Numerical Weather Prediction Model Output Research Articles

A Convolutional Neural Network for Tropical Cyclone Wind Structure Identification in kilometre-scale forecasts

Abstract Detection and tracking of Tropical Cyclones (TCs) in numerical weather prediction model outputs is essential for many applications, such as forecast guidance and real-time monitoring of events. While this task has been automated in the 90s with heuristic models, relying on a set of empirical rules and thresholds, the recent success of machine learning methods to detect objects in images opens new perspectives. This paper introduces and evaluates the capacity of a convolutional neural network based on the U-Net architecture to detect the TC wind structure, including maximum wind speed area and hurricane-force wind speed area, in the outputs of the convective-scale Arome model. A dataset of 400 Arome forecasts over the West Indies domain has been entirely hand labeled by experts, following a rigorous process to reduce heterogeneities. The U-Net performs well on a wide variety of TC intensities and shapes, with an average intersection-over-union metric around 0.8. Its performances however strongly depend on the TC strength, the detection of weak cyclones being more challenging since their structure is less well defined. The U-Net also significantly outperforms an operational heuristic detection model, with a significant gain for weak TCs, while running much faster. In a last part, the capacity of the U-Net to generalize on slightly different data is demonstrated in the context of a domain change and a resolution increase. In both cases, the pre-trained U-Net achieves similar performances as on the original dataset.

Downscaling of surface wind forecasts using convolutional neural networks

Abstract. Near-surface winds over complex terrain generally feature a large variability at the local scale. Forecasting these winds requires high-resolution numerical weather prediction (NWP) models, which drastically increase the duration of simulations and hinder them in running on a routine basis. Nevertheless, downscaling methods can help in forecasting such wind flows at limited numerical cost. In this study, we present a statistical downscaling of WRF (Weather Research and Forecasting) wind forecasts over southeastern France (including the southwestern part of the Alps) from its original 9 km resolution onto a 1 km resolution grid (1 km NWP model outputs are used to fit our statistical models). Downscaling is performed using convolutional neural networks (CNNs), which are the most powerful machine learning tool for processing images or any kind of gridded data, as demonstrated by recent studies dealing with wind forecast downscaling. The previous studies mostly focused on testing new model architectures. In this study, we aimed to extend these works by exploring different output variables and their associated loss function. We found that there is no one approach that outperforms the others in terms of both the direction and the speed at the same time. Finally, the best overall performance is obtained by combining two CNNs, one dedicated to the direction forecast based on the calculation of the normalized wind components using a customized mean squared error (MSE) loss function and the other dedicated to the speed forecast based on the calculation of the wind components and using another customized MSE loss function. Local-scale, topography-related wind features, which were poorly forecast at 9 km, are now well reproduced, both for speed (e.g., acceleration on the ridge, leeward deceleration, sheltering in valleys) and direction (deflection, valley channeling). There is a general improvement in the forecast, especially during the nighttime stable stratification period, which is the most difficult period to forecast. The result is that, after downscaling, the wind speed bias is reduced from −0.55 to −0.01 m s−1, the wind speed MAE is reduced from 1.02 to 0.69 m s−1 (32 % reduction) and the wind direction MAE is reduced from 25.9 to 15.5∘ (40 % reduction) in comparison with the 9 km resolution forecast.

Identifying and characterising trapped lee waves using deep learning techniques

Trapped lee waves, and resultant turbulent rotors downstream, present a hazard for aviation and land‐based transport. Though high‐resolution numerical weather prediction models can represent such phenomena, there is currently no simple and reliable automated method for detecting the extent and characteristics of these waves in model output. Spectral transform methods have traditionally been used to detect and characterise regions of wave activity in model and observational data; however, these methods can be slow and have their limitations. Machine‐learning (ML) techniques offer a new and potentially fruitful method of tackling this problem. We demonstrate that a deep‐learning model can be trained to accurately recognise and label coherent regions of lee waves from vertical velocity data on a single level from a high‐resolution numerical weather prediction (NWP) model. Using transfer learning, wave characteristics (wavelength, orientation, and amplitude) can be extracted from the trained segmentation model. The use of synthetic wave fields with prescribed wave characteristics makes this transfer learning possible without the need to characterise real complex wave fields. Addition of noise to the synthetic data makes the models more robust when applied to more complex and noisy NWP data. The collection of trained models produced provides a valuable tool to investigate the prevalence and nature of lee wave activity, as well as a new way for forecasters to detect resolved waves. The deep‐learning model was more capable and quicker at detecting and characterising lee waves than a spectral technique was. This work is just one example of how already established ML techniques can be used to detect and characterise complex weather phenomena from NWP model output and observational data, and how the careful use of synthetic data can reduce the requirements for large volumes of hand‐labelled training data for ML models.

Deep Learning–Based Summertime Turbulence Intensity Estimation Using Satellite Observations

Abstract Turbulence is what we want to avoid the most during flight. Numerical weather prediction (NWP) model–based methods for diagnosing turbulence have offered valuable guidance for pilots. NWP-based turbulence diagnostics show high accuracy in detecting turbulence in general. However, there is still room for improvements such as capturing convectively induced turbulence. In such cases, observation data can be beneficial to correctly locate convective regions and help provide corresponding turbulence information. Geostationary satellite data are commonly used for upper-level turbulence detection by utilizing its water vapor band information. The Geostationary Operational Environmental Satellite (GOES)-16 carries the Advanced Baseline Imager (ABI), which enables us to observe further down into the atmosphere with improved spatial, temporal, and spectral resolutions. Its three water vapor bands allow us to observe different vertical parts of the atmosphere, and from its infrared window bands, convective activity can be inferred. Such multispectral information from ABI can be helpful in inferring turbulence intensity at different vertical levels. This study develops U-Net based machine learning models that take ABI imagery as inputs to estimate turbulence intensity at three vertical levels: 10–18, 18–24, and above 24 kft (1 kft ≈ 300 m). Among six different U-Net-based models, U-Net3+ model with a filter size of three showed the best performance against the pilot report (PIREP). Two case studies are presented to show the strengths and weaknesses of the U-Net3+ model. The results tend to be overestimated above 24 kft, but estimates of 10–18 and 18–24 kft agree well with the PIREP, especially near convective regions. Significance Statement Turbulence is directly related to aviation safety as well as cost-effective aircraft operation. To avoid turbulence, turbulence diagnostics are calculated from numerical weather prediction (NWP) model outputs and are provided to pilots. The goal of this study is to develop a satellite data–driven machine learning model that estimates turbulence intensity in three different vertical layers to provide additional information along with the NWP-based turbulence diagnostics. Validation results against pilot reports show that the machine learning model performs comparable to NWP-based turbulence diagnostics. Furthermore, results with different channel selections reveal that using multiple water vapor channels can help extract additional information for estimating turbulence intensity at lower levels.

Multivariate Emulation of Kilometer-Scale Numerical Weather Predictions with Generative Adversarial Networks: A Proof of Concept

Abstract Emulating numerical weather prediction (NWP) model outputs is important to compute large datasets of weather fields in an efficient way. The purpose of the present paper is to investigate the ability of generative adversarial networks (GANs) to emulate distributions of multivariate outputs (10-m wind and 2-m temperature) of a kilometer-scale NWP model. For that purpose, a residual GAN architecture, regularized with spectral normalization, is trained against a kilometer-scale dataset from the AROME Ensemble Prediction System (AROME-EPS). A wide range of metrics is used for quality assessment, including pixelwise and multiscale Earth-mover distances, spectral analysis, and correlation length scales. The use of wavelet-based scattering coefficients as meaningful metrics is also presented. The GAN generates samples with good distribution recovery and good skill in average spectrum reconstruction. Important local weather patterns are reproduced with a high level of detail, while the joint generation of multivariate samples matches the underlying AROME-EPS distribution. The different metrics introduced describe the GAN’s behavior in a complementary manner, highlighting the need to go beyond spectral analysis in generation quality assessment. An ablation study then shows that removing variables from the generation process is globally beneficial, pointing at the GAN limitations to leverage cross-variable correlations. The role of absolute positional bias in the training process is also characterized, explaining both accelerated learning and quality-diversity trade-off in the multivariate emulation. These results open perspectives about the use of GAN to enrich NWP ensemble approaches, provided that the aforementioned positional bias is properly controlled.

Comparing global and regional downscaled NWP models for irradiance and photovoltaic power forecasting: ECMWF versus AROME

Inspecting the literature, much effort has been placed on the verification of irradiance forecasts from numerical weather prediction (NWP) models, as such forecasts are thought to have profound implications on the photovoltaic (PV) power forecasts, which in turn affects grid operators' confidence in integrating such power into the electricity grid. However, perhaps due to the proprietary nature of PV plants and lack of access to state-of-the-art NWP model output, only few have had the chance to conduct head-to-head comparisons of global mesoscale and regional downscaled NWP models, in terms of how their irradiance forecast inaccuracies propagate to PV power forecasts. In this regard, this work presents such a study, in which irradiance and PV power forecasts from the European Centre for Medium-Range Weather Forecasts' High-Resolution (HRES) and Météo-France's Application of Research to Operations at Mesoscale (AROME) models are thoroughly verified against the ground-based measurements from 32 research-grade radiometry stations and 94 actual PV plants in Hungary. A wide range of techniques and case studies concerning verification is herein considered, including variance ratio analysis, Murphy–Winkler decomposition, point-versus-areal verification, and seasonal verification. Despite that the results are too numerous to be summarized in a few sentences, the overarching observation from the verification exercise is that the performance of irradiance forecasts can only be used to infer that of PV power forecasts to a certain extent, which contrasts the conventional wisdom.

Performance of windshear/microburst detection algorithms using numerical weather prediction model data for selected tropical cyclone cases

AbstractThree windshear/microburst detection algorithms are used operationally at the Hong Kong International Airport. Their performance is studied in the present article by assuming the availability of the complete set of meteorological data from numerical weather prediction model output. Only selected tropical cyclone cases are considered. The performance is evaluated using pilot reports of windshear. It turns out that the glide path scan windshear detection algorithm has the best overall performance by considering the area under curve using the relative operating characteristic curves. This may be related to the better geometry in covering the glide paths of the airport for this particular algorithm.

Using Observed and Modelled Heat Fluxes for Improved Extrapolation of Wind Distributions

Modelling the horizontal and vertical variation of wind speed is crucial for wind energy applications. A model frequently used for this purpose is part of the Wind Atlas Analysis and Application program (WAsP). Here, we modify the model in WAsP to account for local atmospheric stability parameters. Atmospheric stability effects are treated by using the impact of a temperature scale on the geostrophic drag law and the diabatic logarithmic wind profile. Using this approach, wind-direction dependent mean and standard deviation of a surface-layer temperature scale and a mean boundary-layer height scale can be obtained from either numerical weather prediction model output or observations to improve vertical extrapolations of Weibull wind speed distribution parameters. The modified atmospheric stability model is validated at six flat sites in northwestern Europe. The surface-layer temperature scale is available from sonic anemometer measurements at three of the sites. At all sites the temperature scale is also estimated from reanalysis data and from mesoscale model output. The modified model improves the aggregated estimations of power density distributions when extrapolating to nearby locations from 5.2 to 3%, when using the temperature scale derived from either observations or mesoscale/reanalysis output compared to the current model.

Windmapper: An Efficient Wind Downscaling Method for Hydrological Models

AbstractEstimates of near‐surface wind speed and direction are key meteorological components for predicting many surface hydrometeorological processes that influence critical aspects of hydrological and biological systems. However, observations of near‐surface wind are typically spatially sparse. The use of these sparse wind fields to force distributed models, such as hydrological models, is greatly complicated in complex terrain, such as mountain headwaters basins. In these regions, wind flows are heavily impacted by overlapping influences of terrain at different scales. This can have a great impact on calculations of evapotranspiration, snowmelt, and blowing snow transport and sublimation. The use of high‐resolution atmospheric models allows for numerical weather prediction (NWP) model outputs to be dynamically downscaled. However, the computation burden for large spatial extents and long periods of time often precludes their use. Here, a wind‐library approach is presented to aid in downscaling NWP outputs and terrain‐correcting spatially interpolated observations. This approach preserves important spatial characteristics of the flow field at a fraction of the computational costs of even the simplest high‐resolution atmospheric models. This approach improves on previous implementations by: scaling to large spatial extents O(1M km2); approximating lee‐side effects; and fully automating the creation of the wind library. Overall, this approach was shown to have a third quartile RMSE of 1.8 and a third quartile RMSE of 58.2° versus a standalone diagnostic windflow model. The wind velocity estimates versus observations were better than existing empirical terrain‐based estimates and computational savings were approximately 100‐fold versus the diagnostic model.

A New Paradigm for Medium-Range Severe Weather Forecasts: Probabilistic Random Forest–Based Predictions

Abstract Historical observations of severe weather and simulated severe weather environments (i.e., features) from the Global Ensemble Forecast System v12 (GEFSv12) Reforecast Dataset (GEFS/R) are used in conjunction to train and test random forest (RF) machine learning (ML) models to probabilistically forecast severe weather out to days 4–8. RFs are trained with ∼9 years of the GEFS/R and severe weather reports to establish statistical relationships. Feature engineering is briefly explored to examine alternative methods for gathering features around observed events, including simplifying features using spatial averaging and increasing the GEFS/R ensemble size with time lagging. Validated RF models are tested with ∼1.5 years of real-time forecast output from the operational GEFSv12 ensemble and are evaluated alongside expert human-generated outlooks from the Storm Prediction Center (SPC). Both RF-based forecasts and SPC outlooks are skillful with respect to climatology at days 4 and 5 with diminishing skill thereafter. The RF-based forecasts exhibit tendencies to slightly underforecast severe weather events, but they tend to be well-calibrated at lower probability thresholds. Spatially averaging predictors during RF training allows for prior-day thermodynamic and kinematic environments to generate skillful forecasts, while time lagging acts to expand the forecast areas, increasing resolution but decreasing overall skill. The results highlight the utility of ML-generated products to aid SPC forecast operations into the medium range. Significance Statement Medium-range severe weather forecasts generated from statistical models are explored here alongside operational forecasts from the Storm Prediction Center (SPC). Human forecasters at the SPC rely on traditional numerical weather prediction model output to make medium-range outlooks and statistical products that mimic operational forecasts can be used as guidance tools for forecasters. The statistical models relate simulated severe weather environments from a global weather model to historical records of severe weather and perform noticeably better than human-generated outlooks at shorter lead times (e.g., day 4 and 5) and are capable of capturing the general location of severe weather events 8 days in advance. The results highlight the value in these data-driven methods in supporting operational forecasting.

Radiative cooling scale method for correcting hourly surface air temperature error in numerical weather prediction models

AbstractGlobally, a variety of numerical weather prediction (NWP) models are employed by national agencies. However, the derived outputs from these models are generally inadequate for precise cultivation management due to bias error. This study uses observational data acquired over several months to develop an equation for correcting surface air temperature (SAT) error in NWP model outputs. With the proposed method, model outputs are converted into potential temperature to cancel the influence of differences between computational elevation and real elevation. The potential temperature errors in the NWP model outputs are corrected using a linear regression equation in which a radiative cooling scale (RCS) serves as the independent variable. The RCS is computed using air temperature observed at a nearby meteorological observation station, together with upper‐level pressure data from the NWP model. Versions of the correction models are developed for specific situations: day, night, and upper‐level meteorology, classified into six groups. In an application of the proposed method, the root mean square error (RMSE) of the corrected SAT output from a global spectral model operated by the Japan Meteorological Agency (JMA) was 1.2–1.6 K; the RMSE of the corrected SAT output from a nonhydrostatic model operated by the JMA was 1.2–1.5 K. The proposed method corrected the NWP model SAT output errors to a level of accuracy similar to that of data corrected using the Kalman filtering technique without the requirement for constant observation at the target site.

A Comparative Study of Various Approaches for Producing Probabilistic Forecasts of Upper-Level Aviation Turbulence

Abstract The skill of operational deterministic turbulence forecasts is impacted by the uncertainties in both weather forecasts from the underlying numerical weather prediction (NWP) models and diagnoses of turbulence from the NWP model output. This study compares various probabilistic turbulence forecasting approaches to quantify these uncertainties and provides recommendations on the most suitable approach for operational implementation. The approaches considered are all based on ensembles of NWP forecasts and/or turbulence diagnostics, and include a multi-diagnostic ensemble (MDE), a time-lagged NWP ensemble (TLE), a forecast-model NWP ensemble (FME), and combined time-lagged MDE (TMDE) and forecast-model MDE (FMDE). Both case studies and statistical analyses are provided. The case studies show that the MDE approach that represents the uncertainty in turbulence diagnostics provides a larger ensemble spread than the TLE and FME approaches that represent the uncertainty in NWP forecasts. The larger spreads of MDE, TMDE, and FMDE allow for higher probabilities of detection for low percentage thresholds at the cost of increased false alarms. The small spreads of TLE and FME result in either hits with higher confidence or missed events, highly dependent on the performance of the underlying NWP model. Statistical evaluations reveal that increasing the number of diagnostics in MDE is a cost-effective and powerful method for describing the uncertainty of turbulence forecasts, considering trade-offs between accuracy and computational cost associated with using NWP ensembles. Combining either time-lagged or forecast-model NWP ensembles with MDE can further improve prediction skill and could be considered if sufficient computational resources are available.

Heavy rainfall forecasting for Dehradun capital city during monsoon season 2020

Dehradun city is Uttarakhand's capital and part of the doon valley, which lies between the Himalayas and the Shivaliks. India Meteorological Department has undertaken the Impact Based Forecasting (IBF) of heavy rainfall events for capital/major cities on an experimental basis from the monsoon season 2020. With the improvement in meteorological observation & monitoring network of Uttarakhand and short & medium range Numerical Weather Prediction (NWP) model outputs, it has become possible to provide impact-based forecast at a very small spatial scale with good accuracy. With the help of a dense network of 5 General Packet Radio Service (GPRS) based Automatic Weather Stations (AWS) & 2 Automatic Rain Gauges (ARG) within a 20 Km radius of Dehradun city center, it has become possible to capture & monitor these events. The spatial analysis of these events suggests that the local orographic features rather than large-scale synoptic features played a major role in heavy rainfalls over Dehradun city. The Heidke skill score of 0.63 and 100% probability of detection is found for the capital city forecast of Dehradun. The presence of monsoon trough to the north of its normal position, a cyclonic circulation south of Uttarakhand over Haryana or East-Rajasthan or West Uttar Pradesh and a Western disturbance around 70°E were found to be the synoptic systems favorable for the occurrence of heavy rainfall in Dehradun city. The occasional strong south-westerly wind flow from the Arabian Sea converging over the foothills of the Western Himalayas is also found to be responsible for the heavy downpour.

Relating forecast and satellite precipitation to generate future skillful ensemble forecasts over the northwest Himalayas at major avalanche and glacier sites

Availability of skillful precipitation forecast is crucial for regions such as the Himalayas which directly influence millions of people in the Indian subcontinent. The deterministic raw forecasts obtained from the numerical weather prediction (NWP) models are often biased. In this work, we present a novel approach of considering the spatial mismatch between the precipitation forecast and observation data in developing a statistical model. To deal with the mismatch in the spatial scale, a famous alternative approach could be ‘spatial downscaling and bias correction (SDBC)’. We attempt to blend the traditional approach of SDBC with a multiple-point geostatistics (MPS) approach to generate better-skilled forecasts. We consider the slope, aspect, and other topographical variables, along with the NWP model output, and observation data in a multivariate setup of the MPS (with SDBC) model to generate skillful forecast ensembles. We evaluate the daily generated 25 ensemble forecast members at avalanche and glacier sites. The generated multiple forecast members allow us to assess the uncertainty in the precipitation forecast. The outcome of the study would be helpful to water resource managers, disaster relief authorities, avalanche forecast agencies, and climate change experts for the sustainable development of the Himalayas.

Comparison of different forecasting tools for short-range lightning strike risk assessment

Thunderstorms, the main generator of lightning on earth, are characterized by the presence of extreme atmospheric conditions (turbulence, hail, heavy rain, wind shear, etc.). Consequently, the atmospheric conditions associated with this kind of phenomenon (in particular the strike itself) can be dangerous for aviation. This study focuses on the estimation of the lightning strike risk induced by thunderstorms over the sea, in a short-range forecast, from 0 to 24 h. In this framework, three methods have been developed and compared. The first method is based on the use of thresholds and weighting functions; the second method is based on a neural network approach, and the third method is based on the use of belief functions. Each method has been applied to the same dataset comprising predictors defined from numerical weather prediction model outputs. In order to assess the different methods, a “ground truth” dataset based on lightning stroke locations supplied by the World Wide Lightning Location Network (WWLLN) has been used. The choice of one method over the others will depend on the compromise that the user is willing to accept between false alarms, missed detections, and runtimes. The first method has a very low missed detection rate but a high false alarm rate, while the other two methods have much lower false alarm rates, but at the cost of a non-negligible missed detection rate. Finally, the third method is much faster than the other two methods.

A NASA–Air Force Precipitation Analysis for Near-Real-Time Operations

Abstract This article describes a new precipitation analysis algorithm developed by NASA for time-sensitive operations at the United States Air Force. Implemented as part of the Land Information System—a land modeling and data assimilation software framework—this NASA–Air Force Precipitation Analysis (NAFPA) combines numerical weather prediction model outputs with rain gauge measurements and satellite estimates to produce global, gridded 3-h accumulated precipitation fields at approximately 10-km resolution. Input observations are subjected to quality control checks before being used by the Bratseth analysis algorithm that converges to optimal interpolation. NAFPA assimilates up to 3.5 million observations without artificial data thinning or selection. To evaluate this new approach, a multiyear reanalysis is generated and intercompared with eight alternative precipitation products across the contiguous United States, Africa, and the monsoon region of eastern Asia. NAFPA yields superior accuracy and correlation over low-latency (up to 14 h) alternatives (numerical weather prediction and satellite retrievals), and often outperforms high-latency (up to 3.5 months) products, although the details for the latter vary by region and product. The development of NAFPA offers a high-quality, near-real-time product for use in meteorological, land surface, and hydrological research and applications. Significance Statement Precipitation is a key input to land modeling systems due to effects on soil moisture and other parts of the hydrologic cycle. It is also of interest to government decision-makers due to impacts on human activities. Here we present a new precipitation analysis based on available near-real-time data. By running the program for prior years and comparing with alternative products, we demonstrate that our analysis provides better accuracy and usually less bias than near-real-time satellite data alone, and better accuracy and correlation than data provided by numerical weather models. Our analysis is also competitive with other products created months after the fact, justifying confidence in using our analysis in near-real-time operations.

A DEEP LEARNING APPROACH FOR CROP TYPE MAPPING BASED ON COMBINED TIME SERIES OF SATELLITE AND WEATHER DATA

Abstract. Global Earth Monitor (GEM, Horizon 2020) takes advantage of the large volumes of available Earth Observation (EO), weather, climate and other non-EO data to establish economically viable continuous monitoring of the Earth. Within the GEM framework, the development of scalable and cost-effective solutions is being tested on several use-cases, with crop identification being one of them.Crop identification uses a combination of EO and weather data to enable automatic identification of crops. The use case supports operational decisions when managing crops and the monitoring of actual vs. planned or reported agricultural land use (e.g., Common Agricultural Policy monitoring). Satellite data and weather data come at very different temporal and spatial resolutions: Sentinel-2 constellation nominally provides an observation of a field every 5 days at 10 m spatial resolution, while weather data has continuous hourly time series at multi-km spatial resolution. We have designed ad-hoc routines to spatially aggregate satellite data at field level and to systematically compose layers of different time discretization series, so that each EO is associated with a complete time series (of opportune length) of weather variables at daily resolution. For each field, we extract the time series of the median over field pixels of Sentinel-2 L1C bands, cloud mask and cloud probability. For doing this we take advantage of Sentinel Hub's Statistical API (Sinergise, 2020), that enables the retrieval of statistics of band values and derived indices over a specified geographic area and time range. Using meteoblue dataset API (meteoblue, 2017), complete time series of daily weather data (NEMS4 model, meteoblue, 2008) are then associated to each field observation, following the systematic layer composition approach mentioned above. An opportune time series length is defined for each of the 17 weather variables we considered. To handle this kind of multi-dimensional layered data, we use a flexible encoding-decoding framework (FlexMod, designed by TUM as part of GEM project): multiple encoders are designed for features of different time length (namely EO data and weather variables) and are then passed to the decoder via a mediator. Thanks to the flexible design of FlexMod framework, different models and architectures can be easily tested by simply defining new encoders and/or decoders. We present results obtained on a dataset in Slovenia, where crop fields are labelled according to a Hierarchical Crop and Agriculture Taxonomy (HCAT). This taxonomy, based on the EAGLE-Matrix and EU regulations, is the one adopted in the EuroCrops project (Schneider et al. 2021). The classification of field crops takes advantage of Sentinel-2 satellite data and Numerical Weather Prediction model output data. We exploit the potential of FlexMod to test different feature extractors, temporal encoding frameworks and decoders and we present a comparison between results obtained training a long-short term memory (LSTM) implementation (Breizhcrops, Rußwurm et al. 2020) and a Self-attention transformer model (Vaswani et al. 2017), the latter showing the best performances with accuracy 0.904 and Cohen’s kappa 0.824. We moreover investigate the role of weather data by benchmarking results against those obtained with just satellite imagery. To better appraise the influence of the weather data we analyse how perturbing weather data in the testing dataset affects the final results. So far, we obtain in both cases very similar accuracies and Cohen’s kappa. A deeper analysis of crop-specific scores (precision, recall, F1) suggests that the training and testing datasets are too limited in terms of size and crop variability to draw any general conclusion over the role of weather. As future developments, once the EuroCrops datasets are ready, we plan to expand the training and testing dataset to cover a higher variability of climatological areas and increase the numerosity of the so far under-represented crops, in the attempt to draw more general conclusions around the influence of weather and the predictability of specific crop classes. Moreover, given the encouraging scores, we aim to perform crop type mapping at least at European scale, thanks to the availability of the EuroCrops data and the cost-effective big data solutions developed during GEM project.

A Collaborative Effort to Better Understand, Measure, and Model Atmospheric Exchange Processes over Mountains

Abstract In this essay, we highlight some challenges the atmospheric community is facing concerning adequate treatment of flows over mountains and their implications for numerical weather prediction (NWP), climate simulations, and impact modeling. With recent increases in computing power (and hence model resolution) numerical models start to face new limitations (such as numerical instability over steep terrain). At the same time there is a growing need for sufficiently reliable NWP model output to drive various impact models (for hydrology, air pollution, agriculture, etc.). The input information for these impact models is largely produced by the boundary layer (BL) parameterizations of NWP models. All known BL parameterizations assume flat and horizontally homogeneous surface conditions, and their performance and interaction with resolved flows is massively understudied over mountains—hence their output may be accidentally acceptable at best. We therefore advocate the systematic investigation of the so-called “mountain boundary layer” (MoBL), introduced to emphasize its many differences to the BL over flat and horizontally homogeneous terrain. An international consortium of scientists has launched a research program, TEAMx (Multi-Scale Transport and Exchange Processes in the Atmosphere over Mountains–Program and Experiment), to address some of the most pressing scientific challenges. TEAMx is endorsed by World Weather Research Programme (WWRP) and the Global Energy and Water Exchanges (GEWEX) project as a “cross-cutting project.” A program coordination office was established at the University of Innsbruck (Austria). This essay introduces the background to and content of a recently published white paper outlining the key research questions of TEAMx.

ПРИМЕНЕНИЕ ИСКУССТВЕННЫХ НЕЙРОННЫХ СЕТЕЙ ДЛЯ ОПРЕДЕЛЕНИЯ ОБЩЕГО СОДЕРЖАНИЯ ВОДЯНОГО ПАРА В АТМОСФЕРЕ ПО ДАННЫМ МИКРОВОЛНОВОГО РАДИОМЕТРА МТВЗА-ГЯ СО СПУТНИКА “МЕТЕОР-М” № 2-2

The paper presents the application of the artificial neural network algorithm for the retrieval of total precipitable water in the atmosphere over water and land from the measurements of MTVZA-GYa microwave radiometer on board the Meteor-M No. 2-2 satellite. Satellite-based estimates of total precipitable water were compared with radiosonde and AERONET data, as well as with the ECMWF numerical weather prediction model output. According to the comparison, the root-mean-square error (RMSE) does not exceed 4.5 mm for radiosonde data and is less than 4 mm for the ECMWF and AERONET data. The best accuracy is provided over water with the RMSE not exceeding 3 mm. The total precipitable water estimates retrieved from MTVZA-GYa and NOAA-20/ATMS radiometer data are consistent over water, while the MTVZА-GYa based estimates are more accurate over land.

A comprehensive country-based day-ahead wind power generation forecast model by coupling numerical weather prediction data and CatBoost with feature selection methods for Turkey

A country-based day-ahead wind power generation forecast (WPGF) model with a grid selection algorithm and feature selection models was proposed in this study. Atmospheric variables extracted from 300, 500, 700 hPa pressure levels, and surface level of ERA5 reanalysis data with 2.5° spatial resolution were used to train/validate the categorical boosting (CatBoost) model. A special grid selection algorithm was proposed by considering Turkey’s spatial distribution of wind power plants. The day-ahead forecasts of ECMWF’s HRES (High-resolution) were used as the test subset, therefore, paving the way for the operational use of the model. The proposed model could be considered much as a specialized machine learning based downscaling method for country-based WPGF due to using numerical weather prediction model outputs as its input. Results showed that the proposed model that uses fewer features has outperformed the other models with a normalized root mean square error of 7.6% and coefficient of determination of 0.8989.