Abstract

The identification of atmospheric rivers (ARs) is crucial for weather and climate predictions as they are often associated with severe storm systems and extreme precipitation, which can cause large impacts on society. This study presents a deep learning model, termed ARDetect, for image segmentation of ARs using ERA5 data from 1960 to 2020 with labels obtained from the TempestExtremes tracking algorithm. ARDetect is a convolutional neural network (CNN)-based U-Net model, with its structure having been optimized using automatic hyperparameter tuning. Inputs to ARDetect were selected to be the integrated water vapor transport (IVT) and total column water (TCW) fields, as well as the AR mask from TempestExtremes from the previous time step to the one being considered. ARDetect achieved a mean intersection-over-union (mIoU) rate of 89.04% for ARs, indicating its high accuracy in identifying these weather patterns and a superior performance than most deep learning–based models for AR detection. In addition, ARDetect can be executed faster than the TempestExtremes method (seconds vs minutes) for the same period. This provides a significant benefit for online AR detection, especially for high-resolution global models. An ensemble of 10 models, each trained on the same dataset but having different starting weights, was used to further improve on the performance produced by ARDetect, thus demonstrating the importance of model diversity in improving performance. ARDetect provides an effective and fast deep learning–based model for researchers and weather forecasters to better detect and understand ARs, which have significant impacts on weather-related events such as floods and droughts.

Abstract

The present study aims to identify the precipitation bias associated with the interactions among fast physical processes in the Community Atmospheric Model, version 5 (CAM5), during the abrupt onset of the South China Sea (SCS) summer monsoon, a key precursor of the overall East Asia summer monsoon (EASM). The multiyear hindcast approach is utilized to obtain the well-constrained synoptic-scale horizontal circulation each year during the onset period from the years 1998 to 2012. In the pre-onset period, the ocean precipitation over the SCS is insufficiently suppressed in CAM5 hindcasts and thus weaker land–ocean precipitation contrasts. This is associated with the weaker and shallower convection simulated over the surrounding land, producing weaker local circulation within the SCS basin. In the post-onset period, rainfall of the organized convection over the Philippine coastal ocean is underestimated in the hindcasts, with overestimated upper-level heating. These biases are further elaborated as the underrepresentation of the convection diurnal cycle and coastal convection systems, as well as the issue of precipitation sensitivity to environmental moisture during the SCS onset period. The biases identified in hindcasts are consistent with the general bias of the EASM in the climate simulation of CAM5. The current results highlight that the appropriate representation of land–ocean–convection interactions over coastal areas can potentially improve the simulation of seasonal transition over the monsoon regions.

Abstract

This study examines the sensitivity of the global climate to land surface processes (LSP) using an atmospheric general circulation model both uncoupled (with prescribed SSTs) and coupled to an oceanic general circulation model. The emphasis is on the interactive soil moisture and vegetation biophysical processes, which have first-order influence on the surface energy and water budgets. The sensitivity to those processes is represented by the differences between model simulations, in which two land surface schemes are considered: 1) a simple land scheme that specifies surface albedo and soil moisture availability and 2) the Simplified Simple Biosphere Model (SSiB), which allows for consideration of interactive soil moisture and vegetation biophysical process. Observational datasets are also employed to assess the extent to which results are realistic.

The mean state sensitivity to different LSP is stronger in the coupled mode, especially in the tropical Pacific. Furthermore, the seasonal cycle of SSTs in the equatorial Pacific, as well as the ENSO frequency, amplitude, and locking to the seasonal cycle of SSTs, is significantly modified and more realistic with SSiB. This outstanding sensitivity of the atmosphere–ocean system develops through changes in the intensity of equatorial Pacific trades modified by convection over land. The results further demonstrate that the direct impact of land–atmosphere interactions on the tropical climate is modified by feedbacks associated with perturbed oceanic conditions (“indirect effect” of LSP). The magnitude of such an indirect effect is strong enough to suggest that comprehensive studies on the importance of LSP on the global climate have to be made in a system that allows for atmosphere–ocean interactions.

Abstract

The correspondence between short- and long-time-scale systematic errors in the Community Atmospheric Model, version 4 (CAM4) and version 5 (CAM5), is systematically examined. The analysis is based on the annual-mean data constructed from long-term “free running” simulations and short-range hindcasts. The hindcasts are initialized every day with the ECMWF analysis for the Year(s) of Tropical Convection. It has been found that most systematic errors, particularly those associated with moist processes, are apparent in day 2 hindcasts. These errors steadily grow with the hindcast lead time and typically saturate after five days with amplitudes comparable to the climate errors. Examples include the excessive precipitation in much of the tropics and the overestimate of net shortwave absorbed radiation in the stratocumulus cloud decks over the eastern subtropical oceans and the Southern Ocean at about 60°S. This suggests that these errors are likely the result of model parameterization errors as the large-scale flow remains close to observed in the first few days of the hindcasts. In contrast, other climate errors are present in the hindcasts, but with amplitudes that are significantly smaller than and do not approach their climate errors during the 6-day hindcasts. These include the cold biases in the lower stratosphere, the unrealistic double–intertropical convergence zone pattern in the simulated precipitation, and an annular mode bias in extratropical sea level pressure. This indicates that these biases could be related to slower processes such as radiative and chemical processes, which are important in the lower stratosphere, or the result of poor interactions of the parameterized physics with the large-scale flow.

Abstract

An evaluation is presented of the impact on tropical climate of continental-scale perturbations given by different representations of land surface processes (LSPs) in a general circulation model that includes atmosphere–ocean interactions. One representation is a simple land scheme, which specifies climatological albedos and soil moisture availability. The other representation is the more comprehensive Simplified Simple Biosphere Model, which allows for interactive soil moisture and vegetation biophysical processes.

The results demonstrate that such perturbations have strong impacts on the seasonal mean states and seasonal cycles of global precipitation, clouds, and surface air temperature. The impact is especially significant over the tropical Pacific Ocean. To explore the mechanisms for such impact, model experiments are performed with different LSP representations confined to selected continental-scale regions where strong interactions of climate–vegetation biophysical processes are present. The largest impact found over the tropical Pacific is mainly from perturbations in the tropical African continent where convective heating anomalies associated with perturbed surface heat fluxes trigger global teleconnections through equatorial wave dynamics. In the equatorial Pacific, the remote impacts of the convection anomalies are further enhanced by strong air–sea coupling between surface wind stress and upwelling, as well as by the effects of ocean memory. LSP perturbations over South America and Asia–Australia have much weaker global impacts. The results further suggest that correct representations of LSP, land use change, and associated changes in the deep convection over tropical Africa are crucial to reducing the uncertainty of future climate projections with global climate models under various climate change scenarios.

Abstract

Using the 9-yr warm-season observations at the Atmospheric Radiation Measurement Southern Great Plains site, we assess the land–atmosphere (LA) coupling in the North American Regional Reanalysis (NARR) and two climate models: hindcasts with the Community Atmosphere Model version 5.1 by Cloud-Associated Parameterizations Testbed (CAM5-CAPT) and nudged runs with the Energy Exascale Earth System Model Atmosphere Model version 1 Regionally Refined Model (EAMv1-RRM). We focus on three local convective regimes and diagnose model behaviors using the local coupling metrics. NARR agrees well with observations except a slightly warmer and drier surface with higher downwelling shortwave radiation and lower evaporative fraction. On clear-sky days, it shows warmer and drier early-morning conditions in both models with significant underestimates in surface evaporation by EAMv1-RRM. On the majority of the ARM-observed shallow cumulus days, there is no or little low-level clouds in either model. When captured in models, the simulated shallow cumulus shows much less cloud fraction and lower cloud bases than observed. On the days with late-afternoon deep convection, models tend to present a stable early-morning lower atmosphere more frequently than the observations, suggesting that the deep convection is triggered more often by elevated instabilities. Generally, CAM5-CAPT can reproduce the local LA coupling processes to some extent due to the constrained early-morning conditions and large-scale winds. EAMv1-RRM exhibits large precipitation deficits and warm and dry biases toward mid-to-late summers, which may be an amplification through a positive LA feedback among initial atmosphere and land states, convection triggering and large-scale circulations.

Abstract

An ensemble seasonal hindcast approach is used to investigate the development of the equatorial Pacific Ocean cold sea surface temperature (SST) bias and its characteristic annual cycle in the Community Earth System Model, version 1 (CESM1). In observations, eastern equatorial Pacific SSTs exhibit a warm phase during boreal spring and a cold phase during late boreal summer–autumn. The CESM1 climatology shows a cold bias during both warm and cold phases. In our hindcasts, the cold bias during the cold phase develops in less than 6 months, whereas the cold bias during the warm phase takes longer to emerge. The fast-developing cold-phase cold bias is associated with too-strong vertical advection and easterly wind stress over the eastern equatorial region. The antecedent boreal summer easterly wind anomalies also appear in atmosphere-only simulations, indicating that the errors are intrinsic to the atmosphere component. For the slower-developing warm-phase cold bias, we find that the too-cold SSTs over the equatorial region are associated with a slowly evolving upward displacement of subsurface ocean zonal currents and isotherms that can be traced to the ocean component.

Abstract

The correspondence between mean sea surface temperature (SST) biases in retrospective seasonal forecasts (hindcasts) and long-term climate simulations from five global climate models is examined to diagnose the degree to which systematic SST biases develop on seasonal time scales. The hindcasts are from the North American Multimodel Ensemble, and the climate simulations are from the Coupled Model Intercomparison Project. The analysis suggests that most robust climatological SST biases begin to form within 6 months of a realistically initialized integration, although the growth rate varies with location, time, and model. In regions with large biases, interannual variability and ensemble spread is much smaller than the climatological bias. Additional ensemble hindcasts of the Community Earth System Model with a different initialization method suggest that initial conditions do matter for the initial bias growth, but the overall global bias patterns are similar after 6 months. A hindcast approach is more suitable to study biases over the tropics and subtropics than over the extratropics because of smaller initial biases and faster bias growth. The rapid emergence of SST biases makes it likely that fast processes with time scales shorter than the seasonal time scales in the atmosphere and upper ocean are responsible for a substantial part of the climatological SST biases. Studying the growth of biases may provide important clues to the causes and ultimately the amelioration of these biases. Further, initialized seasonal hindcasts can profitably be used in the development of high-resolution coupled ocean–atmosphere models.