skip to main content


Title: Statistical constraints on climate model parameters using a scalable cloud-based inference framework
Abstract Atmospheric aerosols influence the Earth’s climate, primarily by affecting cloud formation and scattering visible radiation. However, aerosol-related physical processes in climate simulations are highly uncertain. Constraining these processes could help improve model-based climate predictions. We propose a scalable statistical framework for constraining the parameters of expensive climate models by comparing model outputs with observations. Using the C3.AI Suite, a cloud computing platform, we use a perturbed parameter ensemble of the UKESM1 climate model to efficiently train a surrogate model. A method for estimating a data-driven model discrepancy term is described. The strict bounds method is applied to quantify parametric uncertainty in a principled way. We demonstrate the scalability of this framework with 2 weeks’ worth of simulated aerosol optical depth data over the South Atlantic and Central African region, written from the model every 3 hr and matched in time to twice-daily MODIS satellite observations. When constraining the model using real satellite observations, we establish constraints on combinations of two model parameters using much higher time-resolution outputs from the climate model than previous studies. This result suggests that within the limits imposed by an imperfect climate model, potentially very powerful constraints may be achieved when our framework is scaled to the analysis of more observations and for longer time periods.  more » « less
Award ID(s):
2053804
NSF-PAR ID:
10430515
Author(s) / Creator(s):
; ; ; ; ; ; ;
Date Published:
Journal Name:
Environmental Data Science
Volume:
2
ISSN:
2634-4602
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract. The role of clouds in the Arctic radiation budget is not well understood. Ground-based and airborne measurements provide valuable data to test and improve our understanding. However, the ground-based measurements are intrinsically sparse, and the airborne observations are snapshots in time and space. Passive remote sensing measurements from satellite sensors offer high spatial coverage and an evolving time series, having lengths potentially of decades. However, detecting clouds by passive satellite remote sensing sensors is challenging over the Arctic because of the brightness of snow and ice in the ultraviolet and visible spectral regions and because of the small brightness temperature contrast to the surface. Consequently, the quality of the resulting cloud data products needs to be assessed quantitatively. In this study, we validate the cloud data products retrieved from the Advanced Very High Resolution Radiometer (AVHRR) post meridiem (PM) data from the polar-orbiting NOAA-19 satellite and compare them with those derived from the ground-based instruments during the sunlit months. The AVHRR cloud data products by the European Space Agency (ESA) Cloud Climate Change Initiative (Cloud_CCI) project uses the observations in the visible and IR bands to determine cloud properties. The ground-based measurements from four high-latitude sites have been selected for this investigation: Hyytiälä (61.84∘ N, 24.29∘ E), North Slope of Alaska (NSA; 71.32∘ N, 156.61∘ W), Ny-Ålesund (Ny-Å; 78.92∘ N, 11.93∘ E), and Summit (72.59∘ N, 38.42∘ W). The liquid water path (LWP) ground-based data are retrieved from microwave radiometers, while the cloud top height (CTH) has been determined from the integrated lidar–radar measurements. The quality of the satellite products, cloud mask and cloud optical depth (COD), has been assessed using data from NSA, whereas LWP and CTH have been investigated over Hyytiälä, NSA, Ny-Å, and Summit. The Cloud_CCI COD results for liquid water clouds are in better agreement with the NSA radiometer data than those for ice clouds. For liquid water clouds, the Cloud_CCI COD is underestimated roughly by 3 optical depth (OD) units. When ice clouds are included, the underestimation increases to about 5 OD units. The Cloud_CCI LWP is overestimated over Hyytiälä by ≈7 g m−2, over NSA by ≈16 g m−2, and over Ny-Å by ≈24 g m−2. Over Summit, CCI LWP is overestimated for values ≤20 g m−2 and underestimated for values >20 g m−2. Overall the results of the CCI LWP retrievals are within the ground-based instrument uncertainties. To understand the effects of multi-layer clouds on the CTH retrievals, the statistics are compared between the single-layer clouds and all types (single-layer + multi-layer). For CTH retrievals, the Cloud_CCI product overestimates the CTH for single-layer clouds. When the multi-layer clouds are included (i.e., all types), the observed CTH overestimation becomes an underestimation of about 360–420 m. The CTH results over Summit station showed the highest biases compared to the other three sites. To understand the scale-dependent differences between the satellite and ground-based data, the Bland–Altman method is applied. This method does not identify any scale-dependent differences for all the selected cloud parameters except for the retrievals over the Summit station. In summary, the Cloud_CCI cloud data products investigated agree reasonably well with those retrieved from ground-based measurements made at the four high-latitude sites.

     
    more » « less
  2. Abstract. Due to its remote location and extreme weather conditions, atmospheric in situmeasurements are rare in the Southern Ocean. As a result, aerosol–cloudinteractions in this region are poorly understood and remain a major source ofuncertainty in climate models. This, in turn, contributes substantially topersistent biases in climate model simulations such as the well-known positiveshortwave radiation bias at the surface, as well as biases in numericalweather prediction models and reanalyses. It has been shown in previousstudies that in situ and ground-based remote sensing measurements across theSouthern Ocean are critical for complementing satellite data sets due to theimportance of boundary layer and low-level cloud processes. These processesare poorly sampled by satellite-based measurements and are often obscured bymultiple overlying cloud layers. Satellite measurements also do not constrainthe aerosol–cloud processes very well with imprecise estimation of cloudcondensation nuclei. In this work, we present a comprehensive set of ship-basedaerosol and meteorological observations collected on the 6-weekSouthern Ocean Ross Sea Marine Ecosystem and Environment voyage(TAN1802) voyage of RV Tangaroa across the Southern Ocean, from Wellington, New Zealand, tothe Ross Sea, Antarctica. The voyage was carried out from 8 February to21 March 2018. Many distinct, but contemporaneous, data sets were collectedthroughout the voyage. The compiled data sets include measurements from arange of instruments, such as (i) meteorological conditions at the sea surfaceand profile measurements; (ii) the size and concentration of particles; (iii)trace gases dissolved in the ocean surface such as dimethyl sulfide andcarbonyl sulfide; (iv) and remotely sensed observations of low clouds. Here,we describe the voyage, the instruments, and data processing, and provide a briefoverview of some of the data products available. We encourage the scientificcommunity to use these measurements for further analysis and model evaluationstudies, in particular, for studies of Southern Ocean clouds, aerosol, andtheir interaction. The data sets presented in this study are publiclyavailable at https://doi.org/10.5281/zenodo.4060237 (Kremser et al., 2020). 
    more » « less
  3. Abstract. One of the challenges inrepresenting warm rain processes in global climate models (GCMs) is relatedto the representation of the subgrid variability of cloud properties, such ascloud water and cloud droplet number concentration (CDNC), and the effectthereof on individual precipitation processes such as autoconversion. Thiseffect is conventionally treated by multiplying the resolved-scale warm rainprocess rates by an enhancement factor (Eq) which is derived fromintegrating over an assumed subgrid cloud water distribution. The assumedsubgrid cloud distribution remains highly uncertain. In this study, we derivethe subgrid variations of liquid-phase cloud properties over the tropicalocean using the satellite remote sensing products from Moderate ResolutionImaging Spectroradiometer (MODIS) and investigate the correspondingenhancement factors for the GCM parameterization of autoconversion rate. Wefind that the conventional approach of using only subgrid variability ofcloud water is insufficient and that the subgrid variability of CDNC, as wellas the correlation between the two, is also important for correctlysimulating the autoconversion process in GCMs. Using the MODIS data whichhave near-global data coverage, we find that Eq shows a strongdependence on cloud regimes due to the fact that the subgrid variability ofcloud water and CDNC is regime dependent. Our analysis shows a significantincrease of Eq from the stratocumulus (Sc) to cumulus (Cu) regions.Furthermore, the enhancement factor EN due to the subgrid variation ofCDNC is derived from satellite observation for the first time, and resultsreveal several regions downwind of biomass burning aerosols (e.g., Gulf ofGuinea, east coast of South Africa), air pollution (i.e., East China Sea),and active volcanos (e.g., Kilauea, Hawaii, and Ambae, Vanuatu), where theEN is comparable to or even larger than Eq, suggesting an importantrole of aerosol in influencing the EN. MODIS observations suggest thatthe subgrid variations of cloud liquid water path (LWP) and CDNC aregenerally positively correlated. As a result, the combined enhancementfactor, including the effect of LWP and CDNC correlation, is significantlysmaller than the simple product of EqEN. Given the importanceof warm rain processes in understanding the Earth's system dynamics and watercycle, we conclude that more observational studies are needed to provide abetter constraint on the warm rain processes in GCMs.

     
    more » « less
  4. null (Ed.)
    Deriving aerosol optical depth (AOD) from space-borne observations is still challenging due to uncertainties associated with sensor calibration drift, cloud screening, aerosol type classification, and surface reflectance characterization. As an initial step to understanding the physical processes impacting these uncertainties in satellite AOD retrievals, this study outlines a theoretical approach to estimate biases in the satellite aerosol retrieval algorithm affected by surface albedo and prescribed aerosol optical properties using a simplified radiative transfer model with a traditional error propagation approach. We expand the critical surface reflectance concept to obtain the critical surface albedo (CSA), critical single scattering albedo (CSSA), and critical asymmetry parameter (CAP). The top-of-atmosphere (TOA) reflectance is not sensitive to significant variability in aerosol loading (AOD) at the critical value; thus, the AOD cannot be determined. Results show that 5% bias in surface albedo (A), single scattering albedo (SSA), or asymmetry parameter (g) lead to large retrieved AOD errors, especially high under conditions when A, SSA, or g are close to their critical values. The results can be useful for future research related to improvements of satellite aerosol retrieval algorithms and provide a preliminary framework to analytically quantify AOD uncertainties from satellite retrievals. 
    more » « less
  5. The goal of generative models is to learn the intricate relations between the data to create new simulated data, but current approaches fail in very high dimensions. When the true data-generating process is based on physical processes, these impose symmetries and constraints, and the generative model can be created by learning an effective description of the underlying physics, which enables scaling of the generative model to very high dimensions. In this work, we propose Lagrangian deep learning (LDL) for this purpose, applying it to learn outputs of cosmological hydrodynamical simulations. The model uses layers of Lagrangian displacements of particles describing the observables to learn the effective physical laws. The displacements are modeled as the gradient of an effective potential, which explicitly satisfies the translational and rotational invariance. The total number of learned parameters is only of order 10, and they can be viewed as effective theory parameters. We combine N-body solver fast particle mesh (FastPM) with LDL and apply it to a wide range of cosmological outputs, from the dark matter to the stellar maps, gas density, and temperature. The computational cost of LDL is nearly four orders of magnitude lower than that of the full hydrodynamical simulations, yet it outperforms them at the same resolution. We achieve this with only of order 10 layers from the initial conditions to the final output, in contrast to typical cosmological simulations with thousands of time steps. This opens up the possibility of analyzing cosmological observations entirely within this framework, without the need for large dark-matter simulations.

     
    more » « less