Demonstrate improved DOE-E3SM simulation and scientific
understanding of mesoscale convective systems over North America
OVERALL PERFORMANCE MEASURES
1ST QUARTER METRIC COMPLETED:
Mesoscale convective systems (MCS) are a complex of thunderstorms that develop into a single entity with precipitation covering hundreds of kilometers and persist for several hours or more (Houze 2004, 2018). They influence the climate system by (1) releasing latent heat that drives the general circulation, (2) providing a dominant source of precipitation in many regions, (3) redistributing and removing moisture, clouds, and aerosols, and (4) affecting the radiative transfer in the atmosphere. An important element for accurately representing the features of MCS in Earth system models is the simulation of cloud microphysical processes and their interaction with relevant dynamical processes. Significant shortcomings exist in the Gettelman and Morrison (2015) (MG2) cloud microphysics model currently used in the Energy Exascale Earth System Model (E3SM) model. These include: (1) MG2 does not consider rimed precipitating ice particles (graupel/hail), which are important contributors to precipitation in MCS, and (2) MG2 artificially partitions frozen particles into cloud ice and snow based on fixed parameters, which is unphysical (Morrison and Milbrandt 2015). These limitations are becoming untenable in simulating cloud and precipitation for addressing water cycle science questions as E3SM increasingly resolves strong updrafts that produce rimed particles.
The Predicted Particle Properties (P3) is a recently developed cloud microphysics model (Morrison and Milbrandt 2015, Morrison et al. 2015) that addresses the above limitations. It has more advanced treatments of ice microphysics parameterization in which ice particle properties such as the relative degree of riming and vapor deposition in their growth are predicted and evolve locally. P3 has been implemented into the E3SM model in order to improve the simulation of microphysical processes associated with mesoscale convective systems.
In this document, we test the performance of E3SM using the P3 microphysics compared to the original MG2 microphysics at a regionally refined mesh of ¼-degree grid spacing over the Contiguous United States. E3SM simulations are evaluated against observational data from the Next Generation Weather Radar (NEXRAD) network. The results show that the inclusion of rimed hydrometeors in the P3 microphysics improves the probability distribution function of simulated precipitation rates and MCS characteristics such as number, duration, and size. However, improvements are still needed to better simulate the diurnal cycle of MCSs and the partitioning of precipitation between MCS and non-MCS events. In future work, P3 will be extended to better represent additional microphysical processes such as hydrometeor sedimentation and evaporation. Additionally, non-hydrostatic dynamics are being implemented in E3SM, which will allow the model to be applied at higher resolution down to grid spacing of a few kilometers on which the role of cloud microphysics parametrization in MCS simulations can be better estimated.
Since P3 was originally designed for high-resolution weather models at kilometer-scale grid spacings, changes are necessary to adapt the scheme to the coarser grid spacings and larger time steps used in E3SM. Changes include cloud/precipitation fraction treatment, subgrid distribution of hydrometeors, and homogenous aerosol freezing for cirrus clouds. We have also modified the coefficients for autoconversion and accretion process rates to be consistent with the adjustments recently made in MG2 for climatology simulations.
Evaluation of MCS in E3SM was carried out for simulations using a regionally refined mesh (RRM). The Great Plains of the U.S. provide an excellent venue for studying continental long-lasting and propagating MCSs. Therefore, an RRM of ¼-degree grid spacing was set up for the Continental United States (CONUS) using the E3SM Atmosphere Model Version 1 (EAMv1) as shown in Figure 1. In this configuration, CONUS has a grid spacing of 25 km and the rest of the globe has a grid spacing of 110 km. To reduce model uncertainties caused by modeled large-scale circulation, a linear nudging technique is employed to constrain the large-scale circulation from reanalysis data (Zhang et al. 2014, Sun et al. 2019). Two model simulations were run using the microphysics schemes of P3 and MG2, respectively, with a simulation time period from 1 January to 30 September 2011. Analysis is performed for the spring season (March-April-May 2011) during which there were several MCSs observed.
2ND QUARTER METRIC COMPLETED:
Mesoscale convective systems (MCSs) consist of an assembly of cumulonimbus clouds on scales of 100 km or more and produce mesoscale circulations (Houze 2004, 2018). As the largest form of deep convective storms, MCSs contribute to 30%–70% of annual and warm season rainfall as well as over half of the extreme daily rainfall events in the U.S. east of the Rocky Mountains (Stevenson and Schumacher 2014, Feng et al. 2019, Haberlie and Ashley 2019). MCSs are notoriously difficult to simulate in global climate models (GCMs). Failure in simulating MCSs in the Central U.S. is manifested in the erroneous diurnal cycle of precipitation and large warm bias in the near-surface temperature (Lin et al. 2017).
The Energy Exascale Earth System Model (E3SM) has been developed to support the U. S. Department of Energy (DOE)’s energy mission by providing an improved capability to predict future changes in the water cycle, biogeochemical cycle, and cryosphere systems that affect energy production and use. As part of the water-cycle experiments, E3SM v1 has been configured at low (~ 100 km) and high (~ 25 km) resolution to evaluate the impacts of model resolution on simulating water-cycle processes such as precipitation, snowpack, and runoff (Caldwell et al. 2019). This document summarizes analyses performed to evaluate a high-resolution E3SM simulation at 25-km resolution, focusing particularly on its ability to reproduce the observed MCSs and their characteristics in the central U.S. An MCS tracking algorithm is applied to both observations and the model simulation and various features such as MCS number, rain frequency and intensity, and their variability at seasonal and diurnal timescales are evaluated. As the resolution of E3SM continues to increase in the future, evaluating the model skill in simulating MCSs over multiple versions with improved physics parameterizations and model resolution is important for documenting progress towards advancing modeling of water-cycle processes in E3SM.
As part of the water-cycle experiments, E3SM v1 has been used to perform simulations using a low resolution (LR) and a high-resolution (HR) configuration following the Coupled Model Intercomparison Project Phase 6 (CMIP6) HighResMIP protocol (Haarsma et al. 2016). The LR and HR configurations feature the atmosphere and land models at ~100-km and ~25-km grid spacing and the ocean and sea ice models at 30–60-km and 6–18-km grid spacing, respectively. In the control simulations of HighResMIP, time-invariant 1950 forcings are prescribed. Caldwell et al. (2019) reported analysis and comparison of two 50-year control simulations at LR and HR. In these simulations, the ocean and sea ice were initialized based on standalone simulations of the E3SM ocean and sea ice models driven by climatological atmospheric forcing. The HR simulation was then extended for 20 years to archive high frequency model output, including hourly precipitation and 6-hourly, three-dimensional atmospheric temperature, moisture, and winds, among other variables. These high-frequency results are analyzed and evaluated in this document.
To evaluate the E3SM HR simulation of MCSs, the FLEXTRKR algorithm (Feng et al. 2018, 2019) is used to identify and track MCSs in both simulation and observations. An MCS is defined as a large cold cloud system (CCS) with brightness temperature (Tb) < 241 K and an area exceeding 6 x 104 km2 that contain a precipitation feature (PF) with major axis length > 100 km. Furthermore, the PF area, mean rain rate, and rain rate skewness must be larger than the lifetime-dependent thresholds. An MCS is tracked when both conditions of CCS and PF are met continuously for longer than six hours. For the E3SM HR simulation, hourly precipitation and outgoing longwave radiation are used to track MCSs. For observations, hourly satellite infrared data and Next-Generation Radar Network (NEXRAD) precipitation data for 2004−2016, coarsened to 25-km resolution, are used to track MCSs for comparison. A schematic illustration of MCS tracking using FLEXTRKR is shown in Figure 1. As MCS features are not well defined at 100-km grid spacing, only the E3SM HR simulation is analyzed and reported here.
3RD QUARTER METRIC COMPLETED:
Mesoscale convective systems (MCSs) consist of assemblies of cumulonimbus clouds on scales of 100 km or more and produce mesoscale circulations (Houze 2004, 2018). As the largest form of deep convective storms, MCSs contribute to 30%–70% of annual and warm-season rainfall as well as over half of the extreme daily rainfall events in the U.S. east of the Rocky Mountains (Stevenson and Schumacher 2014, Feng et al. 2019, Haberlie and Ashley 2019). As part of the water cycle experiments, the Energy Exascale Earth System Model (E3SM) v1 has been configured at low (~ 100 km) and high (~ 25 km) resolution to evaluate the impacts of model resolution on simulating water cycle processes such as precipitation, snowpack, and runoff (Caldwell et al. 2019). Since MCSs contribute importantly to mean and extreme precipitation in the U.S. and many other regions around the world, understanding how well they are simulated by the model may guide future development towards more skillful modeling of convective storms and associated hydrologic impacts. The FY2020 Second Quarter Performance Metric Report documented comparisons of MCSs in the central and eastern U.S. in a high-resolution simulation produced by E3SM v1 at 25-km resolution with observations. MCSs in the simulation occur less frequently and produce less intense precipitation, resulting in a large underestimation of MCS volumetric rain-rate compared to observations. Model biases in simulating MCSs may be attributed to model limitations in parameterizing convection, clouds, and other related processes but model biases in simulating the MCS large-scale environment may also play an important role. This document summarizes analyses performed to evaluate the springtime MCS large-scale environment in the E3SM v1 high-resolution simulation, making use of a 20-year segment of the simulation with a high-frequency model output of atmospheric circulation to understand the contribution of large-scale circulation biases to modeling MCSs in the U.S. Combined with the MCS tracking used in the second quarter metric report, the analysis of large-scale environment described in this report lays the foundation for investigating how MCSs and their large-scale environment may change in the future, with implications for water availability and floods.
E3SM v1 has been used to perform simulation at high resolution (HR) following the Coupled Model Intercomparison Project Phase 6 (CMIP6) HighResMIP protocol (Haarsma et al. 2016). The HR configuration features the atmosphere and land models at ~25-km grid spacing and the ocean and sea ice models at 6–18-km grid spacing. Following HighResMIP, a 100-year control simulation with time-invariant 1950 forcing has been completed. Caldwell et al. (2019) documented the model configuration and evaluation of key climatological features of the HR simulation. Here, 6-hourly model outputs of atmospheric circulation including winds, specific humidity, temperature, and geopotential height at 925-, 500-, and 200-hPa pressure levels from a 20-year segment of the HR simulation are used to evaluate the springtime MCS large-scale environment over the U.S. to understand the model biases in simulating MCSs.
Song et al. (2019) identified four types of synoptic environments that support the development of MCSs east of the Rocky Mountains during springtime. These environments are associated with frontal systems that provide a lifting mechanism and an enhanced Great Plains low-level jet (GPLLJ) that provides anomalous moisture for convection. During summer, MCSs often develop in the presence of high-pressure systems over North America that suppress convection, so smaller-scale dynamical and/or thermodynamic perturbations are needed to initiate MCSs. Summer MCSs are, therefore inherently less predictable because large-scale circulation plays a less important role in their development. The analysis reported here focuses on springtime to elucidate the role of large-scale circulation biases on simulating MCSs.
To determine how well the four types of large-scale environments favorable for springtime MCSs are simulated by the model, a similarity metric is used to determine how closely the atmospheric circulation of each 6-hour snapshot from the 20-year segment of E3SM HR simulation resembles each of the four observed patterns of favorable large-scale environments. This allows us to quantify the frequency of occurrence of each type of favorable MCS environment in the simulation. Comparison of the frequency of occurrence, as well as the spatial pattern of the favorable MCS environment in the simulation and observation based on the North American Regional Reanalysis (NARR) (Mesinger et al. 2006), provides an indication of the contribution of large-scale circulation biases to the biases in modeling MCS frequency and characteristics. The similarity metric can also be designed to attribute the biases in the frequency of occurrence of the favorable MCS environments in the model to biases in simulating the winds and moisture fields of the favorable environments to gain some insights on the sources of the large scale circulation biases.
4TH QUARTER METRIC COMPLETED:
Mesoscale convective systems (MCSs) consist of assemblies of cumulonimbus clouds on scales of 100 km or more and produce mesoscale circulations (Houze 2004, 2018). As the largest form of deep convective storms, MCSs contribute to 30%–70% of annual and warm season rainfall as well as over half of the extreme daily rainfall events in the U.S. east of the Rocky Mountains (Stevenson and Schumacher 2014, Feng et al. 2019, Haberlie and Ashley 2019). Since MCSs contribute importantly to mean and extreme precipitation in the U.S. and many other regions around the world, understanding how well they are simulated by E3SM may guide future development towards more skillful modeling of convective storms and associated hydrologic impacts.
The FY2020 Second Quarter Performance Metric Report documented comparisons of MCSs in the central and eastern U.S. in a high-resolution simulation produced by E3SM v1 at 25-km resolution (Caldwell et al. 2019) with observations. MCSs in the simulation occur less frequently and produce less intense precipitation, resulting in large underestimation of MCS volumetric rain-rate compared to observations. The FY2020 First and Third Quarter Performance Metric Reports indicated that these model biases in simulating MCSs can be attributed to model limitations in parameterizing convection, clouds, and other related processes, as well as model biases in simulating the MCS large-scale environment. Recently several new developments in parameterizing deep convection have been implemented in E3SM targeted for use in future E3SM versions. In this FY2020 Fourth Quarter Performance Metric Report, we evaluate MCSs simulated by E3SM with and without the new developments in convection parameterizations. The goal is to highlight aspects of MCSs that have been improved with the new model features and identify aspects that need more work to produce more realistic simulations of MCSs in the future. Overall, the new developments show a clear improvement in capturing the diurnal timing of MCS initiation and mature stages in the central U.S. as well as the frequency and intensity of MCSs over ocean in the Indo-Pacific sector. The frequency and intensity of MCSs are still substantially underestimated: effort is underway to address this bias by the incorporation of a stochastic convection scheme, an advanced cloud microphysical scheme, and higher-resolution representations of the MCSs.
Impacts of two recent developments in convection parameterizations for E3SM v2/v3 on simulating MCSs are examined using E3SM simulations at ~25-km grid spacing. The tested new features are (1) a new convective triggering function that combines the dynamic Convective Available Potential Energy (dCAPE) trigger and the Unrestricted air parcel Launch Level (ULL) approach, i.e., the dCAPE-ULL trigger, described in Xie et al. (2019); and (2) the Multiscale Coherent Structures Parameterization (MCSP) described in Moncreff (2019).
The dCAPE-ULL trigger introduces a dynamic constraint on the initiation of convection that emulates the collective dynamical effects to prevent convection from being triggered too frequently, as well as to allow air parcels to launch above the boundary layer (BL) to capture nocturnal elevated convection. The former is to address the well-known problem with the deep convection scheme developed by Zhang and McFarlane (1995) (ZM hereafter) commonly used in earth system models including E3SM, which triggers convection too frequently, leading to spurious precipitation during daytime, particularly over land. The latter is to improve the model capability to capture nocturnal elevated convection that is often associated with propagating MCSs frequently found east of the Rocky Mountains in the U.S. and in other regions but missing in most climate models. Using the dCAPE-ULL trigger in E3SM has resulted in significant improvements in simulating the phase of the diurnal cycle of precipitation in the central U.S. by suppressing spurious convection during daytime and better capturing nocturnal elevated precipitation (Xie et al. 2019). However, no clear improvement is found in the amplitude of model precipitation, which is too weak compared to observations.
The MCSP scheme is developed specifically for representing MCSs in climate models in which they are currently missing as they are neither parameterized nor resolved (Moncreff 2019). The basic idea is to add a heating term that is representative of mesoscale heating observed in organized convection, which consists of heating in a stratiform region (in the upper troposphere), and equal magnitude evaporative cooling below (lower troposphere), as well as to consider mesoscale momentum transport. This additional vertical heating structure is proportional to the heating in the ZM scheme. Combining the MCSP scheme with the ZM scheme, initial analysis shows a reduction of heating produced by the ZM scheme, indicating the impact of the MCSP scheme in suppressing deep convection in the model. Overall, the impact of MCSP in E3SM is primarily on tropical waves (e.g., Kelvin wave) and Madden-Julian Oscillation (MJO), which are much improved. Its impact on mean precipitation is minor, generally with slightly weaker precipitation seen over land and slightly stronger precipitation over ocean.
The results discussed in this metric report are based on 3-year-long atmosphere-land simulations with present-day forcing for the year 2010 along with weekly sea surface temperature (SST) and sea ice prescribed from the observations for years 2010−2012. In addition to examining the individual impacts of the two new features (dCAPE-ULL and MCSP) on MCSs, their combination is also tested to understand their overall effect. Table 1 summarizes the simulations performed for this report. Notably, previous efforts only tested dCAPE-ULL and MCSP in low-resolution simulations (~100 km). Although the impacts of dCAPE-ULL and MCSP on precipitation have been evaluated in previous studies, their influence on MCS has not been specifically investigated. Furthermore, these new model features have not been tuned for simulations at 25-km grid spacing, which is important for optimizing model performance (e.g., Caldwell et al. 2019). Hence the results presented in this report represent an initial effort to evaluate the impacts of the new features on MCS simulation at 25-km grid spacing, which will be followed by more extensive testing and evaluation as guided by the preliminary results reported here.
Similar to the second quarter metric report, MCSs in observations and model simulations are tracked using an updated version of the MCS tracking algorithm (Feng et al. 2020), which is adjusted for use with datasets at 25-km grid spacing. Given that both dCAPE-ULL and MCSP have shown a larger impact on mean precipitation in the tropics, the Indo-Pacific sector where MCSs occur frequently is also selected in the analysis besides the central U.S. In this report, an MCS is defined as a large cold cloud system (CCS) with brightness temperature (Tb) < 241 K and an area exceeding 6 x 104 km2 that contains a precipitation feature (PF) with major axis length > 100 km. Furthermore, the PF area, mean rain rate, and rain rate skewness must be larger than the lifetime-dependent thresholds. An MCS is identified when both conditions of CCS and PF are met continuously for longer than 6 hours. For the E3SM simulation, hourly precipitation and outgoing longwave radiation are used to track MCSs. In addition, MCS criteria for E3SM have been relaxed compared to observations to allow more tracked MCSs for diagnostic purposes. The CCS area and duration are reduced by 50% to 3 x 104 km2 and 3 hours, and the PF area, mean rain rate and rain rate skewness are also relaxed. For observations in the central U.S., hourly satellite infrared data and Next-Generation Radar Network (NEXRAD) precipitation data for the same years as the model, coarsened to 25-km resolution, are used to track MCSs for comparison.
In the Indo-Pacific sector, the 10-km hourly GPM satellite IMERG V06B precipitation data is used. The GPM IMERG precipitation data is a unified precipitation retrieval data set from a network of partner satellites in the GPM constellation (Huffman et al. 2019). The primary precipitation estimates in IMERG is from passive microwave (PMW) sensors. A quasi-Lagrangian interpolation scheme is applied to the gridded PMW precipitation estimates to fill in the gaps between PMW overpasses using motion vectors derived from total precipitable water vapor from numerical models in V06 (Tan et al. 2019). The IMERG V06B data used in this report has 0.1° x 0.1° and hourly resolution. MCS tracking in the Indo-Pacific region uses slightly different PF criteria as that in the U.S. to adapt for use with the IMERG precipitation data set. The PF thresholds for the IMERG satellite data are chosen to best match NEXRAD-based MCS statistics over the U.S. Results reported here are for the boreal summer season (June-July-August, JJA) that represents an even bigger challenge for E3SM to capture MCSs than the spring season (see the second performance metrics report), but similar model behaviors are found in the spring season.