Generalizing Model Output Statistics with Recurrent Neural Networks

Total Page:16

File Type:pdf, Size:1020Kb

Generalizing Model Output Statistics with Recurrent Neural Networks Generalizing Model Output Statistics With Recurrent Neural Networks Joan Creus-Costa Departments of Computer Science and Physics Stanford University [email protected] Abstract We present an architecture using a single neural network to learn a mapping between the raw outputs of a numerical prediction model and model output statistics of interest, in this case temperature and dewpoint temperature. Instead of learning a model for each of thousands of weather stations, we enable data reuse and better learning by using a single recurrent neural network that learns the biases based on the errors of the model in the previous days. This greater generalization enables for better adaptation to changes in NWP models and makes it easier to add new locations for which predictions of model output statistics are desired. It achieves comparable performance to NOAA’s state of the art model output statistics, with a root-mean-square error of 1.88 and 1.70 degrees Celsius in predicting temperature and dewpoint 24 hours in advance. 1 Introduction Global numerical weather prediction models, such as the American GFS or the European ECMWF, produce outputs for various variables—such as humidity or temperature—on a coarse grid with a resolution of approximately 27 km. These raw forecasts, however, do not directly translate to weather forecasts as we know them. The issue with them is threefold: one, it does not capture all the variables humans care about (such as probability of rain, or surface-level quantities); second, it can have a variety local and global biases and systematics; and thirdly, its limited resolution means interpolation is required to produce predictions of meaningful quantities at the points of interest [1]. As a result, a post-processing step, called Model Output Statistics (MOS) is applied to the raw model grid’s values to generate the desired quantities at the desired points. In order to generate this map from the numerical prediction to the refined, debiased forecast, a training set is gathered by looking at ground truth data collected at weather stations around the country, and then a regression is performed between the model outputs and ground truth to learn the biases and corrections. Traditionally, these model output statistics take the form of a multivariate linear regression (with some papers trying out ensemble probabilistic techniques [2], random forests, or neural networks [3]), usually applied independently to each weather station that has ground truth data. An interesting observation is the fact that, by the nature of what we’re trying to learn—localized biases and local climatology [1]—it would seem that the amount of training data that can be used for the models is fairly limited, because it scales merely with the cumulative time a given weather station has been operating, instead of the total data collected, which scales with the product of time elapsed and number of stations. In other words, having to create different models for each station means that data cannot be reused, even if intuitively it might make sense that some low-level characteristics are shared. On top of that, the amount of data that can be used for a given station is limited, too, since as the numerical weather model changes, the distribution of the errors shifts, which means that the regression cannot use old datapoints, [4] since the distribution of the error shifts over time. From this observations we motivate a desire to create a single model that can perform the MOS corrections for any station, instead of on a per-station basis. The benefits, besides potentially better accuracy coming from the increased training size, include the fact that MOS can adapt to changes in the NWP model much faster: if it previously took T days’ worth of data to collect enough data to recalibrate each station’s coefficients after a model change, with a unified CS230: Deep Learning, Fall 2019, Stanford University, CA. (LateX template borrowed from NIPS 2017.) Figure 1: Weather stations providing ground truth data, representing the intersection of ISD stations and MOS stations that had valid recorded measurements over recent years. The colors represent how they were used as part of RNN learning: training in blue, dev in red, and test in green. For the station-by-station tests, only the green stations were used. system data could be theoretically collected in T=N days (with N being the total number of stations, which is a few thousand). This realistically might mean that a few weeks may suffice. In particular, this paper will focus in predicting temperature and dewpoint temperature 24 hours ahead, based on features from the raw GFS model run. 2 Related work Most of the previous literature has used traditional learning methods, such as multivariate linear regression. For instance, in [5], one equation per station per predictand is produced as the output of their algorithm that outputs model output statistics for various air quality outputs. Another family of algorithms looks at ensemble outputs of the weather models and outputs probability distributions, as in [2]. This paper will consider the (simpler) problem of merely giving a single output rather than a full distribution. Neural networks have been used for model output statistics before. In [6], one neural network is trained for each of 31 stations in order to predict temperature. Nine input features are normalized and fed to a single-hidden-layer neural network with logistic activation. The number of hidden is varied and they find that the optimal number ranges from zero to eight depending on the station. Another example, applied to solar radiation forecasts, is [7]. Perhaps the most relevant and interesting related work was the 2018 paper in [3]. It creates ensemble forecast using neural networks, but in a departure from the previous work mentioned earlier (which mostly relied on a simple single hidden layer), it uses a more modern architecture, and tries to train a single big network. With a similar motivation of allowing station-specific information while training a single network, the paper uses an embedding—commonly found in natural language processing tasks—of size nemb = 2 to capture station-specific biases. That is, each of the 537 stations considered had two latent features that are learned and then go into the neural network. By combining that and adding auxiliary features besides temperature, it was able to beat other benchmark model output statistics. This interesting approach still used stations as explicit inputs (in that the station selects which embedding is used), but can use a much larger training corpus since it’s otherwise a single network. In this work, we will try to extend those ideas and remove the explicit station dependence by using a recurrent network instead that learns the biases based on the previous few days. In the interest of full disclosure, the author is not experienced in the field of meteorology and the above represents their best efforts in understanding developments and trends in the fields. 3 Dataset and Features Building a dataset and benchmark for the proved to be a challenging task, with many distinct datasets that required extensive post-processing and error checking. Due to the large amounts of data required to process—over a terabyte compressed—we created a custom framework for downloading and processing large amounts of data on a single-node, 2 48-core, 96-thread machine. This framework scaled well and let us process all the data required in parallel and fast, with caching proving particularly useful in the early stages of development. The datasets were mostly acquired from NOAA’s servers that have historical data and some historical predictions, with about 600 lines of Python. They are listed below: 1. Truth measurements performed by weather stations across the United States. These are available sometimes going as far as year 1901, but only recent measurements (from 2014 on) were downloaded since NWP models change every few years. These come from the Integrated Surface Database (ISD) [8]. 2. Global Forecast System predictions on a grid. For each GFS prediction cycle (which runs four times a day, at times 0, 6, 12, and 18 hours from midnight UTC), we download the forecasts every 3 hours up to 24 hours in advance, and we extract variables deemed relevant for our task at hand. The list was crafted based on the features used by [3] in their model output statistics, and includes a total of 19 measurements for each point in the grid. These are: 2-meter temperature, 2-meter dewpoint temperature, 2-meter relative humidity, convective available potential energy, latent and sensible heat net flux; downward short-wave, downward long-wave, upward short-wave, and upward long-wave radiation fluxes; surface pressure, grid elevation, volumetric soil moisture at 0 meters depth, geopotential height at 500 and 850 hPa, and the u and v components of wind at 500 hPa. On top of that, for the recurrent model, we add latitude, longitude, elevation, and fraction of the year (0 on January 1, and 1 on December 31), and the previous error, explained in Section 4.3. A post-processing script extracts those variables and saves them to binary matrices (indexed by the filesystem), which allows for very quick retrieval of values by using Linux memory maps supported by NumPy, with a custom function written to interpolate to a particular set of coordinates. Due to limited data on the NOAA website, only forecasts after June 2018 could be downloaded. The author spent some time trying to download a longer historical set from NCEP’s Research Data Archive but wasn’t successful; using the more limited dataset from the second half of 2018 on was deemed to be enough after getting early results.
Recommended publications
  • Gridded Model Output Statistics: Improving and Expanding
    6A.6 GRIDDED MODEL OUTPUT STATISTICS: IMPROVING AND EXPANDING Kathryn K. Gilbert*, Bob Glahn, Rebecca L. Cosgrove1, Kari L. Sheets, Geoffrey A. Wagner2 Meteorological Development Laboratory National Weather Service, NOAA Silver Spring, MD 1National Center for Environmental Prediction, National Weather Service, Camp Springs, MD 2Wyle Information Systems, McLean, VA 1. INTRODUCTION Since the original implementation, gridded MOS has undergone several adaptations to improve For years, National Weather Service (NWS) consistency as well as regional applicability. MDL forecasters have used Model Output Statistics developers have welcomed assessments of grid- (MOS) guidance (Glahn and Lowry 1972) as an ded MOS from local forecasters in order to make aid in producing text forecast products issued to adjustments to provide guidance which is both the user community. However, the methods fore- representative of the local area, and compatible casters use to generate products have changed with the process for generating forecasts for because of requirements to produce the official NDFD. Some of the improvements resulting from NWS 7-day forecasts on high-resolution grids. forecaster feedback include: removal of unrepre- These gridded, or digital, forecasts are put into a sentative forecasts from the analysis, expansion of National Digital Forecast Database (NDFD; Glahn the grid coverage, and modifications to the and Ruth 2003). NWS forecasters need support- land/water mask. ing guidance available on a grid, at a resolution comparable to the grid resolution used at the local In this paper, we discuss the current state of forecast office. One approach to the problem of the gridded MOS suite, as well as ongoing en- providing guidance on the required grid is to objec- hancements to the operational GFS-based MOS tively analyze forecasts at MOS stations.
    [Show full text]
  • Verification of Model Output Statistics Forecasts Associated with the North
    Allison Monarski, University of Maryland Masters Scholarly Paper, December 6, 2011 1 Department of Atmospheric and Oceanic Science Verification of Model Output Statistics forecasts associated with the North American Mesoscale model upgrade Results of work done during the Student Career Experience Program tour with the Statistical Modeling Branch (SMB) of the Meteorological Development Lab (MDL) of the National Weather Service (NWS), Silver Spring, MD, under the guidance of Mark Antolik & Kathy Gilbert, NWS/MDL/SMB I. Introduction and Motivation For the past four decades, the model output statistics (MOS) technique has been in operation and forecasters and meteorologists globally have used forecasts generated through MOS. MOS is a statistical approach to forecasting which involves determination of a relationship between a predictand and numerical weather prediction (NWP) model output for a variety of variables (predictors) at the same projection times (Glahn and Lowry 1972). Through the years, MOS has provided reliable forecasts, predicting quantities that the underlying NWP model is unable to predict. Specifically, the MOS technique can account for local effects on certain elements that the coarser resolution NWP model cannot detect (Antolik and Baker 2009). Usually, statistical forecasting systems, such as MOS, rely on many years of stable development data for optimal performance. However, due to frequent changes in the underlying structure, physics, and dynamics of the NWP models, this kind of data is difficult to obtain. Fortunately, developers at the National Weather Service’s (NWS) Meteorological Development Laboratory (MDL) have been able to show that reliable MOS forecasts can be produced using as little as just two years of data from models which are evolving (Antolik and Baker 2009).
    [Show full text]
  • Model Output Statistics Cascade to Improve Day Ahead Solar Irradiance Forecast
    Available online at www.sciencedirect.com ScienceDirect Solar Energy 117 (2015) 99–113 www.elsevier.com/locate/solener Model output statistics cascade to improve day ahead solar irradiance forecast M. Pierro a, F. Bucci a, C. Cornaro a,⇑, E. Maggioni b, A. Perotto b, M. Pravettoni c, F. Spada b a Department of Enterprise Engineering, University of Rome Tor Vergata, Via del Politecnico, 1, 00133 Rome, Italy b IDEAM srl, Via Frova 34, Cinisello Balsamo, MI, Italy c Institute of Applied Sustainability to the Build Environment, University of Applied Sciences and Arts of Southern Switzerland, Canobbio CH-6952, Switzerland Received 29 September 2014; received in revised form 9 April 2015; accepted 26 April 2015 Communicated by: Associate Editor Christian A. Gueymard Abstract In this paper a new hybrid Model Output Statistics (MOS), named MOS cascade, is developed to refine the day-ahead forecast of the global horizontal irradiance provided by the Weather Research and Forecast (WRF) model. The proposed approach is based on a sequence of two different MOS. The first, called MOSRH, is a new physically based algorithm, built to correct the treatment of humidity in the WRF radiation schemes. The second, called MOSNN, is based on artificial intelligence techniques and aims to correct the main systematic and learnable errors of the Numerical Weather Prediction output. The 1-day and 2-day forecast accuracies are analyzed via direct comparison with irradiance data measured in two sites, Rome and Lugano. The paper shows that a considerable reduction in error was achieved using MOSRH model and MOS cascade. The differences between the two sites are discussed in details.
    [Show full text]
  • Supporting Gridded Model Output Statistics Forecast Guidance System
    Supporting Gridded Model Output Statistics Forecast Guidance System Kari L. Sheets ABSTRACT The Meteorological Development Laboratory (MDL) of NOAA’s National Weather Service is creating a National Digital Guidance Database at a fine resolution to comple- ment the existing National Digital Forecast Database. To help accomplish this goal, MDL is producing a gridded Model Output Statistics (MOS) forecast guidance system. Currently, gridded MOS populates a 5-km grid with elements needed for weather fore- cast grids covering the contiguous United States. This paper describes the use of Geographic Information System (GIS) to gener- ate and quality control geophysical variables, create a station dictionary including land/water designations for the observing stations, and analyze or troubleshoot problem areas in gridded MOS weather elements. Stations used in traditional MOS development are unevenly distributed, leaving developers searching for additional data to aid in ana- lyzing these values to the aforementioned grid. GIS is integral in supplying the data to move from station-based MOS to gridded MOS. 1. Introduction The Meteorological Development Laboratory (MDL) of NOAA’s National Weather Service (NWS) is developing a National Digital Guidance Database (NDGD) at fine reso- lutions to complement the existing National Digital Forecast Database (NDFD, Glahn and Ruth 2003). To help accomplish this goal, MDL is creating a gridded forecast guid- ance system. Current forecast guidance is produced for the United States and its terri- tories at approximately 1800 hourly observing sites and over 5000 cooperative observ- ing sites by using the Model Output Statistics (MOS) technique (Glahn and Lowry 1972). In the MOS approach, observed predictand data are statistically related to predictors such as forecasts from dynamical models, surface observations, and geoclimatic infor- mation.
    [Show full text]
  • Statistical Post-Processing of Ensemble Forecasts of Temperature in Santiago De Chile
    Statistical post-processing of ensemble forecasts of temperature in Santiago de Chile Mailiu D´ıaza, Orietta Nicolisa;b, Julio Cesar´ Mar´ınc and Sandor´ Barand aDepartment of Statistics, University of Valpara´ıso Gran Breta~na1111, Valpara´ıso,Chile bFaculty of Engineering, Andres Bello University Quillota 980, Vi~nadel Mar, Chile cDepartment of Meteorology, University of Valpara´ıso Gran Breta~na644, Valpara´ıso,Chile dFaculty of Informatics, University of Debrecen Kassai ´ut26, H-4028 Debrecen, Hungary Abstract Currently all major meteorological centres generate ensemble forecasts using their operational ensemble prediction systems; however, it is a general problem that the spread of the ensemble is too small, resulting in underdispersive forecasts, leading to a lack of calibration. In order to correct this problem, different statistical calibration techniques have been developed in the last two decades. In the present work different post-processing techniques are tested for calibrating arXiv:1809.04042v1 [stat.AP] 11 Sep 2018 9 member ensemble forecast of temperature for Santiago de Chile, obtained by the Weather Research and Forecasting (WRF) model using different planetary boundary layer and land surface model parametrization. In particular, the ensemble model out- put statistics (EMOS) and Bayesian model averaging techniques are implemented and since the observations are characterized by large altitude differences, the estimation of model parameters is adapted to the actual conditions at hand. Compared to the raw ensemble all tested post-processing approaches significantly improve the calibration of probabilistic and the accuracy of point forecasts. The EMOS method using parameter estimation based on expert clustering of stations (according to their altitudes) shows the best forecast skill.
    [Show full text]
  • Statistical Postprocessing of Ensemble Forecasts for Severe Weather at Deutscher Wetterdienst
    Nonlin. Processes Geophys., 27, 473–487, 2020 https://doi.org/10.5194/npg-27-473-2020 © Author(s) 2020. This work is distributed under the Creative Commons Attribution 4.0 License. Statistical postprocessing of ensemble forecasts for severe weather at Deutscher Wetterdienst Reinhold Hess Deutscher Wetterdienst, Offenbach, Germany Correspondence: Reinhold Hess ([email protected]) Received: 7 January 2020 – Discussion started: 20 January 2020 Revised: 13 August 2020 – Accepted: 19 August 2020 – Published: 6 October 2020 Abstract. This paper gives an overview of Deutscher Wet- 1 Introduction terdienst’s (DWD’s) postprocessing system called Ensemble- MOS together with its motivation and the design con- Ensemble forecasting rose with the understanding of the lim- sequences for probabilistic forecasts of extreme events ited predictability of weather. This limitation is caused by based on ensemble data. Forecasts of the ensemble systems sparse and imperfect observations, approximating numerical COSMO-D2-EPS and ECMWF-ENS are statistically opti- data assimilation and modelling, and by the chaotic phys- mised and calibrated by Ensemble-MOS with a focus on se- ical nature of the atmosphere. The basic idea of ensemble vere weather in order to support the warning decision man- forecasting is to vary observations, initial and boundary con- agement at DWD. ditions, and physical parameterisations within their assumed Ensemble mean and spread are used as predictors for lin- scale of uncertainty and rerun the forecast model with these ear and logistic multiple regressions to correct for conditional changes. biases. The predictands are derived from synoptic observa- The obtained ensemble of forecasts expresses the distribu- tions and include temperature, precipitation amounts, wind tion of possible weather scenarios to be expected.
    [Show full text]
  • REFORECASTS an Important Dataset for Improving Weather Predictions
    REFORECASTS An Important Dataset for Improving Weather Predictions BY THOMAS M. HAMILL, JEFFREY S. WHITAKER, AND STEVEN L. MULLEN Statistical correction of today’s numerical forecast using a long set of reforecasts (hindcasts) dramatically improves its forecast skill. eanalyses such as the National Centers for reanalysis datasets have facilitated a wide range of Environmental Prediction (NCEP)–National research; for example, the Kalnay et al. (1996) article R Center for Atmospheric Research (NCAR) re- above has been cited more than 3200 times. analysis (Kalnay et al. 1996) and the European Centre In this article, we explore the value of a companion for Medium-Range Weather Forecasts (ECMWF) dataset to reanalyses, which we shall call “reforecasts.” 40-year reanalysis (ERA-40; Uppala et al. 2005) have These are retrospective weather forecasts generated become heavily used products for geophysical science with a fixed numerical model. Model developers research. These reanalyses run a practical, consistent could use them for diagnosing model bias, thereby data assimilation and short-range forecast system facilitating the development of new, improved ver- over a long period of time. While the observation sions of the model. Others could use them as data type and quality may change somewhat, the forecast for statistically correcting weather forecasts, thereby model and assimilation system are typically fixed. developing improved, user-specific products [e.g., This facilitates the generation of a reanalysis dataset model output statistics (MOS) Glahn and Lowry that is fairly consistent in quality over time. These 1972; Carter et al. 1989]. Others may use them for studies of atmospheric predictability. Unfortunately, extensive sets of reforecasts are not commonly pro- AFFILIATIONS: HAMILL AND WHITAKER—NOAA Earth System duced utilizing the same model version as is run Research Lab, Physical Sciences Division, Boulder, Colorado; operationally.
    [Show full text]
  • Articles Described by the Community Multiscale Air Datasets, Moderate-To-Large Biases Exist in Hourly Precipita- Quality Modeling System
    Geosci. Model Dev., 14, 3969–3993, 2021 https://doi.org/10.5194/gmd-14-3969-2021 © Author(s) 2021. This work is distributed under the Creative Commons Attribution 4.0 License. Evaluation of the offline-coupled GFSv15–FV3–CMAQv5.0.2 in support of the next-generation National Air Quality Forecast Capability over the contiguous United States Xiaoyang Chen1, Yang Zhang1, Kai Wang1, Daniel Tong2,6, Pius Lee4,3, Youhua Tang3,4, Jianping Huang5,6, Patrick C. Campbell3,4, Jeff Mcqueen5, Havala O. T. Pye7, Benjamin N. Murphy7, and Daiwen Kang7 1Department of Civil and Environmental Engineering, Northeastern University, Boston, MA 02115, USA 2Department of Atmospheric, Oceanic and Earth Sciences, George Mason University, Fairfax, VA 22030, USA 3Center for Spatial Information Science and System, George Mason University, Fairfax, VA 22030, USA 4Air Resources Laboratory, National Oceanic and Atmospheric Administration, College Park, MD 20740, USA 5National Oceanic and Atmospheric Administration/National Centers for Environmental Prediction/Environmental Modeling Center, College Park, MD 20740, USA 6IM Systems Group, Rockville, MD 20852, USA 7Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, NC 27711, USA Correspondence: Yang Zhang ([email protected]) Received: 11 August 2020 – Discussion started: 2 September 2020 Revised: 19 January 2021 – Accepted: 7 February 2021 – Published: 29 June 2021 Abstract. As a candidate for the next-generation National Teorological Aerodrome Reports (METAR) dataset.
    [Show full text]
  • Performance Assessment of General Circulation Model in Simulating Daily Precipitation and Temperature Using Multiple Gridded Datasets
    water Article Performance Assessment of General Circulation Model in Simulating Daily Precipitation and Temperature Using Multiple Gridded Datasets Najeebullah Khan 1,2, Shamsuddin Shahid 1 , Kamal Ahmed 1,2, Tarmizi Ismail 1 , Nadeem Nawaz 2 and Minwoo Son 3,* 1 School of Civil Engineering, Faculty of Engineering, Universiti Teknologi Malaysia (UTM), 81310 Johor Bahru, Malaysia; [email protected] (N.K.); [email protected] (S.S.); [email protected] (K.A.); [email protected] (T.I.) 2 Faculty of Water Resource Management, Lasbela University of Agriculture Water and Marine Sciences (LUAWMS), 90150 Uthal, Balochistan, Pakistan; [email protected] 3 Department of Civil Engineering, Chungnam National University, Daejeon 34134, Korea * Correspondence: [email protected]; Tel.: +82-42-821-5676 Received: 28 September 2018; Accepted: 4 December 2018; Published: 6 December 2018 Abstract: The performance of general circulation models (GCMs) in a region are generally assessed according to their capability to simulate historical temperature and precipitation of the region. The performance of 31 GCMs of the Coupled Model Intercomparison Project Phase 5 (CMIP5) is evaluated in this study to identify a suitable ensemble for daily maximum, minimum temperature and precipitation for Pakistan using multiple sets of gridded data, namely: Asian Precipitation– Highly-Resolved Observational Data Integration Towards Evaluation (APHRODITE), Berkeley Earth Surface Temperature (BEST), Princeton Global Meteorological Forcing (PGF) and Climate Prediction Centre (CPC) data. An entropy-based robust feature selection approach known as symmetrical uncertainty (SU) is used for the ranking of GCM. It is known from the results of this study that the spatial distribution of best-ranked GCMs varies for different sets of gridded data.
    [Show full text]
  • Guidelines on Ensemble Prediction System Postprocessing
    Guidelines on Ensemble Prediction System Postprocessing 2021 edition WEATHER CLIMATE WATER CLIMATE WEATHER WMO-No. 1254 Guidelines on Ensemble Prediction System Postprocessing 2021 edition WMO-No. 1254 WMO-No. 1254 © World Meteorological Organization, 2021 The right of publication in print, electronic and any other form and in any language is reserved by WMO. Short extracts from WMO publications may be reproduced without authorization, provided that the complete source is clearly indicated. Editorial correspondence and requests to publish, reproduce or translate this publication in part or in whole should be addressed to: Chair, Publications Board World Meteorological Organization (WMO) 7 bis, avenue de la Paix Tel.: +41 (0) 22 730 84 03 P.O. Box 2300 Fax: +41 (0) 22 730 81 17 CH-1211 Geneva 2, Switzerland Email: [email protected] ISBN 978-92-63-11254-5 NOTE The designations employed in WMO publications and the presentation of material in this publication do not imply the expression of any opinion whatsoever on the part of WMO concerning the legal status of any country, territory, city or area, or of its authorities, or concerning the delimitation of its frontiers or boundaries. The mention of specific companies or products does not imply that they are endorsed or recommended by WMO in preference to others of a similar nature which are not mentioned or advertised. PUBLICATION REVISION TRACK RECORD Part/chapter/ Date Purpose of amendment Proposed by Approved by section CONTENTS Page ACKNOWLEDGEMENTS . vii EXECUTIVE SUMMARY . viii KEY RECOMMENDATIONS . ix CHAPTER 1 . INTRODUCTION . 1 CHAPTER 2 . PHYSICAL POSTPROCESSING . 5 2.1 Meteorological diagnostic information.
    [Show full text]
  • A Model Output Deep Learning Method for Grid Temperature Forecasts in Tianjin Area
    applied sciences Article A Model Output Deep Learning Method for Grid Temperature Forecasts in Tianjin Area Keran Chen 1,2 , Ping Wang 1,2, Xiaojun Yang 3, Nan Zhang 3 and Di Wang 1,2,* 1 School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China; [email protected] (K.C.); [email protected] (P.W.) 2 Joint Laboratory of Intelligent Identification & Nowcasting Service for Convective System, CMA Public Meteorological Service Center, Beijing 100081, China 3 Tianjin Bureau of Meteorology, Tianjin 300074, China; [email protected] (X.Y.); [email protected] (N.Z.) * Correspondence: [email protected] Received: 25 June 2020; Accepted: 18 August 2020; Published: 22 August 2020 Abstract: In weather forecasting, numerical weather prediction (NWP) that is based on physical models requires proper post-processing before it can be applied to actual operations. Therefore, research on intelligent post-processing algorithms has always been an important topic in this field. This paper proposes a model output deep learning (MODL) method for post-processing, which can improve the forecast effect of numerical weather prediction. MODL is an end-to-end post-processing method based on deep convolutional neural network, which directly learns the mapping relationship between the forecast fields output by numerical model and the observation temperature field in order to obtain more accurate temperature forecasts. MODL modifies the existing deep convolution model according to the post-processing problem’s characteristics, thereby improving the performance of the weather forecast. This paper uses The International Grand Global Ensemble (TIGGE) dataset from European Centre for Medium-Range Weather Forecasts (ECMWF) and the observed air temperature of 2 m obtained from Tianjin meteorological station in order to test the post-processing performance of MODL.
    [Show full text]
  • SCN21-21 Updated: Notification of Dissemination Changes of Models Downstream of the GFS Due To
    NOUS41 KWBC 181800 AAA PNSWSH Service Change Notice 21-21 Updated National Weather Service Headquarters Silver Spring MD 200 PM EDT Thu Mar 18 2021 To: Subscribers: -NOAA Weather Wire Service -Emergency Managers Weather Information Network -NOAAPort Other NWS Partners, Users and Employees FROM: Mike Farrar Director National Centers for Environmental Prediction SUBJECT: Updated: Notification of Dissemination Changes of Models Downstream of the GFS due to the GFSv16 Upgrade: Effective March 22, 2021 Updated to reflect delay in implementation date to Monday, March 22 due to Critical Weather Day. Effective on or about March 22, 2021, beginning with the 1200 Coordinated Universal Time (UTC) cycle of the Global Forecast System (GFS), models that use GFS products for initialization will experience delays associated with the upgrade to GFS V16. Due to timing changes with the GFSv16 upgrade, which are laid out in its own SCN, the following downstream models will have timing changes across all dissemination channels including AWIPS, Satellite Broadcast Network(SBN)/NOAAPort, and NCEP and NWS web services (NOMADS, NOMADS GDS, NOMADS grib filter, FTPPRD and TGFTP). Ensemble Tracker: up to 16 min later https://nomads.ncep.noaa.gov/pub/data/nccf/com/ens_tracker/prod/ ftp://ftp.ncep.noaa.gov/pub/data/nccf/com/ens_tracker/prod/ Extra-Tropical Storm Surge (ETSS): up to 10 min later https://nomads.ncep.noaa.gov/pub/data/nccf/com/etss/prod/ ftp://ftp.ncep.noaa.gov/pub/data/nccf/com/etss/prod/ https://tgftp.nws.noaa.gov/SL.us008001/ST.expr/DF.gr2/DC.ndgd/GT
    [Show full text]