RESUMEN

Transfer learning (TL), which enables neural networks (NNs) to generalize out-of-distribution via targeted re-training, is becoming a powerful tool in scientific machine learning (ML) applications such as weather/climate prediction and turbulence modeling. Effective TL requires knowing (1) how to re-train NNs? and (2) what physics are learned during TL? Here, we present novel analyses and a framework addressing (1)-(2) for a broad range of multi-scale, nonlinear, dynamical systems. Our approach combines spectral (e.g. Fourier) analyses of such systems with spectral analyses of convolutional NNs, revealing physical connections between the systems and what the NN learns (a combination of low-, high-, band-pass filters and Gabor filters). Integrating these analyses, we introduce a general framework that identifies the best re-training procedure for a given problem based on physics and NN theory. As test case, we explain the physics of TL in subgrid-scale modeling of several setups of 2D turbulence. Furthermore, these analyses show that in these cases, the shallowest convolution layers are the best to re-train, which is consistent with our physics-guided framework but is against the common wisdom guiding TL in the ML literature. Our work provides a new avenue for optimal and explainable TL, and a step toward fully explainable NNs, for wide-ranging applications in science and engineering, such as climate change modeling.

RESUMEN

Models of many engineering and natural systems are imperfect. The discrepancy between the mathematical representations of a true physical system and its imperfect model is called the model error. These model errors can lead to substantial differences between the numerical solutions of the model and the state of the system, particularly in those involving nonlinear, multi-scale phenomena. Thus, there is increasing interest in reducing model errors, particularly by leveraging the rapidly growing observational data to understand their physics and sources. Here, we introduce a framework named MEDIDA: Model Error Discovery with Interpretability and Data Assimilation. MEDIDA only requires a working numerical solver of the model and a small number of noise-free or noisy sporadic observations of the system. In MEDIDA, first, the model error is estimated from differences between the observed states and model-predicted states (the latter are obtained from a number of one-time-step numerical integrations from the previous observed states). If observations are noisy, a data assimilation technique, such as the ensemble Kalman filter, is employed to provide the analysis state of the system, which is then used to estimate the model error. Finally, an equation-discovery technique, here the relevance vector machine, a sparsity-promoting Bayesian method, is used to identify an interpretable, parsimonious, and closed-form representation of the model error. Using the chaotic Kuramoto-Sivashinsky system as the test case, we demonstrate the excellent performance of MEDIDA in discovering different types of structural/parametric model errors, representing different types of missing physics, using noise-free and noisy observations.

RESUMEN

The movement of tropical cyclones (TCs), particularly around the time of landfall, can substantially affect the resulting damage. Recently, trends in TC translation speed and the likelihood of stalled TCs such as Harvey have received significant attention, but findings have remained inconclusive. Here, we examine how the June-September steering wind and translation speed of landfalling Texas TCs change in the future under anthropogenic climate change. Using several large-ensemble/multi-model datasets, we find pronounced regional variations in the meridional steering wind response over North America, but-consistently across models-stronger June-September-averaged northward steering winds over Texas. A cluster analysis of daily wind patterns shows more frequent circulation regimes that steer landfalling TCs northward in the future. Downscaling experiments show a 10-percentage-point shift from the slow-moving to the fast-moving end of the translation-speed distribution in the future. Together, these analyses indicate increases in the likelihood of faster-moving landfalling Texas TCs in the late 21st century.

RESUMEN

Numerical weather prediction models require ever-growing computing time and resources but, still, have sometimes difficulties with predicting weather extremes. We introduce a data-driven framework that is based on analog forecasting (prediction using past similar patterns) and employs a novel deep learning pattern-recognition technique (capsule neural networks, CapsNets) and an impact-based autolabeling strategy. Using data from a large-ensemble fully coupled Earth system model, CapsNets are trained on midtropospheric large-scale circulation patterns (Z500) labeled 0-4 depending on the existence and geographical region of surface temperature extremes over North America several days ahead. The trained networks predict the occurrence/region of cold or heat waves, only using Z500, with accuracies (recalls) of 69-45% (77-48%) or 62-41% (73-47%) 1-5 days ahead. Using both surface temperature and Z500, accuracies (recalls) with CapsNets increase to ∼ 80% (88%). In both cases, CapsNets outperform simpler techniques such as convolutional neural networks and logistic regression, and their accuracy is least affected as the size of the training set is reduced. The results show the promises of multivariate data-driven frameworks for accurate and fast extreme weather predictions, which can potentially augment numerical weather prediction efforts in providing early warnings.

RESUMEN

Deep learning techniques such as convolutional neural networks (CNNs) can potentially provide powerful tools for classifying, identifying, and predicting patterns in climate and environmental data. However, because of the inherent complexities of such data, which are often spatio-temporal, chaotic, and non-stationary, the CNN algorithms must be designed/evaluated for each specific dataset and application. Yet CNN, being a supervised technique, requires a large labeled dataset to start. Labeling demands (human) expert time which, combined with the limited number of relevant examples in this area, can discourage using CNNs for new problems. To address these challenges, here we (1) Propose an effective auto-labeling strategy based on using an unsupervised clustering algorithm and evaluating the performance of CNNs in re-identifying and predicting these clusters up to 5 days ahead of time; (2) Use this approach to label thousands of daily large-scale weather patterns over North America in the outputs of a fully-coupled climate model and show the capabilities of CNNs in re-identifying and predicting the 4 clustered regimes up to 5 days ahead of time. The deep CNN trained with 1000 samples or more per cluster has an accuracy of 90% or better for both identification and prediction while prediction accuracy scales weakly with the number of lead days. Accuracy scales monotonically but nonlinearly with the size of the training set, e.g. reaching 94% with 3000 training samples per cluster for identification and 93-76% for prediction at lead day 1-5, outperforming logistic regression, a simpler machine learning algorithm, by  ~ 25%. Effects of architecture and hyperparameters on the performance of CNNs are examined and discussed.

RESUMEN

This study aimed to identify the differentiating parameters of the spinal curves' 2D projections through a hierarchical classification of the 3D spinal curve in adolescent idiopathic scoliosis (AIS). A total number of 103 right thoracic left lumbar pre-operative AIS patients were included retrospectively and consecutively. A total number of 20 non-scoliotic adolescents were included as the control group. All patients had biplanar X-rays and 3D reconstructions of the spine. The 3D spinal curve was calculated by interpolating the center of vertebrae and was isotropically normalized. A hierarchical classification of the normalized spinal curves was developed to group the patients based on the similarity of their 3D spinal curve. The spinal curves' 2D projections and clinical spinal measurements in the three anatomical planes were then statistically compared between these groups and between the scoliotic subtypes and the non-scoliotic controls. A total of 5 patient groups of right thoracic left lumbar AIS patients were identified. The characteristics of the posterior-anterior and sagittal views of the spines were: Type 1: Normal sagittal profile and S shape axial view. T1 is leveled or tilted to the right in the posterior view. Type 2: Hypokyphotic and a V shape axial view. T1 is tilted to the left in the posterior view. Type 3: Hypokyphotic (only T5-T10) and frontal imbalance, S shape axial view. T1 is leveled or tilted to the right, and 3 frontal curves. Type 4: Flat sagittal profile (T1-L2), slight frontal imbalance with a V shape axial view, T1 tilted to the left. Type 5: flat sagittal profile and forward trunk shift with a proximal kyphosis and S shape axial view. T1 is leveled or tilted to the right. In conclusion, a hierarchical classification of the 3D scoliotic spine allowed identifying various distinguishing features of the spinal curves in patients with a right thoracic curve in an orderly fashion. The subtypes' characteristics resulting from this 3D classification can be identified from the pairs of the frontal and sagittal spinal curves i.e. X-rays in right thoracic AIS patients.

Asunto(s)

Escoliosis/diagnóstico por imagen , Adolescente , Análisis por Conglomerados , Femenino , Humanos , Interpretación de Imagen Asistida por Computador , Imagenología Tridimensional , Vértebras Lumbares/diagnóstico por imagen , Vértebras Lumbares/patología , Masculino , Radiografía , Estudios Retrospectivos , Escoliosis/clasificación , Escoliosis/patología , Vértebras Torácicas/diagnóstico por imagen , Vértebras Torácicas/patología

RESUMEN

Climates across both hemispheres are strongly influenced by tropical Pacific variability associated with the El Niño/Southern Oscillation (ENSO). Conversely, extratropical variability also can affect the tropics. In particular, seasonal-mean alterations of near-surface winds associated with the North Pacific Oscillation (NPO) serve as a significant extratropical forcing agent of ENSO. However, it is still unclear what dynamical processes give rise to year-to-year shifts in these long-lived NPO anomalies. Here we show that intraseasonal variability in boreal winter pressure patterns over the Central North Pacific (CNP) imparts a significant signature upon the seasonal-mean circulations characteristic of the NPO. Further we show that the seasonal-mean signature results in part from year-to-year variations in persistent, quasi-stationary low-pressure intrusions into the subtropics of the CNP, accompanied by the establishment of persistent, quasi-stationary high-pressure anomalies over high latitudes of the CNP. Overall, we find that the frequency of these persistent extratropical anomalies (PEAs) during a given winter serves as a key modulator of intraseasonal variability in extratropical North Pacific circulations and, through their influence on the seasonal-mean circulations in and around the southern lobe of the NPO, the state of the equatorial Pacific 9-12 months later.

RESUMEN

A previously unknown instability creates space-filling lattices of 3D vortices in linearly stable, rotating, stratified shear flows. The instability starts from an easily excited critical layer. The layer intensifies by drawing energy from the background shear and rolls up into vortices that excite new critical layers and vortices. The vortices self-similarly replicate to create lattices of turbulent vortices. The vortices persist for all time. This self-replication occurs in stratified Couette flows and in the dead zones of protoplanetary disks where it can destabilize Keplerian flows.