Some dynamical aspects of meso-scale rainfall events as revealed by physical initialization

In recent years, physical initialization has emerged as a powerful tool to improve initial state of dynamical model during assimilation phase. This improved initial state at high resolution global spectral model is able to provide a tropical meso-scale coverage. In this paper, model out-put is used to study some dynamical aspects of meso-scale rainfall events. Major findings of this study are : (i) Meso-scale rainfall event carries a distinct dynamic structure in vertical profiles of divergence and vertical upward motion, (ii) Meso-scale event exhibits a large diurnal variation in these vertical profiles and (iii) Vertical motion field of meso-scale organisation appears to play a significant role in tropical storm formation.

Sensitivity of horizontal surface deformation to mantle dynamics from 3D instantaneous dynamics modeling of the eastern Mediterranean region

Geodetically estimated surface motions contain contributions to crustal deformation from coupled geodynamic processes active at all spatial scales and constitute key data for lithosphere dynamics research. Data interpretation methods should therefore account for the full range of possible processes, otherwise risking misinterpretation of data signal and incorrect estimation of lithosphere rheology, stress, or deformation fields. Here we explore the sensitivity of surface deformation to sub-lithospheric processes such as viscous plate-mantle and slab-mantle coupling, variations in slab pull, and buoyancy-driven mantle flow. To this end, we perform 3D instantaneous-dynamics numerical modelling of an elaborately structured compressible crust-mantle system designed for the Eastern Mediterranean Aegean-Anatolian region. We first determine a reference model driven by the absolute motions of the major plates, regional slab pull, a 3D mantle buoyancy field, and modulated by plate boundary coupling and mantle viscosity. The RMS motion data fit of ~5.9 mm/yr of predicted and observed Aegean-Anatolian horizontal surface motions demonstrates that the bulk amplitude of surface motion can be explained by these combined mantle processes. Next, by systematically perturbing reference model features, we assess the crustal sensitivity to each geodynamic driver and to mantle rheology. We find significant changes in crustal velocity gradient amplitudes, often between 10% and 40% of the reference model, with slab morphology effects of up to 93%. This demonstrates the key importance of carefully accounting for each process in modelling lithosphere dynamics. For the Aegean-Anatolia region, we present geodynamic evidence that the Aegean slab pull is the primary driver of the crustal motion field, as was previously suggested from kinematic analysis.

Audio2Head: Audio-driven One-shot Talking-head Generation with Natural Head Motion

We propose an audio-driven talking-head method to generate photo-realistic talking-head videos from a single reference image. In this work, we tackle two key challenges: (i) producing natural head motions that match speech prosody, and (ii)} maintaining the appearance of a speaker in a large head motion while stabilizing the non-face regions. We first design a head pose predictor by modeling rigid 6D head movements with a motion-aware recurrent neural network (RNN). In this way, the predicted head poses act as the low-frequency holistic movements of a talking head, thus allowing our latter network to focus on detailed facial movement generation. To depict the entire image motions arising from audio, we exploit a keypoint based dense motion field representation. Then, we develop a motion field generator to produce the dense motion fields from input audio, head poses, and a reference image. As this keypoint based representation models the motions of facial regions, head, and backgrounds integrally, our method can better constrain the spatial and temporal consistency of the generated videos. Finally, an image generation network is employed to render photo-realistic talking-head videos from the estimated keypoint based motion fields and the input reference image. Extensive experiments demonstrate that our method produces videos with plausible head motions, synchronized facial expressions, and stable backgrounds and outperforms the state-of-the-art.

Deep Gradient Prior Regularized Robust Video Super-Resolution

This paper proposes a robust multi-frame video super-resolution (SR) scheme to obtain high SR performance under large upscaling factors. Although the reference low-resolution frames can provide complementary information for the high-resolution frame, an effective regularizer is required to rectify the unreliable information from the reference frames. As the high-frequency information is mostly contained in the image gradient field, we propose to learn the gradient-mapping function between the high-resolution (HR) and the low-resolution (LR) image to regularize the fusion of multiple frames. In contrast to the existing spatial-domain networks, we train a deep gradient-mapping network to learn the horizontal and vertical gradients. We found that adding the low-frequency information (mainly from the LR image) to the gradient-learning network can boost the performance of the network. A forward and backward motion field prior is used to regularize the estimation of the motion flow between frames. For robust SR reconstruction, a weighting scheme is proposed to exclude the outlier data. Visual and quantitative evaluations on benchmark datasets demonstrate that our method is superior to many state-of-the-art methods and can recover better details with less artifacts.

A Short-Term Quantitative Precipitation Forecasting Approach Using Radar Data and a RAP Model

Very short-term (0~3 h) radar-based quantitative precipitation forecasting (QPF), also known as nowcasting, plays an essential role in flash flood warning, water resource management, and other hydrological applications. A novel nowcasting method combining radar data and a model wind field was developed and validated with two hurricane precipitation events. Compared with several existing nowcasting approaches, this work attempts to enhance the prediction capabilities from two major aspects. First, instead of using a radar reflectivity field, this work proposes the use of the rainfall rate field estimated from polarimetric radar variables in the motion field derivation. Second, the derived motion field is further corrected by the Rapid Refresh (RAP) model field. With the corrected motion field, the future rainfall rate field is predicted through a linear extrapolation method. The proposed method was validated using two hurricanes: Harvey and Irma. The proposed work shows an enhanced performance according to statistical scores. Compared with the model only and centroid-tracking only approaches, the average probability of detection (POD) increases about 25% and 50%; the average critical success index (CSI) increases about 20% and 37%; and the average false alarm rate (FAR) decreases about 14% and 16%, respectively.

Precipitation Nowcasting using Data-driven Reduced-order Model

<div>Radar-based precipitation nowcasting refers to predicting rain for a short period of time using radar reflectivity images. For dynamic nowcasting, motion fields can be extrapolated using an approximate and localized reduced-order model. Motion field estimation based on traditional Horn-Schunck (HS) algorithm suffers from over-smoothing at discontinuities in non-rigid and dissolving texture present in precipitation nowcasting products. An attempt to preserve the discontinuities using an L1 norm formulation in HS led to the use of Total Variation L1 norm (TVL1). In this paper, we propose a radar-based precipitation nowcasting model with TVL1-based estimation of motion field and Sparse Identification of Non-linear Dynamics (SINDy)-based</div><div>estimation of non-linear dynamics. TVL1 is effective in preserving the edges especially in the case of the eye of typhoons and squall lines while estimating motion vectors. SINDy captures the non-linear dynamics and generates the subsequent update values for the motion field based on a reduced-order representation. Finally, the SINDy-generated ensemble of motion field is used along with</div><div>the radar reflectivity image for generating precipitation nowcasts. We evaluated the effectiveness of TVL1 in preserving edges while capturing the motion field from non-rigid surfaces. The performance of the proposed TVL1-SINDy model in nowcasting weather events such as Typhoons and Squall lines are evaluated using performance metrics such as Mean Absolute Error (MAE), and Critical Success Index (CSI). Experimental results show that the proposed nowcasting system demonstrates better performance compared to the benchmark nowcasting models with lower MAE, higher CSI at higher lead times.</div>

Precipitation Nowcasting using Data-driven Reduced-order Model

<div>Radar-based precipitation nowcasting refers to predicting rain for a short period of time using radar reflectivity images. For dynamic nowcasting, motion fields can be extrapolated using an approximate and localized reduced-order model. Motion field estimation based on traditional Horn-Schunck (HS) algorithm suffers from over-smoothing at discontinuities in non-rigid and dissolving texture present in precipitation nowcasting products. An attempt to preserve the discontinuities using an L1 norm formulation in HS led to the use of Total Variation L1 norm (TVL1). In this paper, we propose a radar-based precipitation nowcasting model with TVL1-based estimation of motion field and Sparse Identification of Non-linear Dynamics (SINDy)-based</div><div>estimation of non-linear dynamics. TVL1 is effective in preserving the edges especially in the case of the eye of typhoons and squall lines while estimating motion vectors. SINDy captures the non-linear dynamics and generates the subsequent update values for the motion field based on a reduced-order representation. Finally, the SINDy-generated ensemble of motion field is used along with</div><div>the radar reflectivity image for generating precipitation nowcasts. We evaluated the effectiveness of TVL1 in preserving edges while capturing the motion field from non-rigid surfaces. The performance of the proposed TVL1-SINDy model in nowcasting weather events such as Typhoons and Squall lines are evaluated using performance metrics such as Mean Absolute Error (MAE), and Critical Success Index (CSI). Experimental results show that the proposed nowcasting system demonstrates better performance compared to the benchmark nowcasting models with lower MAE, higher CSI at higher lead times.</div>

The behavior of high-CAPE (convective available potential energy) summer convection in large-domain large-eddy simulations with ICON

Current state-of-the-art regional numerical weather prediction (NWP) models employ kilometer-scale horizontal grid resolutions, thereby simulating convection within the grey zone. Increasing resolution leads to resolving the 3D motion field and has been shown to improve the representation of clouds and precipitation. Using a hectometer-scale model in forecasting mode on a large domain therefore offers a chance to study processes that require the simulation of the 3D motion field at small horizontal scales, such as deep summertime moist convection, a notorious problem in NWP. We use the ICOsahedral Nonhydrostatic weather and climate model in large-eddy simulation mode (ICON-LEM) to simulate deep moist convection and distinguish between scattered, large-scale dynamically forced, and frontal convection. We use different ground- and satellite-based observational data sets, which supply information on ice water content and path, ice cloud cover, and cloud-top height on a similar scale as the simulations, in order to evaluate and constrain our model simulations. We find that the timing and geometric extent of the convectively generated cloud shield agree well with observations, while the lifetime of the convective anvil was, at least in one case, significantly overestimated. Given the large uncertainties of individual ice water path observations, we use a suite of observations in order to better constrain the simulations. ICON-LEM simulates a cloud ice water path that lies between the different observational data sets, but simulations appear to be biased towards a large frozen water path (all frozen hydrometeors). Modifications of parameters within the microphysical scheme have little effect on the bias in the frozen water path and the longevity of the anvil. In particular, one of our convective days appeared to be very sensitive to the initial and boundary conditions, which had a large impact on the convective triggering but little impact on the high frozen water path and long anvil lifetime bias. Based on this limited set of sensitivity experiments, the evolution of locally forced convection appears to depend more on the uncertainty of the large-scale dynamical state based on data assimilation than of microphysical parameters. Overall, we judge ICON-LEM simulations of deep moist convection to be very close to observations regarding the timing, geometrical structure, and cloud ice water path of the convective anvil, but other frozen hydrometeors, in particular graupel, are likely overestimated. Therefore, ICON-LEM supplies important information for weather forecasting and forms a good basis for parameterization development based on physical processes or machine learning.