Abstract This paper describes the development of U-net++ models, a type of neural network that performs deep learning, to emulate the shortwave Rapid Radiative-transfer Model (RRTM). The goal is to emulate the RRTM accurately in a small fraction of the computing time, creating a U-net++ that could be used as a parameterization in numerical weather prediction (NWP). Target variables are surface downwelling flux, top-of-atmosphere upwelling flux ( ), net flux, and a profile of radiative-heating rates. We have devised several ways to make the U-net++ models knowledge-guided, recently identified as a key priority in machine learning (ML) applications to the geosciences. We conduct two experiments to find the best U-net++ configurations. In Experiment 1, we train on non-tropical sites and test on tropical sites, to assess extreme spatial generalization. In Experiment 2, we train on sites from all regions and test on different sites from all regions, with the goal of creating the best possible model for use in NWP. The selected model from Experiment 1 shows impressive skill on the tropical testing sites, except four notable deficiencies: large bias and error for heating rate in the upper stratosphere, unreliable for profiles with single-layer liquid cloud, large heating-rate bias in the mid-troposphere for profiles with multi-layer liquid cloud, and negative bias at lowzenith angles for all flux components and tropospheric heating rates. The selected model from Experiment 2 corrects all but the first deficiency, and both models run ~10 4 times faster than the RRTM. Our code is available publicly.
more »
« less
Estimating Full Longwave and Shortwave Radiative Transfer with Neural Networks of Varying Complexity
Radiative transfer (RT) is a crucial but computationally expensive process in numerical weather/climate prediction. We develop neural networks (NN) to emulate a common RT parameterization called the Rapid Radiative Transfer Model (RRTM), with the goal of creating a faster parameterization for the Global Forecast System (GFS) v16. In previous work we emulated a highly simplified version of the shortwave RRTM only—excluding many predictor variables, driven by Rapid Refresh forecasts interpolated to a consistent height grid, using only 30 sites in the Northern Hemisphere. In this work we emulate the full shortwave and longwave RRTM—with all predictor variables, driven by GFSv16 forecasts on the native pressure–sigma grid, using data from around the globe. We experiment with NNs of widely varying complexity, including the U-net++ and U-net3+ architectures and deeply supervised training, designed to ensure realistic and accurate structure in gridded predictions. We evaluate the optimal shortwave NN and optimal longwave NN in great detail—as a function of geographic location, cloud regime, and other weather types. Both NNs produce extremely reliable heating rates and fluxes. The shortwave NN has an overall RMSE/MAE/bias of 0.14/0.08/−0.002 K day−1for heating rate and 6.3/4.3/−0.1 W m−2for net flux. Analogous numbers for the longwave NN are 0.22/0.12/−0.0006 K day−1and 1.07/0.76/+0.01 W m−2. Both NNs perform well in nearly all situations, and the shortwave (longwave) NN is 7510 (90) times faster than the RRTM. Both will soon be tested online in the GFSv16.
Radiative transfer is an important process for weather and climate. Accurate radiative transfer models exist, such as the RRTM, but these models are computationally slow. We develop neural networks (NNs), a type of machine learning model that is often computationally fast after training, to mimic the RRTM. We wish to accelerate the RRTM by orders of magnitude without sacrificing much accuracy. We drive both the NNs and RRTM with data from the GFSv16, an operational weather model, using locations around the globe during all seasons. We show that the NNs are highly accurate and much faster than the RRTM, which suggests that the NNs could be used to solve radiative transfer inside the GFSv16.
- NSF-PAR ID:
- 10473583
- Publisher / Repository:
- American Meteorological Society
- Date Published:
- Journal Name:
- Journal of Atmospheric and Oceanic Technology
- Volume:
- 40
- Issue:
- 11
- ISSN:
- 0739-0572
- Format(s):
- Medium: X Size: p. 1407-1432
- Size(s):
- ["p. 1407-1432"]
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract. Mineral dust aerosols cool and warm the atmosphere byscattering and absorbing solar (shortwave: SW) and thermal (longwave: LW)radiation. However, significant uncertainties remain in dust radiativeeffects, largely due to differences in the dust size distribution andspectral optical properties simulated in Earth system models. Dust modelstypically underestimate the coarse dust load (more than 2.5 µm indiameter) and assume a spherical shape, which leads to an overestimate ofthe fine dust load (less than 2.5 µm) after the dust emissions in themodels are scaled to match observed dust aerosol optical depth at 550 nm(DAOD550). Here, we improve the simulated dust properties with data setsthat leverage measurements of size-resolved dust concentration, asphericityfactor, and refractive index in a coupled global chemical transport modelwith a radiative transfer module. After the adjustment of size-resolved dustconcentration and spectral optical properties, the global and annual averageof DAOD550 from the simulation increases from 0.023 to 0.029 and fallswithin the range of a semi-observationally based estimate (0.030 ± 0.005). The reduction of fine dust load after the adjustment leads to areduction of the SW cooling at the top of the atmosphere (TOA). To improveagreement against a semi-observationally based estimate of the radiativeeffect efficiency at TOA, we find that a less absorptive SW dust refractiveindex is required for coarser aspherical dust. Thus, only a minor differenceis estimated for the net global dust radiative effect at TOA (−0.08 vs.−0.00 W m−2 on a global scale). Conversely, our sensitivitysimulations reveal that the surface warming is substantially enhanced nearthe strong dust source regions (less cooling to −0.23 from −0.60 W m−2 on a global scale). Thus, less atmospheric radiativeheating is estimated near the major source regions (less heating to 0.15from 0.59 W m−2 on a global scale), because of enhanced LWwarming at the surface by the synergy of coarser size and aspherical shape.more » « less
-
more » « less
Abstract Most climate models in phase 6 of the Coupled Model Intercomparison Project (CMIP6) still suffer pronounced warm and dry summer biases in the central United States (CUS), even in high-resolution simulations. We found that the cloud base definition in the cumulus parameterization was the dominant factor determining the spread of the biases among models and those defining cloud base at the lifting condensation level (LCL) performed the best. To identify the underlying mechanisms, we developed a physically based analytical bias model (ABM) to capture the key feedback processes of land–atmosphere coupling. The ABM has significant explanatory power, capturing 80% variance of temperature and precipitation biases among all models. Our ABM analysis via counterfactual experiments indicated that the biases are attributed mostly by surface downwelling longwave radiation errors and second by surface net shortwave radiation errors, with the former 2–5 times larger. The effective radiative forcing from these two errors as weighted by their relative contributions induces runaway temperature and precipitation feedbacks, which collaborate to cause CUS summer warm and dry biases. The LCL cumulus reduces the biases through two key mechanisms: it produces more clouds and less precipitable water, which reduce radiative energy input for both surface heating and evapotranspiration to cause a cooler and wetter soil; it produces more rainfall and wetter soil conditions, which suppress the positive evapotranspiration–precipitation feedback to damp the warm and dry bias coupling. Most models using non-LCL schemes underestimate both precipitation and cloud amounts, which amplify the positive feedback to cause significant biases.
-
more » « less
Abstract An advanced aerosol treatment, with a focus on semivolatile nitrate formation, is introduced into the Community Atmosphere Model version 5 with interactive chemistry (CAM5‐chem) by coupling the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) with the 7‐mode Modal Aerosol Module (MAM7). An important feature of MOSAIC is dynamic partitioning of all condensable gases to the different fine and coarse mode aerosols, as governed by mode‐resolved thermodynamics and heterogeneous chemical reactions. Applied in the free‐running mode from 1995 to 2005 with prescribed historical climatological conditions, the model simulates global distributions of sulfate, nitrate, and ammonium in good agreement with observations and previous studies. Inclusion of nitrate resulted in ∼10% higher global average accumulation mode number concentrations, indicating enhanced growth of Aitken mode aerosols from nitrate formation. While the simulated accumulation mode nitrate burdens are high over the anthropogenic source regions, the sea‐salt and dust modes respectively constitute about 74% and 17% of the annual global average nitrate burden. Regional clear‐sky shortwave radiative cooling of up to −5 W m−2due to nitrate is seen, with a much smaller global average cooling of −0.05 W m−2. Significant enhancements in regional cloud condensation nuclei (at 0.1% supersaturation) and cloud droplet number concentrations are also attributed to nitrate, causing an additional global average shortwave cooling of −0.8 W m−2. Taking into consideration of changes in both longwave and shortwave radiation under all‐sky conditions, the net change in the top of the atmosphere radiative fluxes induced by including nitrate aerosol is −0.7 W m−2.
-
Accurate multidecadal radiative flux records are vital to understand Arctic amplification and constrain climate model uncertainties. Uncertainty in the NASA Clouds and the Earth’s Radiant Energy System (CERES)-derived irradiances is larger over sea ice than any other surface type and comes from several sources. The year-long Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) expedition in the central Arctic provides a rare opportunity to explore uncertainty in CERES-derived radiative fluxes. First, a systematic and statistically robust assessment of surface shortwave and longwave fluxes was conducted using in situ measurements from MOSAiC flux stations. The CERES Synoptic 1degree (SYN1deg) product overestimates the downwelling shortwave flux by +11.40 Wm–2 and underestimates the upwelling shortwave flux by –15.70 Wm–2 and downwelling longwave fluxes by –12.58 Wm–2 at the surface during summer. In addition, large differences are found in the upwelling longwave flux when the surface approaches the melting point (approximately 0°C). The biases in downwelling shortwave and longwave fluxes suggest that the atmosphere represented in CERES is too optically thin. The large negative bias in upwelling shortwave flux can be attributed in large part to lower surface albedo (–0.15) in satellite footprint relative to surface sensors. Additionally, the results show that the spectral surface albedo used in SYN1deg overestimates albedo in visible and mid-infrared bands. A series of radiative transfer model perturbation experiments are performed to quantify the factors contributing to the differences. The CERES-MOSAiC broadband albedo differences (approximately 20 Wm–2) explain a larger portion of the upwelling shortwave flux difference than the spectral albedo shape differences (approximately 3 Wm–2). In addition, the differences between perturbation experiments using hourly and monthly MOSAiC surface albedo suggest that approximately 25% of the sea ice surface albedo variability is explained by factors not correlated with daily sea ice concentration variability. Biases in net shortwave and longwave flux can be reduced to less than half by adjusting both albedo and cloud inputs toward observed values. The results indicate that improvements in the surface albedo and cloud data would substantially reduce the uncertainty in the Arctic surface radiation budget derived from CERES data products.more » « less
