Carolyn A. Reynolds’s research while affiliated with University School and other places

Located in the western North Atlantic, the subsurface temperature and salinity structure at the Atlantis II Seamounts is sensitive to fluctuations in the Gulf Stream, and historical ocean structure profiles at this location show a bimodal distribution, making seasonal climatologies a poor predictor of ocean states. Historical CMIP6 ensemble means show dynamic changes in temperature between 1980-2007 (0.5 °C), with no corresponding change in sea surface height anomalies, suggestive of decadal changes independent of Gulf Stream position. While previous Deep Learning models have been able to predict SST globally, the Gulf Stream region has some of the lowest accuracy and shortest forecast horizons. We use a deep learning (convolutional neural network) model to predict sea surface height anomalies and sea surface temperature for one month to two years, focusing on the Atlantis II Seamounts. The training dataset consists of the global 1850-2007 CMIP6 temperature, height, and mixed layer depth that capture climate modes of variability that impact the Gulf Stream. We test our model predictions with ORAS5 reanalysis data. Our model can beat the climatology for up to 24 months by predicting different decadal regimes in the Gulf Stream. Using back-propagation saliency maps, we connect Atlantis II Seamounts predictions to multiple global modes of climate variability, such as the North Atlantic Oscillation, the Indian Ocean Dipole, the Pacific Decadal Oscillation, and ENSO.

Electromagnetic ducting, a phenomenon where atmospheric conditions create a layer that traps and channels electromagnetic waves, is crucial for radio wave propagation in communication, radar, and navigation systems. This study analyzes dropsonde data from the Atmospheric River Reconnaissance flights and finds that the sector of atmospheric rivers (ARs) plays a significant role in determining the frequency and properties of electromagnetic ducts, particularly elevated ducts. Elevated ducting is two to three times more frequent in the AR warm sector and non-AR warm sector than in the AR core or cold sector?. The strongest and deepest ducts are found in the non-AR warm sector, which is consistent with the typical synoptic setting of ARs. Most ducts are within 500 m of the marine boundary layer height, but some can be significantly above or below. The relationships between duct strength, depth, and maximum trapped wavelength are invariant across different AR sectors, in contrast to other studies in non-AR environments. These results suggest that large changes in airmass density lead to deeper ducts that can trap larger wavelengths of electromagnetic radiation.

We examine the environmental conditions that lead to well- and poorly-predicted MJO events in the Navy Earth System Prediction Capability (ESPC) global coupled forecast system. Individual MJO events are tracked using an MJO tracking algorithm following Chikira (2014). Good and poor forecasts are determined by how well the forecasted MJO-object matches with the observed MJO-objects. The primary difference between good and poor MJO forecasts is location and timing of forecasted MJO events. Good MJO forecasts capture the evolution of observed environmental moisture and low-level winds anomalies, while poor MJO forecasts do not build sufficient moisture anomalies to support the MJO’s amplifications and propagation. The poor forecasts struggle to simulate the horizontal advection and evaporation. The errors in the evaporation are likely driven by errors in the environmental and MJO-scale zonal wind anomalies. The errors in the evolution of the horizontal advection are largely driven by the evolution of the environmental wind and moisture gradient, which induces errors in the evolution of the zonal moisture advection. The effect of the good and poor MJO forecasts on the extended range MJO-teleconnections are examined. It is found that MJO-teleconnections are best simulated following the MJO’s enhanced convection over the Maritime Continent. There are systematic biases in the MJO-induced Rossby wave trains in the Navy ESPC, in some cases despite a good representation of the MJO. A northeastward tilt in the subtropical jet exit region is identified in the Navy ESPC and is another possible cause of these biases in the Rossby wave trains.

Artificial Intelligence (AI) weather models are explored for initial condition sensitivity studies to analyze the physicality of the relationships learned. Gradients (or sensitivities) of the target metric of interest are computed with respect to the variable fields at initial time by means of the backpropagation algorithm, which does not assume linear perturbation growth. Here, sensitivities from an AI model at 36-h lead time were compared to those produced by an adjoint of a dynamical model for an extreme weather event, cyclone Xynthia, presenting very similar structures and with the evolved perturbations leading to similar impacts. This demonstrates the ability of the AI weather model to learn physically meaningful spatio-temporal links between atmospheric processes. These findings should enable researchers to conduct initial condition studies in minutes, potentially at lead times into the non-linear regime (typically >5 days), with important applications in observing network design and the study of atmospheric dynamics.

Artificial Intelligence (AI) weather models are explored for initial condition sensitivity studies to analyze the physicality of the relationships learned. Gradients (or sensitivities) of the target metric of interest are computed with respect to the variable fields at initial time by means of the backpropagation algorithm and gradient descent, which do not assume linear perturbation growth. Here, sensitivities from an AI model at 36-hour lead time were compared to those produced by an adjoint of a dynamical model for an extreme weather event, cyclone Xynthia, presenting very similar structures and with the evolved perturbations leading to similar impacts. This demonstrates the ability of the AI model to learn physically-meaningful spatio-temporal links between atmospheric processes. These findings should enable researchers to conduct initial condition studies in minutes, potentially at lead times into the non-linear regime (typically > 5 days), with important applications in observing network design and the study of atmospheric dynamics.

Accurately simulating the Madden-Julian Oscillation (MJO), which dominates intraseasonal (30-90 day) variability in the tropics, is critical to predicting tropical cyclones (TCs) and other phenomena at extended-range (2-3 week) timescales. MJO biases in intensity and propagation speed are a common problem in global coupled models. For example, the MJO in the Navy Earth System Prediction Capability (ESPC), a global coupled model, has been shown to be too strong and too fast, which has implications for the MJO-TC relationship in that model. The biases and extended-range prediction skill in the operational version of the Navy ESPC are compared to experiments applying different versions of Analysis Correction-based Additive Inflation (ACAI) to reduce model biases. ACAI is a method in which time-mean and stochastic perturbations based on analysis increments are added to the model tendencies with the goals of reducing systematic error and accounting for model uncertainty. Over the extended boreal summer (May-November), ACAI reduces the root mean squared error (RMSE) and improves the spread-skill relationship of the total tropical and MJO-filtered OLR and low-level zonal winds. While ACAI improves skill in the environmental fields of low-level absolute vorticity, potential intensity, and vertical wind shear, it degrades the skill in the relative humidity, which increases the positive bias in the Genesis Potential Index (GPI) in the operational Navy ESPC. Northern Hemisphere integrated TC genesis biases are reduced (increased number of TCs) in the ACAI experiments, which is consistent with the positive GPI bias in the ACAI simulations.

In recent years, studies have put forth various theories on the role of oceanic equatorial Rossby waves (OERW) in the subseasonal‐to‐seasonal (S2S) predictability of the Indian Ocean (IO). While much of the scientific literature uses data from in‐situ, satellite, and/or reanalysis datasets, this study focuses on reforecast fields from the European Centre for Medium‐Range Weather Forecasting's (ECMWF) S2S dataset. Evaluation of the model's predictive skill in representing OERWs and the associated variations in subsurface‐to‐surface interaction and air–sea coupling are discussed. This work provides a unique methodology to calculate and evaluate the predictability of OERWs from model forecast data. Our results indicate that the model forecasts OERWs with high skill (anomaly correlation > 0.8 out to 40 days), indicating they are a key source of oceanic subseasonal predictability at extended lead times. Analysis of the wavenumber–frequency spectra for the IO indicates a strong reduction in power throughout the model forecast time period in the oceanic equatorial Kelvin wave (OEKW) regime and modest reduction of the OERW power. Both Kelvin and Rossby waves are modulated by the subseasonal zonal wind stress anomalies and the reduction of power is impacted by biases in winds at longer forecast leads. The erroneous weakening of the OEKWs contributes to the weakening of the reflected OERWs. Previous studies have documented that ocean heat content (OHC), particularly associated with downwelling OERWs, is important to maintaining and amplifying subseasonal precipitation in the IO. The reduced OERW power results in weaker advection of enhanced OHC anomalies by the OERWs, which has numerous implications for air–sea and subsurface‐to‐surface coupling, as discussed. The atmospheric response to the waning westward transport of OHC anomalies in the western IO by OERWs is associated with a weakening of intraseasonal precipitation anomalies associated with the intraseasonal oscillation.

The Makassar Peninsula in southwestern Sulawesi, Indonesia, experienced its largest flood in its recorded history in January 2019. Four-day accumulated rainfall exceeded 350 mm with devastating impacts on the community, including 53 perished and over 14 thousand evacuated. Previous studies find a convectively coupled Kelvin wave and convectively coupled equatorial Rossby wave associated with the Madden-Julian Oscillation to be likely contributors to the onset of the mesoscale convective system responsible for the flooding. We employ an adjoint model to identify and dynamically link specific components of the mesoscale and environmental flow affecting the flooding event. The adjoint simulations indicate that enhancing the moisture and low-level convergence associated with the mesoscale convective system can substantially increase rainfall. The sensitivity patterns are complex, with low-level convergence and vorticity sensitivity in quadrature and projecting onto the larger-scale Kelvin and Rossby waves. The vorticity sensitivity enhances waves along the dynamic equator. Small adjoint-based perturbations made to the initial state can increase the 36-h rainfall maximum by greater than 30%. The sensitivity analysis supports the importance of a mesoscale convective system, orographic ascent, and equatorial wave components in contributing to the flood. The rapid growth of small initial perturbations underscores the need for probabilistic forecasts.

6th WGNE workshop on systematic errors in weather and climate models What: Scientists, ranging from early career to highly experienced, involved in the development of weather and climate models and in the diagnosis of model errors, held an international workshop to discuss the nature, causes and remedies of systematic errors across timescales and across Earth system modeling components. When: 31 Oct - 04 Nov 2022 Where: Reading, UK and online