Demonstrate Improved Simulation of the Complex Dynamics of North American Coastal Systems by Comparing Models to Data
OVERALL PERFORMANCE MEASURES
Wind-generated waves are involved in several important processes in the global climate system. This includes the mediation of momentum, heat, and mass fluxes between the ocean and atmosphere (Cavaleri et al. 2012). Waves also play an important role in the cryosphere, where there are feedbacks between wave dissipation and sea ice fracture in the marginal ice zone (Squire et al. 1995). Of particular interest to U.S. Department of Energy (DOE) mission questions is the response of the water surface level to shoaling and breaking waves in coastal regions. This additional “wave setup” (Longuet-Higgins and Stewart 1964) can represent a large portion of the overall mean water surface elevation in tropical cyclone flooding events, with implications to energy infrastructure in coastal regions. The wave setup process is illustrated in Figure 1(a).
Wind waves occupy a portion of the energy/frequency spectrum that is distinct from the longer-period ocean waves (tides, storm surges, tsunamis, etc.), which are resolved in some ocean circulation models (Wright et al. 1999). Since wind waves have periods on the order of 1-10 seconds and wavelengths on the order of 10-100m, their time and length scales are too fine to be resolved explicitly over the entire globe. Therefore, “phase-averaged” wave models are typically employed in large-scale applications (Cavaleri et al. 2007). These models describe the evolution of wave action (which is closely related to wave energy) as it propagates in latitude, longitude, frequency, and directional space (Tolman 1991). The frequency/direction spectrum can then be used to calculate several statistical quantities describing the wave field such as significant wave height, mean direction, and mean period. Another important quantity is the Stokes drift, shown in Figure 1 (b), which is the mean velocity induced by wave motion in the propagation direction.
Since phase-averaged wave models resolve a frequency/direction spectrum at each model grid point in the ocean, the number of model unknowns is high. This large number of unknowns, combined with the complexity of wave physics parameterizations, which describe generation, dissipation, non-linear interactions, etc., makes these wave models very expensive additions to Earth system models.
Therefore, our goal is to use unstructured meshes in order to economically resolve wave processes globally across the open ocean and coastal regions of interest. Using these meshes, the computational expense associated with the high mesh resolution required for coastal regions can be balanced with efficient use of coarse resolution for open ocean basins. As will be shown in the results section, coarse unstructured meshes with coastal refinement can provide comparable accuracy to global high-resolution structured meshes. Unstructured meshes can achieve this level of accuracy at significantly reduced computational cost. This allows for an unprecedented capability to efficiently include wave processes at both global and coastal scales in Earth system models.
The most recent version (6.07) of the National Oceanic and Atmospheric Administration (NOAA) WAVEWATCHIII® model has been integrated into the DOE Energy Exascale Earth System Model (E3SM) as the wave model component. Initially, the model used the traditional structured mesh configuration. However, in order to enable global-to-coastal wave modeling for E3SM, modifications were made to extend WAVEWATCHIII® to global, unstructured mesh domains. Previously, unstructured meshes had been used successfully in hurricane wave prediction studies but were limited to regional domains (Abdolali et al. 2020). Here, we implement and validate, for the first time, the performance of unstructured meshes for global domains with coastal refinements, which are appropriate for climate modeling applications within E3SM.
The triangular unstructured mesh considered in this study was generated using open-source mesh generation software (Roberts et al. 2019) and was designed to align with early versions of meshes under consideration for version two of E3SM (Hoch et al. 2020). The mesh has 2-degree resolution globally and transitions to ½-degree resolution in regions shallower than 4km. The 4km threshold was chosen to resolve coastal areas between the continental shelf break and the shoreline. A 10% element grade is enforced between the ½- and 2-degree resolution. The resolutions chosen allows the unstructured mesh to be readily compared against structured meshes with global uniform resolutions of ½ degree and 2 degrees. An image of the mesh is shown in Figure 2.
This global unstructured mesh capability has been validated against wave buoy measurements in order to assess overall suitability for use in E3SM, in terms of both accuracy and efficiency. We have compared modeled wave results for June-October 2005 with observations from the National Data Buoy Center (NDBC; Meindl and Hamilton 1992) along the U.S. coast. Additionally, the comparisons between the ½-degree and 2-degree structured meshes provide a sense of how well the unstructured mesh balances the accuracy and efficiency of the two different resolutions. The model was forced using atmospheric data from the Climate Forecast System Reanalysis (CFSR; Saha et al. 2010) product and was not coupled to other Earth system model components for this study.
2ND QUARTER METRIC COMPLETED:
Extreme precipitation and subsequent severe flooding has been implicated as the primary cause of tropical cyclone (TC)-related fatalities over the past 30 years, as well as the leading cause of infrastructural damage related to these storms (Pielke Jr. et al. 2008). As such, TCs and TC-related flooding are responsible for persistent risks to the U.S. east and Gulf coasts. Investments in model improvement and computation at scale have enabled the Department of Energy (DOE) Energy Exascale Earth System Model (E3SM) to produce one of the most realistic Atlantic TC climatologies among global modeling systems (Balaguru et al. 2020). More specifically, global TC frequency, TC lifetime maximum intensities, and the relative distribution of TCs among the different basins are significantly better simulated at high-resolution compared to low-resolution models commonly used in the Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5/6). However, modeling TC coastal impacts also requires realistic simulations of the distribution and characteristics of landfalling TCs, which are sensitive to the large-scale environment in the North Atlantic basin as well as near the coasts. Furthermore, modeling TC‑related precipitation remains an ongoing challenge due to sensitivity of TC rainfall to the simulated TC structure and the physics parameterizations used in the models.
DOE investments in software products such as TempestExtremes (Ullrich and Zarzycki 2017) have allowed tracking and characterization of TCs to become a standard part of the E3SM workflow. Building upon this work, recent efforts have also led to a comprehensive and automated evaluation capability for TCs (Zarzycki et al. 2021, Stansfield et al. 2020), thus enabling developers to quickly identify potential model biases. Further efforts are underway to develop evaluation metrics and diagnostics that evaluate the underlying processes and large-scale environment of TCs, and impacts related to TCs.
In this document, we evaluate the performance of E3SM for modeling landfalling TC precipitation and demonstrate improvement in TC-related precipitation in E3SM at high resolution (28km grid spacing) compared to the standard low resolution (110km grid spacing typical for CMIP models). Our analysis shows that high model resolution enables more accurate simulation of the properties of landfalling storms along the Atlantic and Gulf coasts, particularly with regards to the storm structure. We also demonstrate that, under a variety of salient metrics, E3SM storm structure is particularly realistic, while bias in TC climatology largely comes from subtle biases in the large-scale environment of the atmosphere and ocean, requiring continued efforts in improving the coupled model.
The simulations evaluated in this study use E3SM version 1, as described by Caldwell et al. (2019), in its high-resolution (E3SM-HR) and low-resolution (E3SM-LR) configurations. The model is a fully coupled atmosphere-ocean global climate model (GCM), developed to support DOE’s energy mission (Leung et al. 2020). The atmospheric model is described in Rasch et al. (2019). It consists of a spectral‑element dynamical core with 72 vertical levels (Dennis et al. 2012) and parameterized physics processes including deep convection (Neale et al. 2008, Richter and Rasch 2008, Zhang and McFarlane 1995); macrophysics, turbulence, and shallow convection (Golaz et al. 2002, Larson 2017, Larson and Golaz 2005); microphysics (Gettelman and Morrison 2015, Gettelman et al. 2015); aerosol treatment (Liu et al. 2016); and radiative transfer (Iacono et al. 2008, Mlawer et al. 1997). The ocean and sea ice components employ the Model for Prediction Across Scales (Petersen et al. 2019, Ringler et al. 2013). A mesoscale eddy parameterization (Gent and McWilliams 1990) is used only for the E3SM-LR simulation but disabled in E3SM-HR. Neither the E3SM-HR nor the E3SM-LR configurations use a sub-mesoscale eddy transport scheme. The land model is similar to the Community Land Model version 4.5 (Oleson et al. 2013). The Model for Scale Adaptive River Transport (Li et al. 2013, 2015) is used for river routing.
The E3SM-LR simulation is conducted using an atmospheric grid spacing of 110km (1) and an ocean grid spacing that varies between 30 and 60km. The E3SM-HR simulation uses an atmospheric grid spacing of 28km (0.25) and an ocean grid spacing that varies between 8 and 16km. These simulations employ transient forcings following the High Resolution Model Intercomparison Project (Haarsma et al. 2016) protocol for the years spanning 1950 through 1969. Both the E3SM-HR and E3SM‑LR simulations share the same tuning parameter values − namely, the low-resolution configuration mirrors the “LRtunedHR” simulation described in Caldwell et al. (2019).
TCs were tracked in both data sets using the TempestExtremes software (Ullrich and Zarzycki 2017). The tracking criteria are identical to those recommended by Zarzycki and Ullrich (2017), first identifying candidates as minima in the sea-level pressure field, then culling candidates that do not have an upper‑level warm core (defined as a thickness anomaly in the upper-level geopotential). Candidates are then stitched together in time to form trajectories by seeking pairs of candidates at adjacent time levels that are within a specified maximum distance of one another.
3RD QUARTER METRIC COMPLETED:
Coastal zones, a tiny portion of the land-ocean continuum, are hotspots for ecological and socioeconomic activities. Coastal ecosystem dynamics are primarily controlled by coastal biogeochemistry, which is subject to the strong influences of terrestrial sediment and nutrient fluxes transported by rivers. For example, excess nutrient discharge from the Mississippi River has been a primary cause of the recurring hypoxic dead zone in the Gulf of Mexico (Rabalais et al. 2002, Turner et al. 2008, McLellan et al. 2015, Feng et al. 2019). In 2017, this hypoxic zone reached the size of 22,720 km2, the largest ever measured and as big as the entire state of New Jersey (Van Meter et al. 2018). Moreover, coastal carbon cycling is largely regulated by terrestrial sediment and nutrient fluxes due to the close physical and chemical interactions between various carbon elements and sediment and nutrient substances (Zaehle et al. 2011, Browning et al. 2017). Last but not least, the decline of territorial sediment flux is one of the major reasons for the shifting coastline in the past decades (Valderrama-Landeros et al. 2019, Warrick et al. 2019).
Sediment and nutrient transport from land to the ocean via rivers is generally lacking or underrepresented in Earth System models (ESMs) because ESMs lack physically based representation of riverine hydraulic and thermodynamic conditions critical for representing the transport and transformation of riverine sediment and nutrient fluxes. The riverine component of Energy Exascale Earth System Model (E3SM), Model for Scale Adaptive River Transport (MOSART), has been well tested and validated for simulating riverine hydraulic and thermodynamic processes in both natural and managed river systems (Li et al. 2013, 2015a, b, Voisin et al. 2013a, b, Yigzaw et al. 2018, 2019). In particular, MOSART incorporates reservoir regulation and thermal stratification processes that have significant effects on riverine hydraulic and thermodynamic conditions and, consequently, on riverine sediment and nutrient fluxes. Therefore E3SM has unique capabilities to represent sediment and nutrient fluxes from land to the ocean for modeling coastal biogeochemistry under climate and other human-induced changes.
This metric report summarizes recent efforts supported by the U.S. Department of Energy (DOE) in developing and evaluating new coastal modeling capabilities in E3SM including: (1) a process-based soil erosion model within the E3SM land model (ELM), ELM-Erosion, to simulate sediment yield from land surface to rivers (Tan et al. 2017, 2018), (2) a physically based riverine suspended sediment transport module within MOSART, MOSART-sediment (Li et al. in prep), together with a newly developed median sediment particle size map over the contiguous United States (CONUS) (Abeshu et al. in prep), and (3) a module to simulate erosional carbon, nitrogen, and phosphorus fluxes from land to rivers and coast, highlighting the significant role of extreme rainfall (Tan et al. 2020, 2021). These modules have been successfully validated against observations in the contiguous U.S., demonstrating the ability of E3SM in simulating soil erosion and hillslope sediment yield in CONUS and reproducing the observed long-term suspended sediment discharge in a number of U.S. Geological Survey (USGS) stations. The development of these components fills a key gap in coastal modeling and provides the foundation for further development in representing coastal biogeochemistry in E3SM.
The overall framework of modeling sediment and nutrient transport in E3SM is summarized in Figure 1. This document describes the development and testing of key components, including ELM‑Erosion, MOSART-sediment, and erosional carbon (C), nitrogen (N), and phosphorus (P) fluxes, representing processes shown by the solid lines, while ongoing and future development focuses on processes shown by the dashed lines.
ELM-Erosion is an event-scale soil erosion model developed based on the improved Morgan‑Morgan-Finney (MMF) soil erosion scheme (Tan et al. 2018). It simplifies the complex water erosion processes into rainfall-driven and runoff-driven erosion and defines hillslope sediment flux as the sum of the two erosional fluxes, capped by the sediment transport capacity (solid pink arrow in Figure 1 marked “hillslope routing” from ELM to MOSART). Soil erosion and sediment transport capacity in ELM-Erosion are represented as functions of E3SM simulated hydrological conditions (i.e., direct throughfall, leaf drainage, and surface runoff), vegetation conditions (i.e., plant leaf area index and canopy height), and soil texture, land slope, and land use. Hillslope fluxes of particulate organic carbon (POC), particulate nitrogen (PN), and particulate phosphorus (PP) are also represented in ELM-Erosion (solid dark green arrow in Figure 1 marked “hillslope routing” from ELM to MOSART), based on the modeled sediment flux and the E3SM-simulated surface soil organic C, solid-form N, and solid-form P content, and soil C, N, and P pools are updated based on the simulated POC, PN, and PP fluxes (Tan et al. 2020, 2021).
MOSART-sediment is a river sediment dynamics model developed on top of the MOSART river routing model and water management (WM) model. As illustrated in Figure 1 (solid pink arrows within MOSART), MOSART-sediment receives hillslope sediment flux calculated by ELM-Erosion and routes it through river channels as mud sediment (median diameter ≤ 0.0625 mm) that is assumed to be well mixed with water and not deposited on the channel bed. MOSART-sediment also simulates the transport of fine sand sediment (0.0625 < median diameter ≤ 0.25 mm), which is assumed to be produced only from channel erosion and is frequently exchanged with the material in the channel bed and in-channel bars and controlled by the local hydrodynamical and grain size conditions (Figure 1). The effects of reservoirs on suspended sediment, including the trapping of suspended sediment within the reservoirs and the regulation of the hydraulic conditions controlling the suspended sediment transport, are also explicitly represented in MOSART sediment through coupling between MOSART and WM. ELM-Erosion has been incorporated in E3SM v2 and MOSART-sediment will be released with E3SM v3. Enhancement of ELM-Erosion for the cropland management effect on soil erosion is ongoing and planned for release with E3SM v3. For MOSART, sediment deposition on floodplains and the transport and reactions of river POC and dissolved organic C (DOC) are being developed under the E3SM and Integrated Coastal Modeling (ICoM) projects.
Both ELM-Erosion and MOSART-sediment have been validated over CONUS using Phase 2 of the North American Land Data Assimilation System (NLDAS-2) atmospheric forcing data and the 0.125° NLDAS grid configuration. The model evaluation focused on the Mississippi river basin that contributes the most sediment, C, and nutrient fluxes to the U.S. coastal waters. Model parameters of soil erosion and hillslope sediment and POC fluxes were calibrated based on sensitivity experiments for 1991−2012. The simulated soil erosion and hillslope sediment and POC fluxes were compared with the U.S. Department of Agriculture (USDA) National Resources Inventory (NRI) state-level soil erosion estimate and the pre‑dam river sediment and POC yield data in large U.S. river basins, respectively. The hillslope PN and PP fluxes simulated by ELM-Erosion from 1991 to 2019 were used as input to an empirical nutrient spiraling model to estimate river PN and PP yields to oceans (Tan et al. 2021). The simulated C/N and C/P ratios in river-suspended particles and the amount of river PN and PP yields in large U.S. river basins were compared with published and USGS data.
Coupled with ELM, MOSART was run from 1991 to 2012 using the diffusive wave routing method and a new median sediment particle size map to evaluate the simulated streamflow and suspended sediment against measurements at 63 USGS river gauges with at least five years of daily records (Li et al. in review). In addition, three representative USGS gauges (one in the Missouri River and two in the Mississippi River) were selected to evaluate the effect of reservoir operation on suspended sediment. Three MOSART-sediment simulations were performed to isolate the net effects of reservoir regulation and trapping on suspended sediment transport. To represent the effects of reservoir operations, the locations of 1839 relatively large reservoirs (storage capacity ≥ 0.1 km3) were extracted from the GRanD database.