scholarly journals A Deep Learning Method for Bias Correction of ECMWF 24–240 h Forecasts

2021 ◽  
Author(s):  
Lei Han ◽  
Mingxuan Chen ◽  
Kangkai Chen ◽  
Haonan Chen ◽  
Yanbiao Zhang ◽  
...  
Keyword(s):  
Deep Learning ◽  
Relative Humidity ◽  
Wind Direction ◽  
Prediction Models ◽  
Weather Prediction ◽  
Ground Truth ◽  
Learning Method ◽  
Weather Variables ◽  
Medium Range Weather Forecast ◽  
Forecast Lead Time

AbstractCorrecting the forecast bias of numerical weather prediction models is important for severe weather warnings. The refined grid forecast requires direct correction on gridded forecast products, as opposed to correcting forecast data only at individual weather stations. In this study, a deep learning method called CU-net is proposed to correct the gridded forecasts of four weather variables from the European Centre for Medium-Range Weather Forecast Integrated Forecasting System global model (ECMWF-IFS): 2-m temperature, 2-m relative humidity, 10-m wind speed, and 10-m wind direction, with a forecast lead time of 24 h to 240 h in North China. First, the forecast correction problem is transformed into an image-to-image translation problem in deep learning under the CU-net architecture, which is based on convolutional neural networks. Second, the ECMWF-IFS forecasts and ECMWF reanalysis data (ERA5) from 2005 to 2018 are used as training, validation, and testing datasets. The predictors and labels (ground truth) of the model are created using the ECMWF-IFS and ERA5, respectively. Finally, the correction performance of CU-net is compared with a conventional method, anomaly numerical correction with observations (ANO). Results show that forecasts from CU-net have lower root mean square error, bias, mean absolute error, and higher correlation coefficient than those from ANO for all forecast lead times from 24 h to 240 h. CU-net improves upon the ECMWF-IFS forecast for all four weather variables in terms of the above evaluation metrics, whereas ANO improves upon ECMWF-IFS performance only for 2-m temperature and relative humidity. For the correction of the 10-m wind direction forecast, which is often difficult to achieve, CU-net also improves the correction performance.

scholarly journals HGPT2: An ERA5-Based Global Model to Estimate Relative Humidity

2021 ◽  
Vol 13 (11) ◽  
pp. 2179
Author(s):  
Pedro Mateus ◽  
Virgílio B. Mendes ◽  
Sandra M. Plecha
Keyword(s):  
Water Vapor ◽  
Relative Humidity ◽  
Real Time ◽  
Prediction Models ◽  
Weather Prediction ◽  
Precipitable Water ◽  
Precipitable Water Vapor ◽  
Global Navigation Satellite Systems ◽  
International Gnss Service ◽  
Weighted Mean Temperature

The neutral atmospheric delay is one of the major error sources in Space Geodesy techniques such as Global Navigation Satellite Systems (GNSS), and its modeling for high accuracy applications can be challenging. Improving the modeling of the atmospheric delays (hydrostatic and non-hydrostatic) also leads to a more accurate and precise precipitable water vapor estimation (PWV), mostly in real-time applications, where models play an important role, since numerical weather prediction models cannot be used for real-time processing or forecasting. This study developed an improved version of the Hourly Global Pressure and Temperature (HGPT) model, the HGPT2. It is based on 20 years of ERA5 reanalysis data at full spatial (0.25° × 0.25°) and temporal resolution (1-h). Apart from surface air temperature, surface pressure, zenith hydrostatic delay, and weighted mean temperature, the updated model also provides information regarding the relative humidity, zenith non-hydrostatic delay, and precipitable water vapor. The HGPT2 is based on the time-segmentation concept and uses the annual, semi-annual, and quarterly periodicities to calculate the relative humidity anywhere on the Earth’s surface. Data from 282 moisture sensors located close to GNSS stations during 1 year (2020) were used to assess the model coefficients. The HGPT2 meteorological parameters were used to process 35 GNSS sites belonging to the International GNSS Service (IGS) using the GAMIT/GLOBK software package. Results show a decreased root-mean-square error (RMSE) and bias values relative to the most used zenith delay models, with a significant impact on the height component. The HGPT2 was developed to be applied in the most diverse areas that can significantly benefit from an ERA5 full-resolution model.

scholarly journals Understanding the relationship between clouds and surface downward radiation forecast errors with Unsupervised Deep Learning

2021 ◽  
Author(s):  
Matthias Zech ◽  
Lueder von Bremen
Keyword(s):  
Deep Learning ◽  
Unsupervised Learning ◽  
Prediction Models ◽  
Weather Prediction ◽  
Reanalysis Data ◽  
Cloud Formation ◽  
Trade Wind ◽  
Forecast Errors ◽  
Model Learning ◽  
Unsupervised Deep Learning

<p>Cloudiness is a difficult parameter to forecast and has improved relatively little over the last decade in numerical weather prediction models as the EMCWF IFS. However, surface downward solar radiation forecast (ssrd) errors are becoming more important with higher penetration of photovoltaics in Europe as forecasts errors induce power imbalances that might lead to high balancing costs. This study continues recent approaches to better understand clouds using satellite images with Deep Learning. Unlike other studies which focus on shallow trade wind cumulus clouds over the ocean, this study investigates the European land area. To better understand the clouds, we use the daily MODIS optical cloud thickness product which shows both water and ice phase of the cloud. This allows to consider both cloud structure and cloud formation during learning. It is also much easier to distinguish between snow and cloud in contrast to using visible bands. Methodologically, it uses the Unsupervised Learning approach <em>tile2vec</em> to derive a lower dimensional representation of the clouds. Three cloud regions with two similar neighboring tiles and one tile from a different time and location are sampled to learn lower-rank embeddings. In contrast to the initial <em>tile2vec</em> implementation, this study does not sample arbitrarily distant tiles but uses the fractal dimension of the clouds in a pseudo-random sampling fashion to improve model learning.</p><p>The usefulness of the cloud segments is shown by applying them in a case study to investigate statistical properties of ssrd forecast errors over Europe which are derived from hourly ECMWF IFS forecasts and ERA5 reanalysis data. This study shows how Unsupervised Learning has high potential despite its relatively low usage compared to Supervised Learning in academia. It further shows, how the generated land cloud product can be used to better characterize ssrd forecast errors over Europe.</p>

scholarly journals Comparison of Snow Cover from Satellite and Numerical Weather Prediction Models in the Northern Hemisphere and Northern Europe

2009 ◽  
Vol 48 (6) ◽  
pp. 1199-1216 ◽  
Author(s):  
Otto Hyvärinen ◽  
Kalle Eerola ◽  
Niilo Siljamo ◽  
Jarkko Koskinen
Keyword(s):  
Snow Cover ◽  
Prediction Models ◽  
Weather Prediction ◽  
Ground Truth ◽  
In Situ Measurements ◽  
Lower Atmosphere ◽  
Limited Area ◽  
In Situ Observations ◽  
Using Data

Abstract Snow cover has a strong effect on the surface and lower atmosphere in NWP models. Because the progress of in situ observations has stalled, satellite-based snow analyses are becoming increasingly important. Currently, there exist several products that operationally map global or continental snow cover. In this study, satellite-based snow cover analyses from NOAA, NASA, and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), and NWP snow analyses from the High-Resolution Limited-Area Model (HIRLAM) and ECMWF, were compared using data from January to June 2006. Because no analyses were independent and since available in situ measurements were already used in the NWP analyses, no independent ground truth was available and only the consistency between analyses could be compared. Snow analyses from NOAA, NASA, and ECMWF were similar, but the analysis from NASA was greatly hampered by clouds. HIRLAM and EUMETSAT deviated most from other analyses. Even though the analysis schemes of HIRLAM and ECMWF were quite similar, the resulting snow analyses were quite dissimilar, because ECMWF used the satellite information of snow cover in the form of NOAA analyses, while HIRLAM used none. The differences are especially prominent in areas around the snow edge where few in situ observations are available. This suggests that NWP snow analyses based only on in situ measurements would greatly benefit from inclusion of satellite-based snow cover information.

Probabilistic fire-danger forecasting: A framework for week-two forecasts using statistical post-processing techniques and the Global ECMWF Fire Forecast System (GEFF)

2021 ◽  
Author(s):  
Rochelle P. Worsnop ◽  
Michael Scheuerer ◽  
Francesca Di Giuseppe ◽  
Christopher Barnard ◽  
Thomas M. Hamill ◽  
...  
Keyword(s):  
Prediction Models ◽  
Weather Prediction ◽  
Weather Forecast ◽  
Reanalysis Data ◽  
Post Processing ◽  
Medium Range Weather Forecast ◽  
Skill Scores ◽  
Probabilistic Forecasts ◽  
Systematic Biases ◽  
Processing Techniques

AbstractWildfire guidance two weeks ahead is needed for strategic planning of fire mitigation and suppression. However, fire forecasts driven by meteorological forecasts from numerical weather prediction models inherently suffer from systematic biases. This study uses several statistical-postprocessing methods to correct these biases and increase the skill of ensemble fire forecasts over the contiguous United States 8–14 days ahead. We train and validate the post-processing models on 20 years of European Centre for Medium-range Weather Forecast (ECMWF) reforecasts and ERA5 reanalysis data for 11 meteorological variables related to fire, such as surface temperature, wind speed, relative humidity, cloud cover, and precipitation. The calibrated variables are then input to the Global ECMWF Fire Forecast (GEFF) system to produce probabilistic forecasts of daily fire-indicators which characterize the relationships between fuels, weather, and topography. Skill scores show that the post-processed forecasts overall have greater positive skill at Days 8–14 relative to raw and climatological forecasts. It is shown that the post-processed forecasts are more reliable at predicting above- and below-normal probabilities of various fire indicators than the raw forecasts and that the greatest skill for Days 8–14 is achieved by aggregating forecast days together.

Insights into the atmospheric state through observations of infrasound from a ground-truth source at regional distance

2021 ◽  
Author(s):  
Karl Koch ◽  
Christoph Pilger
Keyword(s):  
Sound Speed ◽  
Prediction Models ◽  
Sound Propagation ◽  
Middle Atmosphere ◽  
Weather Prediction ◽  
Winter Season ◽  
Ground Truth ◽  
Speed Ratio ◽  
German Aerospace ◽  
Infrasound Signal

<p>From the more than 160 tests of the ARIANE-5 main engine carried out by the German Aerospace Center (DLR) facility near Heilbronn, Germany, a large overall portion was detected at IMS infrasound station IS26 in the Bavarian forest. Located at a distance of about 320 km in an easterly direction (99° east-southeast from North) these observations were mostly made in the winter season between October and April with a detection rate of more than 70% , as stratospheric winds favor infrasound propagating through the atmosphere within the stratospheric duct. Only two exceptions were found for the summer season when stratospheric ducting is not predicted neither by climatologies nor the applied weather prediction models, due to a reversal of the middle atmosphere wind pattern.</p><p>Numerical weather prediction models for summer and winter seasons, or times with detections or non-detections were compared. It is then found that these models differ significantly in the sound speed profiles producing either a strong stratospheric duct for altitudes between 30 and 60 km in the case of detection, i.e. in winter months – or a lack thereof inhibiting regional sound propagation in summer months. It is of course reflected by the effective sound speed ratio, mostly exceeding a value of 1 for detections and less than 1 for non-detections. A significant portion of profiles representing non-detections, however, exhibit a sound speed profile that should enable infrasound signal observations. These cases are analyzed in detail to identify which fine structures within the sound speed profiles could explain the lack of observations.</p>

scholarly journals The Gridding of MOS

2009 ◽  
Vol 24 (2) ◽  
pp. 520-529 ◽  
Author(s):  
Bob Glahn ◽  
Kathryn Gilbert ◽  
Rebecca Cosgrove ◽  
David P. Ruth ◽  
Kari Sheets
Keyword(s):  
United States ◽  
Prediction Models ◽  
Weather Prediction ◽  
Correction Method ◽  
Maximum Temperature ◽  
Weather Variables ◽  
Weather Forecasts ◽  
Surface Weather ◽  
Model Output Statistics ◽  
Numerical Weather Prediction Models

Abstract Model output statistics (MOS) guidance forecasts have been produced at stations and provided to National Weather Service forecasters and private entities for over three decades. As the numerical weather prediction models became more accurate, MOS followed that trend. Up until a few years ago, the MOS produced at observation locations met the basic need for guidance. With the advent of the Interactive Forecast Preparation System and the National Digital Forecast Database, gridded MOS forecasts became needed as guidance for forecasters. One method of providing such grids is to objectively analyze the MOS forecasts for points. A basic successive correction method has been extended to analyze MOS forecasts and surface weather variables. This method is being applied to MOS forecasts to provide guidance for producing grids of sensible weather elements such as temperature, clouds, and snow amount. Guidance forecasts have been implemented for the conterminous United States for most weather elements contained in routine weather forecasts. This paper describes the method applied to daytime maximum temperature over the conterminous United States and gives example results.

Evaluation of Forecasts of the Water Vapor Signature of Atmospheric Rivers in Operational Numerical Weather Prediction Models

2013 ◽  
Vol 28 (6) ◽  
pp. 1337-1352 ◽  
Author(s):  
Gary A. Wick ◽  
Paul J. Neiman ◽  
F. Martin Ralph ◽  
Thomas M. Hamill
Keyword(s):  
Water Vapor ◽  
Lead Time ◽  
Prediction Models ◽  
Weather Prediction ◽  
Time Model ◽  
Atmospheric Rivers ◽  
Position Bias ◽  
Forecast Lead Time ◽  
Northeastern Pacific ◽  
Numerical Weather Prediction Models

Abstract The ability of five operational ensemble forecast systems to accurately represent and predict atmospheric rivers (ARs) is evaluated as a function of lead time out to 10 days over the northeastern Pacific Ocean and west coast of North America. The study employs the recently developed Atmospheric River Detection Tool to compare the distinctive signature of ARs in integrated water vapor (IWV) fields from model forecasts and corresponding satellite-derived observations. The model forecast characteristics evaluated include the prediction of occurrence of ARs, the width of the IWV signature of ARs, their core strength as represented by the IWV content along the AR axis, and the occurrence and location of AR landfall. Analysis of three cool seasons shows that while the overall occurrence of ARs is well forecast out to a 10-day lead, forecasts of landfall occurrence are poorer, and skill degrades with increasing lead time. Average errors in the position of landfall are significant, increasing to over 800 km at 10-day lead time. Also, there is a 1°–2° southward position bias at 7-day lead time. The forecast IWV content along the AR axis possesses a slight moist bias averaged over the entire AR but little bias near landfall. The IWV biases are nearly independent of forecast lead time. Model spatial resolution is a factor in forecast skill and model differences are greatest for forecasts of AR width. This width error is greatest for coarser-resolution models that have positive width biases that increase with forecast lead time.

scholarly journals Deep Learning Method for Prediction of Patient-Specific Dose Distribution in Breast Cancer

2021 ◽  
Author(s):  
Sang Hee Ahn ◽  
EunSook Kim ◽  
Chankyu Kim ◽  
Wonjoong Cheon ◽  
Myeongsoo Kim ◽  
...  
Keyword(s):  
Breast Cancer ◽  
Neural Network ◽  
Deep Learning ◽  
Dose Distribution ◽  
Prediction Models ◽  
The Body ◽  
Patient Specific ◽  
Learning Method ◽  
Dose Prediction ◽  
Test Sets

Abstract BackgroundPatient-specific dose prediction improves the efficiency and quality of radiation treatment planning and reduces the time required to find the optimal plan. In this study, a patient-specific dose prediction model was developed for a left-sided breast clinical case using deep learning and its performance was compared with that of conventional knowledge-based planning using RapidPlan™.MethodsPatient-specific dose prediction was performed using a contour image of the planning target volume (PTV) and organs at risk (OARs) with a U-net-based modified dose prediction neural network. A database of 50 volumetric modulated arc therapy (VMAT) plans for left-sided breast cancer patients was utilized to produce training and validation datasets. The dose prediction deep neural network (DpNet) feature weights of the previously learned convolution layers were applied to the test on a cohort of 10 test sets. With the same patient data set, dose prediction was performed for the 10 test sets after training in RapidPlan. The 3D dose distribution, absolute dose difference error, dose-volume histogram, 2D gamma index, and iso-dose dice similarity coefficient were used for quantitative evaluation of the dose prediction.ResultsThe mean absolute error (MAE) and one standard deviation (SD) between the clinical and deep learning dose prediction models were 0.02 ± 0.04%, 0.01 ± 0.83%, 0.16 ±0.82%, 0.52 ± 0.97% -0.88 ± 1.83%, -1.16 ± 2.58%, and -0.97 ± 1.73% for D95%, Dmean in the PTV, and the OARs of the body, left breast, heart, left lung, and right lung, respectively, and those measured between the clinical and RapidPlan dose prediction models were 0.02 ± 0.04%, 0.87 ± 0.63%, -0.29 ± 0.98%, 1.30 ± 0.86%, -0.32 ± 1.10%, 0.12 ± 2.13, and -1.74 ±1.79%, respectively.ConclusionsIn this study, a deep learning method for dose prediction was developed and was demonstrated to predict patient-specific doses for left-sided breast cancer accurately. Using the deep learning framework, the efficiency and accuracy of the dose prediction were compared to those of RapidPlan. The doses predicted by deep learning were superior to the results of the RapidPlan-generated VMAT plan. The proposed model will be able to maintain treatment plan quality and increase efficiency through patient dose prediction.

Sign in / Sign up

Export Citation Format

Share Document