Two-Stage Spatiotemporal Context Refinement Network for Precipitation Nowcasting

Precipitation nowcasting by radar echo extrapolation using machine learning algorithms is a field worthy of further study, since rainfall prediction is essential in work and life. Current methods of predicting the radar echo images need further improvement in prediction accuracy as well as in presenting the predicted details of the radar echo images. In this paper, we propose a two-stage spatiotemporal context refinement network (2S-STRef) to predict future pixel-level radar echo maps (deterministic output) more accurately and with more distinct details. The first stage is an efficient and concise spatiotemporal prediction network, which uses the spatiotemporal RNN module embedded in an encoder and decoder structure to give a first-stage prediction. The second stage is a proposed detail refinement net, which can preserve the high-frequency detailed feature of the radar echo images by using the multi-scale feature extraction and fusion residual block. We used a real-world radar echo map dataset of South China to evaluate the proposed 2S-STRef model. The experiments showed that compared with the PredRNN++ and ConvLSTM method, our 2S-STRef model performs better on the precipitation nowcasting, as well as at the image quality evaluating index and the forecasting indices. At a given 45dBZ echo threshold (heavy precipitation) and with a 2 h lead time, the widely used CSI, HSS, and SSIM indices of the proposed 2S-STRef model are found equal to 0.195, 0.312, and 0.665, respectively. In this case, the proposed model outperforms the OpticalFlow method and PredRNN++ model.

Strong Spatiotemporal Radar Echo Nowcasting Combining 3DCNN and Bi-Directional Convolutional LSTM

In order to solve the existing problems of easy spatiotemporal information loss and low forecast accuracy in traditional radar echo nowcasting, this paper proposes an encoding-forecasting model (3DCNN-BCLSTM) combining 3DCNN and bi-directional convolutional long short-term memory. The model first constructs dimensions of input data and gets 3D tensor data with spatiotemporal features, extracts local short-term spatiotemporal features of radar echoes through 3D convolution networks, then utilizes constructed bi-directional convolutional LSTM to learn global long-term spatiotemporal feature dependencies, and finally realizes the forecast of echo image changes by forecasting network. This structure can capture the spatiotemporal correlation of radar echoes in continuous motion fully and realize more accurate forecast of moving trend of short-term radar echoes within a region. The samples of radar echo images recorded by Shenzhen and Hong Kong meteorological stations are used for experiments, the results show that the critical success index (CSI) of this proposed model for eight predicted echoes reaches 0.578 when the echo threshold is 10 dBZ, the false alarm ratio (FAR) is 20% lower than convolutional LSTM network (ConvLSTM), and the mean square error (MSE) is 16% lower than the real-time optical flow by variational method (ROVER), which outperforms the current state-of-the-art radar echo nowcasting methods.

Evolution of Severe and Nonsevere Convection Inferred from GOES-Derived Cloud Properties

Geostationary satellites [e.g., the Geostationary Operational Environmental Satellite (GOES)] provide high temporal resolution of cloud development and motion, which is essential to the study of many mesoscale phenomena, including thunderstorms. Initial research on thunderstorm growth with geostationary imagery focused on the mature stages of storm evolution, whereas more recent research on satellite-observed storm growth has concentrated on convective initiation, often defined arbitrarily as the presence of a given radar echo threshold. This paper seeks to link the temporal trends in robust GOES-derived cloud properties with the future occurrence of severe-weather radar signatures during the development phase of thunderstorm evolution, which includes convective initiation. Two classes of storms (severe and nonsevere) are identified and tracked over time in satellite imagery, providing distributions of satellite growth rates for each class. The relationship between the temporal trends in satellite-derived cloud properties and Next Generation Weather Radar (NEXRAD)-derived storm attributes is used to show that this satellite-based approach can potentially be used to extend severe-weather-warning lead times (with respect to radar-derived signatures), without a substantial increase in false alarms. In addition, the effect of varying temporal sampling is investigated on several storms during a period of GOES super-rapid-scan operations (SRSOR). It is found that, from a satellite perspective, storms evolve significantly on time scales shorter than the current GOES operational scan strategies.

Short-Latency, Goal-Directed Movements of the Pinnae to Sounds That Produce Auditory Spatial Illusions

The precedence effect (PE) is an auditory spatial illusion whereby two identical sounds presented from two separate locations with a delay between them are perceived as a fused single sound source whose position depends on the value of the delay. By training cats using operant conditioning to look at sound sources, we have previously shown that cats experience the PE similarly to humans. For delays less than ±400 μs, cats exhibit summing localization, the perception of a “phantom” sound located between the sources. Consistent with localization dominance, for delays from 400 μs to ∼10 ms, cats orient toward the leading source location only, with little influence of the lagging source. Finally, echo threshold was reached for delays >10 ms, where cats first began to orient to the lagging source. It has been hypothesized by some that the neural mechanisms that produce facets of the PE, such as localization dominance and echo threshold, must likely occur at cortical levels. To test this hypothesis, we measured both pinnae position, which were not under any behavioral constraint, and eye position in cats and found that the pinnae orientations to stimuli that produce each of the three phases of the PE illusion was similar to the gaze responses. Although both eye and pinnae movements behaved in a manner that reflected the PE, because the pinnae moved with strikingly short latencies (∼30 ms), these data suggest a subcortical basis for the PE and that the cortex is not likely to be directly involved.

Influence of Sound Source Location on the Behavior and Physiology of the Precedence Effect in Cats

Psychophysical experiments on the precedence effect (PE) in cats have shown that they localize pairs of auditory stimuli presented from different locations in space based on the spatial position of the stimuli and the interstimulus delay (ISD) between the stimuli in a manner similar to humans. Cats exhibit localization dominance for pairs of transient stimuli with |ISDs| from ∼0.4 to 10 ms, summing localization for |ISDs| < 0.4 ms and breakdown of fusion for |ISDs| > 10 ms, which is the approximate echo threshold. The neural correlates to the PE have been described in both anesthetized and unanesthetized animals at many levels from auditory nerve to cortex. Single-unit recordings from the inferior colliculus (IC) and auditory cortex of cats demonstrate that neurons respond to both lead and lag sounds at ISDs above behavioral echo thresholds, but the response to the lag is reduced at shorter ISDs, consistent with localization dominance. Here the influence of the relative locations of the leading and lagging sources on the PE was measured behaviorally in a psychophysical task and physiologically in the IC of awake behaving cats. At all configurations of lead-lag stimulus locations, the cats behaviorally exhibited summing localization, localization dominance, and breakdown of fusion. Recordings from the IC of awake behaving cats show neural responses paralleling behavioral measurements. Both behavioral and physiological results suggest systematically shorter echo thresholds when stimuli are further apart in space.