Making forecasters SAD: Verification of Scale, Anisotropy and Direction using wavelets.

<p>Many atmospheric phenomena like fronts, convection and turbulence leave a distinct imprint on the spatial structure of meteorological fields such as precipitation, wind and temperature. Whether or not a forecast model is able to realistically simulate the resulting spatial correlation patterns is therefore a relevant question for model developers, forecasters and end users alike. Highly resolved numerical models have the potential to achieve this goal, but their realism is often difficult to assess objectively due to the sheer amount of data and wide variety of possible error contributions. </p><p>While some existing verification methods measure an overall “structure” error, most of these approaches are limited to precipitation fields and fail to produce specific, interpretable judgements. Here, we introduce a new structural verification technique based on the dual-tree complex wavelet transformation: The SAD-scores explicitly quantify how well the observed spatial Scales, degrees of Anisotropy and preferred Directions are represented by the simulation. Directional aspects in particular have previously often been neglected, but can be important in assessing the realism of predicted fronts, convergence lines and organized convection. </p><p>Unlike many established techniques, SAD is applicable not only to precipitation but to any meteorological field of interest. General verdicts like “the structure was predicted poorly” can be resolved into specific statements like “the modelled convection was too small in scale” or “the simulated front was too linear and rotated by an angle of X degrees”. The localized nature of the wavelets furthermore allows us to conveniently display the structural properties on a map. Lastly, making use of the inverse wavelet transform, we show how the detected structural errors can potentially be corrected, thereby leading the way towards future post-processing applications.</p>

Neural network for seasonal climate precipitation prediction on the Brazil

Precipitation is the hardest meteorological field to be predicted. An approach based on and optimal neural network is applied for climate precipitation prediction for the Brazil. A self-configurated multi-layer perceptron neural network (MLP-NN) is used as a predictor tool. The MLP-NN topology is found by solving an optimization problem by the Multi-Particle Collision Algorithm (MPCA). Prediction for Summer and Winter seasons are shown. The neural forecasting is evaluated by using the reanalysis data from the NCEP/NCAR and data from satellite GPCP (Global Precipitation Climatology Project -- monthly precipitation dataset).

NetCDF: Performance and Storage Optimization of Meteorological Data

<p>Any software relies on data, and the meteorological field is not an exception. The importance of using correct and accurate data is as important as using it efficiently. GRIB and NetCDF are the most popular file formats used in Meteorology, being able to store exactly the same data in any of them. However, they differ in how they internally treat the data, and transforming from GRIB (a simpler file format) to NetCDF is not enough to ensure the best efficiency for final applications.</p><p>In this study, we improved the performance and storage of <em>ARPEGE cloud cover forecasts post-processing with convolutional neural network</em> and <em>Precipitation Nowcasting using Deep Neural Network</em> projects (proposed in other sessions for the EGU general assembly). The data treatments of both projects were studied and different NetCDF capabilities were applied in order to obtain significantly faster execution times (up to 60 times faster) and more efficient space usage.</p>

Challenges in constructing a high-resolution, multi-site, and multivariate weather generator and potential solutions

<p>A reliable simulation of the spatiotemporal characteristics of the meteorological field is of great significance for hydrological impact studies. To approach this target, a number of weather generators (WGs) have been developed over the past few decades. However, a detailed literature review shows that currently developed WGs are subject to one or several aspects of the following limitations: (1) low spatial and temporal resolutions to describe the real spatiotemporal dynamics of meteorological processes; 2) incapability to simulate a spatially coherent, temporally consistent, and physically meaningful meteorological field; and 3) inability to extend into the future in a climate change context. To tackle these problems, this study proposes some potential solutions: (1) using the multi-site multivariate WGs (MMWGs) to simulate the spatial, temporal, and inter-variable dependencies in the meteorological field; (2) coupling the MMWGs with the resampling-based algorithms to generate high-resolution spatiotemporal meteorological data; and (3) perturbing the parameters of the distribution and dependency models based on the future climate projection. A case study is carried out and shows that the proposed solutions are effective in addressing the aforementioned challenges. These findings could assist in developing high-resolution MMWGs for weather simulation and impact assessment.</p>

Comparative Evaluation of Algorithms for Spatial Interpolation of Atmospheric State Parameters Based on a Dynamic Stochastic Model Taking into Account the Vertical Variation of a Meteorological Field

The paper presents a comparative analysis of two algorithms for the spatial interpolation of meteorological fields. Both algorithms are based on a four-dimensional low-order parametric dynamic stochastic model, taking into account the vertical variation of a meteorological field. The algorithms are characterized by different representations of the forecast model in state and observation space equations for the Kalman filter. The authors studied the accuracy of the spatial interpolation of temperature and wind fields for the developed algorithms. The results of the study are presented in this paper. Numerical simulation was conducted using long-term upper-air observations obtained for a typical mesometeorological range. The results of the study demonstrate that the accuracy of interpolation for the two considered algorithms is comparable.

Prediction of Uncertain Spatiotemporal Data Based on XML Integrated With Markov Chain

Since XML could benefit data management greatly and Markov chains have an advantage in data prediction, the authors study the methodology of predicting uncertain spatiotemporal data based on XML integrated with Markov chain. To accomplish this, first, the researchers devise an uncertain spatiotemporal data model based on XML. Then, the researchers put forward the method based on Markov chains to predict spatiotemporal data, which has taken the uncertainty into consideration. Next, the researchers apply the prediction method to meteorological field. Finally, the experimental results demonstrate the advantages the authors approach. Such a method of prediction could broaden the research field of spatiotemporal data, and provide a significant reference in the study of forecasting uncertain spatiotemporal data.

The Rapid Deployments to Wildfires Experiment (RaDFIRE): Observations from the Fire Zone

The Rapid Deployments to Wildfires Experiment (RaDFIRE) was a meteorological field campaign aimed at observing fire–atmosphere interactions during active wildfires. Using a rapidly deployable scanning Doppler lidar, airborne Doppler radar, and a suite of other instruments, the field campaign sampled 21 wildfires from 2013 to 2016 in the western United States. Observations include rotating convective plumes, plume interactions with stable layers and multilayered smoke detrainment, convective plume entrainment processes, smoke-induced density currents, and aircraft in situ observations of developing pyrocumulus. Collectively, these RaDFIRE observations highlight the range of meteorological phenomena associated with wildfires, especially plume dynamics, and will provide a valuable dataset for the modeling community.