone. This complex systems can affect the local weather and air-quality of coastal areas. These mesoscale phenomena and complexities are not well represented through a global model. Though the low resolution (80 km) data (ERA-interim) shows the presence of Shamal winds, the data misses the detail structure and their spatial variability. This necessitates for setting up of a high resolution mesoscale simulation approach.A non-hydrostatic mesoscale model, Weather Research and Forecast (WRF) developed by NCAR (Skamarok et al 2008) is used here for simulating and forecasting the wind systems associated with the Arabian Gulf. The model is run on an operational basis to forecast the wind fields, for a 48 hr. The modeling suite consists of a preprocessor, WPS, the model, WRF and a post processor, ARWpsot. The total suite has been compiled and installed successfully using FORTRAN and C compilers. A two way nested WRF domains with a grid size ratio of 1:3 is configured. The resolutions for the outer domain and inner domain are set to 9 km, and 3 km respectively. Various physical parametrization schemes are available within WRF modelling system. The physical parameterization schemes used in the current configuration includes MYJ boundary layer parametrization, Janjic Eta Monin–Obukhov surface layer scheme, Dudhia shortwave radiation, Rapid Radiative Transfer Model (RRTM) long wave radiation, WRF Single-Moment 6-Class (WSM6) microphysics, and the NOAH land surface scheme. Cumulus scheme, Grell is used only for the outer domain. Both of the model domains have a total of 39 vertical levels, with the topmost level at 50 hPa and lowermost level at approximately 30 m above the ground level. The initial and boundary conditions are obtained from the National Centers for Environmental Prediction's (NCEP) Global forecast System (GFS) available at resolution 0.50 (∼50 km), and the boundary conditions are updated at 6 hr interval. The entire process is automatized and run on RAAD Linux HPC cluster computing system.A two day simulation of WRF is analyzed and the sea breeze system of Qatar peninsula is examined and compared with the automatic weather station (AWS) data of Qatar Meteorological Department. The time of onset and duration of sea breeze are identified. The simulation of WRF clearly shows the development of a thermal internal boundary layer over the coastal city Doha. The existence of a convergence zone with high vertical updraft was also found. This may be due to the interaction of sea breezes coming from opposite sides of Qatar peninsula. Understanding the system of sea breeze would benefit the overall understanding pollutant transport and the dispersion mechanisms.

The spectral method is generally assumed to provide better numerical accuracy than the finite difference method. However, the majority of regional models use finite discretization methods due to the difficulty of specifying time-dependent lateral boundary conditions in spectral models. This study evaluates the behavior of nonhydrostatic dynamics with a spectral discretization. To this end, Juang’s nonhydrostatic dynamical core for the National Centers for Environmental Prediction (NCEP) regional spectral model has been implemented into the Regional Model Program (RMP) of the Global/Regional Integrated Model system (GRIMs). The behavior of the nonhydrostatic RMP is validated, and compared with that of the hydrostatic core in 2-D idealized experiments: the mountain wave, rising thermal bubble, and density current experiments. The nonhydrostatic effect in the RMP is further validated in comparison with the results from the Weather Research and Forecasting (WRF) model, which uses a finite difference method. The analyses of the experimental results from the RMP generally follow the characteristics found in previous studies without any discernible difference. For example, in both the RMP and the WRF model, the eastward-tilted propagation of mountain waves is very similar in the nonhydrostatic core experiments. Both nonhydrostatic models also efficiently reproduce the motion and deformation of the warm and cold bubbles, but the RMP results contain more small-scale noise. In a 1-km real-case simulation testbed, the lee waves that originate over the eastern flank of the Korean peninsula travel further eastward in the WRF model than in the RMP. It is found that differences of small-scale wave characteristics between the RMP and WRF model are mainly from the numerical techniques used, such as the accuracy of the advection scheme and the magnitude of the numerical diffusion, rather than from discrepancies in the spatial discretization.

This study is intended to simulate and better understand mean and extreme precipitation over the Loess Plateau (LP) in China using the Weather Research and Forecasting (WRF) model. We performed a long-term (1980–2011) simulation with the WRF model at a 10 km horizontal resolution, forced by the ERA-Interim reanalysis. A series of sensitivity experiments were conducted to investigate the influence of resolution and physical parameterization schemes on the simulation ability of the WRF model over the LP. The region has a dominant semi-arid climate and is adversely affected by extreme precipitation. Results show that the WRF model produced better simulations at a 10 km resolution for the LP with complex terrain than it did at coarser resolutions. WRF simulations for precipitation over the LP were most sensitive to cumulus schemes, moderately sensitive to planetary boundary layer schemes, and less sensitive to microphysics schemes. The WRF model not only adequately captured the spatial distribution of precipitation in the LP but also reproduced its magnitude and variability at different time scales. Moreover, the WRF model captured 73% of the observed extreme precipitation events with an intensity of more than 12 mm day− 1, which cause substantial soil erosion in the LP. Although it was unable to capture 27% of the observed events, the WRF model reasonably reproduced ERA-Interim extreme precipitation and its dynamic processes. These analyses strongly suggest that the biased extreme precipitation generated with the WRF model most likely resulted from inaccurate ERA-Interim forcing data. This study provides a better understanding of extreme precipitation simulations as well as insight into predictions of soil erosion in the LP.

Deficiencies of the nesting technique for regional models and their internal variability represent a significant source of uncertainties in regional model outputs. Presented numerical experiments on four different‐size domains, two of them being midlatitude channel domains, show that the spatial structure and magnitude of uncertainties strongly depend on the domain size. The experiments are performed by using the Weather Research and Forecasting (WRF) model nested into the operational European Centre for Medium‐Range Weather Forecasts (ECMWF) analyses on the same horizontal resolution 0.25°× 0.25° and by nesting WRF into a larger WRF domain with the same resolution. Uncertainties are quantified by the root‐mean‐square differences (rmsd) between the WRF results and their driving lateral boundary fields. The results from the midlatitude channel domain show that uncertainties in the tropospheric wind associated with the imperfect nesting method are amplified in the baroclinically active regions of the Atlantic and Pacific. The zonal wind rmsd have a dipole structure in the Atlantic in both the midlatitude channel and the half‐channel simulations nested into ECMWF. The dipole is absent when WRF is nested into itself. On the contrary, the maximal rmsd for the meridional wind are always located in the domain center. When the domain centered on Europe excludes the western Atlantic and North America, the simulated uncertainties become spatially nearly homogeneous, and the magnitude of rmsd due to the imperfect nesting technique greatly reduces.

A summertime convective event that developed on 5 September 2017 over the United Arab Emirates (UAE) is investigated in this study. Atmospheric profiles from a ground‐based microwave radiometer along with satellite observations and in situ data from three weather stations in the UAE were used. The event was simulated using the Weather Research and Forecasting (WRF) model, forced with four input datasets: Global Forecast System (GFS), Climate Forecast System Reanalysis (CFSR), and the European Centre for Medium‐Range Weather Forecasts ERA‐5 and ERA‐Interim reanalyses. The afternoon and evening convection was triggered by the intrusion of a mid‐level trough from midlatitudes and was favoured by the convergence over land of thermally forced maritime air masses. Near‐surface observations at a weather station revealed a 7 °C drop in air temperature, a doubling of the wind speed to 9 m·s−1 in 30 min, and a shift in wind direction from easterly, to southerly and then westerly in about 45 min, associated with the passage of the cold pool emanating from a Mesoscale Convective System (MCS). At the location where the microwave radiometer was deployed, the pre‐squall low and wake low signatures are also captured, with a 5 m·s−1 increase in the wind speed in just 5 min. The observed features of the studied MCS were found to compare with those reported for MCSs in the Tropics. The four experiments gave a similar performance, although the GFS simulation generally generated higher skill scores. The investigated MCS event was not captured by WRF, which was attributed to a misrepresentation of soil moisture in the model. This study highlights the difficulties regional models like WRF may have in reproducing MCSs over arid/hyperarid regions, which may result in a misrepresentation of their impacts in climate projection studies.

Impact of different reanalysis data on WRF dynamical downscaling over China

Abstract. Italy is a territory characterized by complex topography with the Apennines mountain range crossing the entire peninsula and its highest peaks in central Italy. Using the latter as our area of interest and the snow seasons 2018/19, 2019/20 and 2020/21, the goal of this study is to investigate the ability of a simple single-layer and a more sophisticated multi-layer snow cover numerical model to reproduce the observed snow height, snow water equivalent and snow extent in the central Apennines, using for both models the same forecast weather data as meteorological forcing. We here consider two well-known ground surface and soil models: (i) Noah LSM, an Eulerian model which simulates the snowpack as a bulk single layer, and (ii) Alpine3D, a multi-layer Lagrangian model which simulates the snowpack stratification. We adopt the Weather Research and Forecasting (WRF) model to produce the meteorological data to drive both Noah LSM and Alpine3D at a regional scale with a spatial resolution of 3 km. While Noah LSM is already online-coupled with the WRF model, we develop here a dedicated offline coupling between WRF and Alpine3D. We validate the WRF simulations of surface meteorological variables in central Italy using a dense network of automatic weather stations, obtaining correlation coefficients higher than 0.68, except for wind speed, which suffered from the model underestimation of the real elevation. The performances of both WRF–Noah and WRF–Alpine3D are evaluated by comparing simulated and measured snow height, snow height variation and snow water equivalent, provided by a quality-controlled network of automatic and manual snow stations located in the central Apennines. We find that WRF–Alpine3D can predict better than WRF–Noah the snow height and the snow water equivalent, showing a correlation coefficient with the observations of 0.9 for the former and 0.7 for the latter. Both models show similar performances in reproducing the observed daily snow height variation; nevertheless WRF–Noah is slightly better at predicting large positive variations, while WRF–Alpine3D can slightly better simulate large negative variations. Finally we investigate the abilities of the models in simulating the snow cover area fraction, and we show that WRF–Noah and WRF–Alpine3D have almost equal skills, with both models overestimating it. The equal skills are also confirmed by Jaccard and the average symmetric surface distance indices.

In this study, we conduct a 15‐year long (2008–2022) convection‐permitting simulation over the Tibetan Plateau (TP) using the Weather Research and Forecasting (WRF) model to evaluate its added value compared to ERA5 in characterizing subdaily precipitation and its extremes over the TP. Our findings indicate that the overestimation of precipitation amounts caused by overestimated precipitation frequency in ERA5 is mitigated in WRF mainly through a reduction in light precipitation occurrences. This improvement is likely linked to increased temperatures and decreased relative humidity in the lower troposphere. Moreover, the much earlier precipitation peak in ERA5 is corrected in WRF, a result attributed to differences in related net water vapor transport onto the TP. Additionally, WRF provides a more accurate estimation of the percentage of short‐ and long‐duration precipitation events with only slight underestimations and overestimations, respectively, whereas ERA5 exhibits significantly larger errors. The relatively larger temperature changes in WRF, indicating increased solar heating, are likely responsible for the higher occurrence of short‐duration precipitation events. In contrast, the anticyclonic anomalies and northeastern wind anomalies in WRF reduce the persistence of long‐duration precipitation. Importantly, WRF also produces more realistic timings of precipitation with different durations compared to ERA5. For extreme precipitation, short‐duration extreme events dominate accounting for 50% of extreme precipitation. These events are better reproduced by WRF than by ERA5. Although WRF outperforms ERA5 in estimating the proportion and contribution of extreme events to total precipitation and in capturing their triggering timing, a slight underestimation persists in the simulated proportion and contribution.

This paper evaluates microphysics and cumulus parameterization schemes of the Weather Research and Forecasting (WRF) model for high-resolution weather forecasting in tropics. WRF forecasts at 5-km resolution between 8 and 9 hours after initial time for 5 representative storm systems over Thailand and nearby regions for 24 different combinations of microphysics and cumulus parameterization schemes are evaluated using brightness temperatures coincidentally observed by the passive millimeter-wave spectrometer Advanced Microwave Sounding Unit (AMSU) aboard the U.S. National Oceanic and Atmospheric Administration (NOAA)-16 satellite and corresponding retrieved precipitation parameters from the AMSU MIT Precipitation retrieval (AMP) products. Results show that forecasts are very sensitive to microphysics and cumulus parameterization schemes. The combination of the WRF Double-Moment 6-class microphysics scheme and the Betts-Miller-Janjic cumulus parameterization scheme performs best. By using the best physics combination, WRF forecasts generally agree with AMSU observations in terms of morphologies, locations, and intensities. Main forecast errors include storm location offsets and under-forecasting of graupel.