Assessment of meteorological parameters in predicting seasonal temperature of Dhaka city using ANN.

In this study, an attempt has been made to investigate the possibility of a machine learning model, Artificial Neural Network (ANN) for seasonal prediction of the temperature of Dhaka city. Prior knowledge of temperature is essential, especially in tropical regions like Dhaka, as it aids in forecasting heatwaves and implementing effective preparedness schemes. While various machine learning models have been employed for the prediction of hot weather across the world, research specially focused on Bangladesh is limited. Additionally, the application of machine learning models needs to be curated to suit the particular weather features of any region. Therefore, this study approaches ANN method for prediction of the temperature of Dhaka exploring the underlying role of related weather variables. Using the daily data for the months of February to July collected from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data (0.25° × 0.25° global grid) for the years 2011-2020, this study focuses on finding the combination of weather variables in predicting temperatures. The densely populated city, Dhaka, has faced severe consequences due to extreme climate conditions in recent years, and this study will pave a new dimension for further research regarding the topic.

Enhancement of ANN-based wind power forecasting by modification of surface roughness parameterization over complex terrain.

Wind energy plays an important role in the sustainable energy transition towards a low-carbon society. Proper assessment of wind energy resources and accurate wind energy prediction are essential prerequisites for balancing electricity supply and demand. However, these remain challenging, especially for onshore wind farms over complex terrains, owing to the interplay between surface heterogeneities and intermittent turbulent flows in the planetary boundary layer. This study aimed to improve wind characteristic assessment and medium-term wind power forecasts over complex hilly terrain using a numerical weather prediction (NWP) model. The NWP model reproduced the wind speed distribution, duration, and spatio-temporal variabilities of the observed hub-height wind speed at 24 wind turbines in onshore wind farms when incorporating more realistic surface roughness effects, such as the subgrid-scale topography, roughness sublayer, and canopy height. This study also emphasizes the good features for machine learning that represent heterogeneities in the surface roughness elements in the atmospheric model. We showed that medium-term forecasting using the NWP model output and a simple artificial neural network (ANN) improved day-ahead wind power forecasts by 14% in terms of annual normalized mean absolute error. Our results suggest that better parameterizations of surface friction in atmospheric models are important for wind power forecasting and resource assessment using NWP models, especially when combined with machine learning techniques, and shed light on onshore wind power forecasting and wind energy assessment in mountainous regions.

Rare bird forecast: A combined approach using a long-term dataset of an Arctic seabird and a numerical weather prediction model.

Wildlife observation is a popular activity, and sightings of rare or difficult-to-find animals are often highly desired. However, predicting the sighting probabilities of these animals is a challenge for many observers, and it may only be possible by limited experts with intimate knowledge and skills. To tackle this difficulty, we developed user-friendly forecast systems of the daily observation probabilities of a rare Arctic seabird (Ross's Gull Rhodostethia rosea) in a coastal area in northern Japan. Using a dataset gathered during 16 successive winters, we applied a machine learning technique of self-organizing maps and explored how days with gull sightings were related to the meteorological pressure patterns over the Sea of Okhotsk (Method A). We also built a regression model that explains the relationship between gull sightings and local-scale environmental factors (Method B). We then applied these methods with the operational global numerical weather prediction model (a computer simulation application about the fluid dynamics of Earth's atmosphere) to forecast the daily observation probabilities of our target. Method A demonstrated a strong dependence of gull sightings on the 16 representative weather patterns and forecasted stepwise observation probabilities ranging from 0% to 85.7%. Method B also showed that the strength of the northerly wind and the advancement of the season explained gull sightings and forecasted continuous observation probabilities ranging from 0% to 95.5%. Applying these two methods with the operational global numerical weather prediction model successfully forecasted the varied observation probabilities of Ross's Gull from 1 to 5 days ahead from November to February. A 2-year follow-up observation also validated both forecast systems to be effective for successful observation, especially when both systems forecasted higher observation probabilities. The developed forecast systems would therefore allow cost-effective animal observation and may facilitate a better experience for a variety of wildlife observers.

ZPD Retrieval Performances of the First Operational Ship-Based Network of GNSS Receivers over the North-West Mediterranean Sea.

This work presents the design and implementation of an operational infrastructure for the monitoring of atmospheric parameters at sea through GNSS meteorology sensors installed on liners operating in the north-west Mediterranean Sea. A measurement system, capable of operationally and continuously providing the values of surface parameters, is implemented together with software procedures based on a float-PPP approach for estimating zenith path delay (ZPD) values. The values continuously registered over a three year period (2020-2022) from this infrastructure are compared with the data from a numerical meteorological reanalysis model (MERRA-2). The results clearly prove the ability of the system to estimate the ZPD from ship-based GNSS-meteo equipment, with the accuracy evaluated in terms of correlation and root mean square error reaching values between 0.94 and 0.65 and between 18.4 and 42.9 mm, these extreme values being from the best and worst performing installations, respectively. This offers a new perspective on the operational exploitation of GNSS signals over sea areas in climate and operational meteorological applications.

Revealing causality in the associations between meteorological variables and air pollutant concentrations.

Understanding the role of meteorology in determining air pollutant concentrations is an important goal for better comprehension of air pollution dispersion and fate. It requires estimating the strength of the causal associations between all the relevant meteorological variables and the pollutant concentrations. Unfortunately, many of the meteorological variables are not routinely observed. Furthermore, the common analysis methods cannot establish causality. Here we use the output of a numerical weather prediction model as a proxy for real meteorological data, and study the causal relationships between a large suite of its meteorological variables, including some rarely observed ones, and the corresponding nitrogen dioxide (NO2) concentrations at multiple observation locations. Time-lagged convergent cross mapping analysis is used to ascertain causality and its strength, and the Pearson and Spearman correlations are used to study the direction of the associations. The solar radiation, temperature lapse rate, boundary layer height, horizontal wind speed and wind shear were found to be causally associated with the NO2 concentrations, with mean time lags of their maximal impact at -3, -1, -2 and -3 hours, respectively. The nature of the association with the vertical wind speed was found to be uncertain and region-dependent. No causal association was found with relative humidity, temperature and precipitation.

Impaired implicit learning in adults with anorexia nervosa.

OBJECTIVE: Cognitive alterations play an important role in the pathophysiology and treatment of anorexia nervosa (AN). Previous studies suggest that some implicit learning processes may be inhibited in AN. However, this has not yet been fully explored. The purpose of this study is to analyze implicit learning in patients with AN in comparison to healthy controls. METHODS: In this pilot-study, a total of 21 patients diagnosed with AN and 21 matched controls were administered the weather prediction task (WPT), a probabilistic implicit category learning task that consists of two sub-variants. During the feedback (FB) version of the task, participants learn associations between tarot cards and weather outcomes via an operant learning model through which they receive immediate FB on their answers, whereas during the paired associate (PA) variant, participants are directly asked to memorize given associations. RESULTS: AN patients showed selective impairment on the FB task where they scored significantly lower both in comparison to controls (p = .001) who completed the same task and when compared to their own performance on the PA variant (p = .006). Clinical measures showed no significant correlations with test scores. DISCUSSION: Our results demonstrate implicit FB learning deficiencies in adult patients with AN. These impairments may have an impact on the effect of psychotherapeutic interventions and could partially explain the lack of treatment response in AN. Further studies are necessary to derive when and through which mechanisms these alterations originate, and to what extent they should be considered during treatment of the disorder. PUBLIC SIGNIFICANCE: Cognitive impairments pose a challenge in the management of anorexia nervosa. Improved comprehension of cognitive alterations could lead to a greater understanding of the disease and adaptation of psychotherapeutic treatments. In this study, we found that implicit feedback learning in anorexia nervosa is impaired compared to healthy controls. This could indicate the necessity of treatment adaptations in the form of therapy tools without feedback and a larger focus on psychoeducation.

The Visual Advantage Effect in Comparing Uni-Modal and Cross-Modal Probabilistic Category Learning.

People rely on multiple learning systems to complete weather prediction (WP) tasks with visual cues. However, how people perform in audio and audiovisual modalities remains elusive. The present research investigated how the cue modality influences performance in probabilistic category learning and conscious awareness about the category knowledge acquired. A modified weather prediction task was adopted, in which the cues included two dimensions from visual, auditory, or audiovisual modalities. The results of all three experiments revealed better performances in the visual modality relative to the audio and audiovisual modalities. Moreover, participants primarily acquired unconscious knowledge in the audio and audiovisual modalities, while conscious knowledge was acquired in the visual modality. Interestingly, factors such as the amount of training, the complexity of visual stimuli, and the number of objects to which the two cues belonged influenced the amount of conscious knowledge acquired but did not change the visual advantage effect. These findings suggest that individuals can learn probabilistic cues and category associations across different modalities, but a robust visual advantage persists. Specifically, visual associations can be learned more effectively, and are more likely to become conscious. The possible causes and implications of these effects are discussed.

A New Approach to Predict Tributary Phosphorus Loads Using Machine Learning- and Physics-Based Modeling Systems.

Tributary phosphorus (P) loads are one of the main drivers of eutrophication problems in freshwater lakes. Being able to predict P loads can aid in understanding subsequent load patterns and elucidate potential degraded water quality conditions in downstream surface waters. We demonstrate the development and performance of an integrated multimedia modeling system that uses machine learning (ML) to assess and predict monthly total P (TP) and dissolved reactive P (DRP) loads. Meteorological variables from the Weather Research and Forecasting (WRF) Model, hydrologic variables from the Variable Infiltration Capacity model, and agricultural management practice variables from the Environmental Policy Integrated Climate agroecosystem model are utilized to train the ML models to predict P loads. Our study presents a new modeling methodology using as testbeds the Maumee, Sandusky, Portage, and Raisin watersheds, which discharge into Lake Erie and contribute to significant P loads to the lake. Two models were built, one for TP loads using 10 environmental variables and one for DRP loads using nine environmental variables. Both models ranked streamflow as the most important predictive variable. In comparison with observations, TP and DRP loads were predicted very well temporally and spatially. Modeling results of TP loads are within the ranges of those obtained from other studies and on some occasions more accurate. Modeling results of DRP loads exceed performance measures from other studies. We explore the ability of both ML-based models to further improve as more data become available over time. This integrated multimedia approach is recommended for studying other freshwater systems and water quality variables using available decadal data from physics-based model simulations.

Informer-Based Temperature Prediction Using Observed and Numerical Weather Prediction Data.

This paper proposes an Informer-based temperature prediction model to leverage data from an automatic weather station (AWS) and a local data assimilation and prediction system (LDAPS), where the Informer as a variant of a Transformer was developed to better deal with time series data. Recently, deep-learning-based temperature prediction models have been proposed, demonstrating successful performances, such as conventional neural network (CNN)-based models, bi-directional long short-term memory (BLSTM)-based models, and a combination of both neural networks, CNN-BLSTM. However, these models have encountered issues due to the lack of time data integration during the training phase, which also lead to the persistence of a long-term dependency problem in the LSTM models. These limitations have culminated in a performance deterioration when the prediction time length was extended. To overcome these issues, the proposed model first incorporates time-periodic information into the learning process by generating time-periodic information and inputting it into the model. Second, the proposed model replaces the LSTM with an Informer as an alternative to mitigating the long-term dependency problem. Third, a series of fusion operations between AWS and LDAPS data are executed to examine the effect of each dataset on the temperature prediction performance. The performance of the proposed temperature prediction model is evaluated via objective measures, including the root-mean-square error (RMSE) and mean absolute error (MAE) over different timeframes, ranging from 6 to 336 h. The experiments showed that the proposed model relatively reduced the average RMSE and MAE by 0.25 °C and 0.203 °C, respectively, compared with the results of the CNN-BLSTM-based model.

Pedestrian Positioning Using an Enhanced Ensemble Transform Kalman Filter.

Due to the unavailability of GPS indoors, various indoor pedestrian positioning approaches have been designed to estimate the position of the user leveraging sensory data measured from inertial measurement units (IMUs) and wireless signal receivers, such as pedestrian dead reckoning (PDR) and received signal strength (RSS) fingerprinting. This study is similar to the previous study in that it estimates the user position by fusing noisy positional information obtained from the PDR and RSS fingerprinting using the Bayes filter in the indoor pedestrian positioning system. However, this study differs from the previous study in that it uses an enhanced state estimation approach based on the ensemble transform Kalman filter (ETKF), called QETKF, as the Bayes filer for the indoor pedestrian positioning instead of the SKPF proposed in the previous study. The QETKF estimates the updated user position by fusing the predicted position by the PDR and the positional measurement estimated by the RSS fingerprinting scheme using the ensemble transformation, whereas the SKPF calculates the updated user position by fusing them using both the unscented transformation (UT) of UKF and the weighting method of PF. In the field of Earth science, the ETKF has been widely used to estimate the state of the atmospheric and ocean models. However, the ETKF algorithm does not consider the model error in the state prediction model; that is, it assumes a perfect model without any model errors. Hence, the error covariance estimated by the ETKF can be systematically underestimated, thereby yielding inaccurate state estimation results due to underweighted observations. The QETKF proposed in this paper is an efficient approach to implementing the ETKF applied to the indoor pedestrian localization system that should consider the model error. Unlike the ETKF, the QETKF can avoid the systematic underestimation of the error covariance by considering the model error in the state prediction model. The main goal of this study is to investigate the feasibility of the pedestrian position estimation for the QETKF in the indoor localization system that uses the PDR and RSS fingerprinting. Pedestrian positioning experiments performed using the indoor localization system implemented on the smartphone in a campus building show that the QETKF can offer more accurate positioning results than the ETKF and other ensemble-based Kalman filters (EBKFs). This indicates that the QETKF has great potential in performing better position estimation with more accurately estimated error covariances for the indoor pedestrian localization system.

The Potential Benefits of Handling Mixture Statistics via a Bi-Gaussian EnKF: Tests With All-Sky Satellite Infrared Radiances.

The meteorological characteristics of cloudy atmospheric columns can be very different from their clear counterparts. Thus, when a forecast ensemble is uncertain about the presence/absence of clouds at a specific atmospheric column (i.e., some members are clear while others are cloudy), that column's ensemble statistics will contain a mixture of clear and cloudy statistics. Such mixtures are inconsistent with the ensemble data assimilation algorithms currently used in numerical weather prediction. Hence, ensemble data assimilation algorithms that can handle such mixtures can potentially outperform currently used algorithms. In this study, we demonstrate the potential benefits of addressing such mixtures through a bi-Gaussian extension of the ensemble Kalman filter (BGEnKF). The BGEnKF is compared against the commonly used ensemble Kalman filter (EnKF) using perfect model observing system simulated experiments (OSSEs) with a realistic weather model (the Weather Research and Forecast model). Synthetic all-sky infrared radiance observations are assimilated in this study. In these OSSEs, the BGEnKF outperforms the EnKF in terms of the horizontal wind components, temperature, specific humidity, and simulated upper tropospheric water vapor channel infrared brightness temperatures. This study is one of the first to demonstrate the potential of a Gaussian mixture model EnKF with a realistic weather model. Our results thus motivate future research toward improving numerical Earth system predictions though explicitly handling mixture statistics.

Wet Bulb Globe Temperature: Indicating Extreme Heat Risk on a Global Grid.

The Wet Bulb Globe Temperature (WBGT) is an international standard heat index used by the health, industrial, sports, and climate sectors to assess thermal comfort during heat extremes. Observations of its components, the globe and the wet bulb temperature (WBT), are however sparse. Therefore WBGT is difficult to derive, making it common to rely on approximations, such as the ones developed by Liljegren et al. (2008, https://doi.org/10.1080/15459620802310770, W B G T L i l j e g r e n ) and by the American College of Sports Medicine ( W B G T A C S M 87 ). In this study, a global data set is created by implementing an updated WBGT method using ECMWF ERA5 gridded meteorological variables and is evaluated against existing WBGT methods. The new method, W B G T B r i m i c o m b e , uses globe temperature calculated using mean radiant temperature and is found to be accurate in comparison to W B G T L i l j e g r e n across three heatwave case studies. In addition, it is found that W B G T A C S M 87 is not an adequate approximation of WBGT. Our new method is a candidate for a global forecasting early warning system.

Upper-air meteorological dataset for Uyo, using radiosonde.

Weather pattern anomalies and climate change have greatly impacted human activities and the environment in varying ways. Whether induced naturally or by anthropogenic activities, it remains a menace to global public health. A foreknowledge of the weather/climate change can help in mitigating the impact of disasters emanating from these changes. Upper-air meteorological data play an exceptionally large role in weather and climate prediction. However, there is a paucity of ground truth meteorological data in Nigeria and many parts of Africa. Consequently, the need to measure and archive these data. Internet of things and blockchain technologies are employed to build a system that captures and records meteorological data at up to 9,000 metres above sea level. Spanning between January 18, 2021 and July 26, 2021, in Uyo local government area, upper air pressure, temperature, dew point, time and the elevation at which they were captured, are the meteorological data presented in this data article.

Data-driven techniques for temperature data prediction: big data analytics approach.

For extrapolation, climate change and other meteorological analysis, a study of past and current weather events is a prerequisite. NASA (National Aeronautics and Space Administration) has been able to develop a model capable of predicting various weather data for any location on the Earth, including locations lacking weather stations, weather satellite coverage, and other weather measuring instruments. This paper evaluates the prediction accuracy of the NASA temperature data with respect to NiMet (Nigerian Meteorological Agency) ground truth measurement, using Akwa Ibom Airport as a case study. Exploratory data analysis (descriptive and diagnostic analyses) of temperature retrieved from NiMet and NASA was performed to give a clear path to follow for predictive and prescriptive analyses. Using 2783 days of weather data retrieved from NiMet as ground truth, the accuracy of NASA predictions with the corresponding resolution was calculated. Mean absolute error (MAE) of 2.184 °C and root mean square error (RMSE) of 2.579 °C, with a coefficient of determination (R2) of 0.710 for maximum temperature, then MAE of 0.876 °C, RMSE of 1.225 °C with a coefficient of determination (R2) of 0.620 for minimum temperature was discovered. There is a good correlation between the two datasets; hence, a model can be developed to generate more accurate predictions, using the NASA data as input. Predictive and prescriptive analyses were performed by employing five prediction algorithms: decision tree regression, XGBoost regression and MLP (multilayer perceptron) with LBFGS (limited-memory Broyden-Fletcher-Goldfarb-Shanno) optimizer, MLP with SGD (stochastic gradient) optimizer and MLP with Adam optimizer. The MLP LBFGS algorithm performed best, by significantly reducing the MAE by 35.35% and RMSE by 31.06% for maximum temperature, accordingly, MAE by 10.05% and RMSE by 8.00% for minimum temperature. Results obtained show that given sufficient data, plugging NASA predictions as input to an LBFGS-MLP model gives more accurate temperature predictions for the study area.

Testing Variational Bias Correction of Satellite Radiance Data in the ACCESS-C: Australian Convective-Scale NWP System.

Radiance observations are typically affected by biases that come mainly from instrument error (scanning or calibration) and inaccuracies of the radiative transfer model. These biases need to be removed for successful assimilation, so a bias correction scheme is crucial in the Numerical Weather Prediction (NWP) system. Today, most NWP centres, including the Bureau of Meteorology (hereafter, "the Bureau"), correct the biases through variational bias correction (VarBC) schemes, which were originally developed for global models. However, there are difficulties in estimating the biases in a limited-area model (LAM) domain. As a result, the Bureau's regional NWP system, ACCESS-C (Australian Community Climate and Earth System Simulator-City), uses variational bias coefficients obtained directly from its global NWP system ACCESS-G (Global). This study investigates independent radiance bias correction in the data assimilation system for ACCESS-C. We assessed the impact of using independent bias correction for the LAM compared with the operational bias coefficients derived in ACCESS-G between February and April 2020. The results from our experiment show no significant difference between the control and test, suggesting a neutral impact on the forecast. Our findings point out that the VarBC-LAM strategy should be further explored with different settings of predictors and adaptivity for a more extended period and over additional domains.

Machine-Learning-Based Near-Surface Ozone Forecasting Model with Planetary Boundary Layer Information.

Surface ozone is one of six air pollutants designated as harmful by National Ambient Air Quality Standards because it can adversely impact human health and the environment. Thus, ozone forecasting is a critical task that can help people avoid dangerously high ozone concentrations. Conventional numerical approaches, as well as data-driven forecasting approaches, have been studied for ozone forecasting. Data-driven forecasting models, in particular, have gained momentum with the introduction of machine learning advancements. We consider planetary boundary layer (PBL) height as a new input feature for data-driven ozone forecasting models. PBL has been shown to impact ozone concentrations, making it an important factor in ozone forecasts. In this paper, we investigate the effectiveness of utilization of PBL height on the performance of surface ozone forecasts. We present both surface ozone forecasting models, based on multilayer perceptron (MLP) and bidirectional long short-term memory (LSTM) models. These two models forecast hourly ozone concentrations for an upcoming 24-h period using two types of input data, such as measurement data and PBL height. We consider the predicted values of PBL height obtained from the weather research and forecasting (WRF) model, since it is difficult to gather actual PBL measurements. We evaluate two ozone forecasting models in terms of index of agreement (IOA), mean absolute error (MAE), and root mean square error (RMSE). Results showed that the MLP-based and bidirectional LSTM-based models yielded lower MAE and RMSE when considering forecasted PBL height, but there was no significant changes in IOA when compared with models in which no forecasted PBL data were used. This result suggests that utilizing forecasted PBL height can improve the forecasting performance of data-driven prediction models for surface ozone concentrations.

A canary, a coal mine, and imperfect data: determining the efficacy of open-source climate change models in detecting and predicting extreme weather events in Northern and Western Kenya.

Climate models, by accurately forecasting future weather events, can be a critical tool in developing countermeasures to reduce crop loss and decrease adverse effects on animal husbandry and fishing. In this paper, we investigate the efficacy of various regional versions of the climate models, RCMs, and the commonly available weather datasets in Kenya in predicting extreme weather patterns in northern and western Kenya. We identified two models that may be used to predict flood risks and potential drought events in these regions. The combination of artificial neural networks (ANNs) and weather station data was the most effective in predicting future drought occurrences in Turkana and Wajir with accuracies ranging from 78 to 90%. In the case of flood forecasting, isolation forests models using weather station data had the best overall performance. The above models and datasets may form the basis of an early warning system for use in Kenya's agricultural sector.

Recent advancements for tropical cyclone data assimilation.

In this review, data assimilation (DA) techniques used for tropical cyclones (TCs) are briefly overviewed. The strength and weakness of variational methods, ensemble methods, hybrid methods, and particle filter methods are also discussed. Several global numerical weather prediction models and their corresponding DA systems frequently used for TC forecasting and verification are described first. The DA research and development efforts in the operational regional model from the National Centers for Environmental Prediction's Hurricane Weather Research and Forecasting are then discussed in greater detail. Focused remarks on TC observations from reconnaissance, ground-based radar, enhanced satellite-derived atmospheric motion vectors and all-sky satellite radiances and their impacts on TC analyses and forecasts are addressed. Recent TC DA advancements and challenges on better use of observations and more advanced DA methods for TC application are also briefly reviewed.

Spatial variability of forward modelled attenuated backscatter in clear-sky conditions over a megacity: Implications for observation network design.

Sensors that measure the attenuated backscatter coefficient (e.g., automatic lidars and ceilometers [ALCs]) provide information on aerosols that can impact urban climate and human health. To design an observational network of ALC sensors for supporting data assimilation and to improve prediction of urban weather and air quality, a methodology is needed. In this study, spatio-temporal patterns of aerosol-attenuated backscatter coefficient are modelled using Met Office numerical weather prediction (NWP) models at two resolutions, 1.5 km (UKV) and 300 m (London Model [LM]), for 28 clear-sky days and nights. Initially, attenuated backscatter coefficient data are analysed using S-mode principal component analysis (PCA) with varimax rotation. Four to seven empirical orthogonal functions (EOFs) are produced for each model level, with common EOFs found across different heights (day and night) for both NWP models. EOFs relate strongly to orography, wind, and emissions source location, highlighting these as critical controls of attenuated backscatter coefficient spatial variability across the megacity. Urban-rural differences are largest when wind speeds are low and vertical boundary-layer dynamics can more effectively distribute near-surface aerosol emissions vertically. In several night-time EOFs, gravity-wave features are found for both NWP models. Increasing the horizontal resolution of native ancillaries (model input parameters) and improving the urban surface scheme in the LM may enhance the urban signal in the EOFs. PCA output, with agglomerative Ward cluster analysis (CA), minimises intra-group variance. The UKV and LM CA shape and size results are similar and strongly related to orography. PCA-CA is a simple, but adaptable methodology, allowing close alignment with observation network design goals. Here, CA is used with wind roses to suggest the optimised ALC deployment is one in the city to observe the urban plume and others surrounding the city, with priority given to cluster size and frequency of upwind advection.

A Framework for Four-Dimensional Variational Data Assimilation Based on Machine Learning.

The initial field has a crucial influence on numerical weather prediction (NWP). Data assimilation (DA) is a reliable method to obtain the initial field of the forecast model. At the same time, data are the carriers of information. Observational data are a concrete representation of information. DA is also the process of sorting observation data, during which entropy gradually decreases. Four-dimensional variational assimilation (4D-Var) is the most popular approach. However, due to the complexity of the physical model, the tangent linear and adjoint models, and other processes, the realization of a 4D-Var system is complicated, and the computational efficiency is expensive. Machine learning (ML) is a method of gaining simulation results by training a large amount of data. It achieves remarkable success in various applications, and operational NWP and DA are no exception. In this work, we synthesize insights and techniques from previous studies to design a pure data-driven 4D-Var implementation framework named ML-4DVAR based on the bilinear neural network (BNN). The framework replaces the traditional physical model with the BNN model for prediction. Moreover, it directly makes use of the ML model obtained from the simulation data to implement the primary process of 4D-Var, including the realization of the short-term forecast process and the tangent linear and adjoint models. We test a strong-constraint 4D-Var system with the Lorenz-96 model, and we compared the traditional 4D-Var system with ML-4DVAR. The experimental results demonstrate that the ML-4DVAR framework can achieve better assimilation results and significantly improve computational efficiency.