Operational Forecasting System Research Articles

Operational Forest-Fire Spread Forecasting Using the WRF-SFIRE Model

In the present research, the open-source WRF-SFIRE model has been used to carry out surface forest fire spread forecasting in the North Sikkim region of the Indian Himalayas. Global forecast system (GFS)-based hourly forecasted weather model data obtained through the National Centers for Environmental Prediction (NCEP) at 0.25 degree resolution were used to provide the initial conditions for running WRF-SFIRE. A landuse–landcover map at 1:10,000 scale was used to define fuel parameters for different vegetation types. The fuel parameters, i.e., fuel depth and fuel load, were collected from 23 sample plots (0.1 ha each) laid down in the study area. Samples of different categories of forest fuels were measured for their wet and dry weights to obtain the fuel load. The vegetation specific surface area-to-volume ratio was referenced from the literature. The atmospheric data were downscaled using nested domains in the WRF model to capture fire–atmosphere interactions at a finer resolution (40 m). VIIRS satellite sensor-based fire alert (375 m spatial resolution) was used as ignition initiation point for the fire spread forecasting, whereas the forecasted hourly weather data (time synchronized with the fire alert) were used for dynamic forest-fire spread forecasting. The forecasted burnt area (1.72 km2) was validated against the satellite-based burnt area (1.07 km2) obtained through Sentinel 2 satellite data. The shapes of the original and forecasted burnt areas matched well. Based on the various simulation studies conducted, an operational fire spread forecasting system, i.e., Sikkim Wildfire Forecasting and Monitoring System (SWFMS), has been developed to facilitate firefighting agencies to issue early warnings and carry out strategic firefighting.

Combining neural networks and data assimilation to enhance the spatial impact of Argo floats in the Copernicus Mediterranean biogeochemical model

Abstract. Biogeochemical-Argo (BGC-Argo) float profiles provide substantial information on key vertical biogeochemical dynamics and have been successfully integrated in biogeochemical models via data assimilation approaches. Although BGC-Argo assimilation results have been encouraging, data scarcity remains a limitation with respect to their effective use in operational oceanography. To address availability gaps in the BGC-Argo profiles, an observing system experiment (OSE) that combines a neural network (NN) and data assimilation (DA) was performed here. A NN was used to reconstruct nitrate profiles, starting from oxygen profiles and associated Argo variables (pressure, temperature, and salinity), while a variational data assimilation scheme (3DVarBio) was upgraded to integrate BGC-Argo and reconstructed observations in the Copernicus Mediterranean operational forecast system (MedBFM). To ensure the high quality of oxygen data, a post-deployment quality control method was developed with the aim of detecting and eventually correcting potential sensors drift. The Mediterranean OSE features three different set-ups: a control run without assimilation; a multivariate run with assimilation of BGC-Argo chlorophyll, nitrate, and oxygen; and a multivariate run that also assimilates reconstructed observations. The general improvement in the skill performance metrics demonstrated the feasibility of integrating new variables (oxygen and reconstructed nitrate). Major benefits have been observed with respect to reproducing specific biogeochemical-process-based dynamics such as the nitracline dynamics, primary production, and oxygen vertical dynamics. The assimilation of BGC-Argo nitrate corrects a generally positive bias of the model in most of the Mediterranean areas, and the addition of reconstructed profiles makes the corrections even stronger. The impact of enlarged nitrate assimilation propagates to ecosystem processes (e.g. primary production) at a basin-wide scale, demonstrating the importance of the assimilation of BGC-Argo profiles in forecasting the biogeochemical ocean state.

Evaluation of WRF model configurations for dynamic downscaling of tropical cyclones activity over the North Atlantic basin for Lagrangian moisture tracking analysis in future climate

This study assessed five well-established physics suites of the Weather Research and Forecasting (WRF) model in operational forecasting systems in the North Atlantic (NATL) basin or from previous sensitive experiments for dynamic downscaling tropical cyclone (TC) activity. We performed long-term simulations for the 2020 TC season in the NATL and compared the WRF tracks against the HURDAT2 dataset from the US National Hurricane Center. Among the tested configurations, the analysis revealed that the Kain-Fritsch, Purdue Lin, BouLac and revised MM5 schemes for cumulus, microphysics, planetary boundary layer and surface layer, respectively (hereafter WT), outperformed all four others in terms of TC frequency, track density and intensity and showed good performance in the cyclone accumulated energy and TC landfalling locations. In addition, WT well-captured the spatial distribution of accumulated TC precipitation and moisture uptake patterns, although it overestimated the precipitation maxima. Likewise, it agreed on the relative moisture contribution from fixed moisture sources (i.e., Gulf of Mexico, Caribbean Sea, tropical NATL, and western NATL) in the NATL basin. Overall, this study highlighted the high potential of using the WT physics suite in WRF for downscaling TC activity over the NATL basin, which will be useful for TC Lagrangian moisture sources analysis in future climate.

Skill assessment of seasonal forecasts of ocean variables

There is growing demand for seasonal forecast products for marine applications. The availability of consistent and sufficiently long observational records of ocean variables permits the assessment of the spatial distribution of the skill of ocean variables from seasonal forecasts. Here we use state-of-the-art temporal records of sea surface temperature (SST), sea surface height (SSH) and upper 300m ocean heat content (OHC) to quantify the distribution of skill, up to 2 seasons ahead, of two operational seasonal forecasting systems contributing to the seasonal multi-model of the Copernicus Climate Change Services (C3S). This study presents the spatial distribution of the skill of the seasonal forecast ensemble mean in terms of anomaly correlation and root mean square error and compares it to the persistence and climatological benchmarks. The comparative assessment of the skill among variables sheds light on sources/limits of predictability at seasonal time scales, as well as the nature of model errors. Beyond these standard verification metrics, we also evaluate the ability of the models to represent the observed long-term trends. Results show that long-term trends contribute to the skill of seasonal forecasts. Although the forecasts capture the long-term trends in general, some regional aspects remain challenging. Part of these errors can be attributed to specific aspects of the ocean initialization, but others, such as the overestimation of the warming in the Eastern Pacific are also influenced by model error. Skill gains can be obtained by improving the trend representation in future forecasting systems. In the meantime, a forecast calibration procedure that corrects the linear trends can produce substantial skill gains. The results show that calibrated seasonal forecasts beat both the climatological and persistence benchmark almost at every location for all initial dates and lead times. Results demonstrate the value of the seasonal forecasts for marine applications and highlight the importance of representing the decadal variability and trends in ocean heat content and sea level.

A flexible tropical cyclone vortex initialization technique for GFDL SHiELD

Tropical cyclone (TC) intensity forecasting poses challenges due to complex dynamical processes and data inadequacies during model initialization. This paper describes efforts to improve TC intensity prediction in the Geophysical Fluid Dynamics Laboratory (GFDL) System for High-resolution prediction on Earth-to-Local Domains (SHiELD) model by implementing a Vortex Initialization (VI) technique. The GFDL SHiELD model, relying on the Global Forecast System (GFS) analysis for initialization, faces deficiencies in initial TC structure and intensity. The VI method involves adjusting the TC vortex inherited from the GFS analysis and merging it back into the environment at the observed location, enhancing the analyzed representation of storm structure. We made modifications to the VI package implemented in the operational Hurricane Analysis and Forecast System, including handling initial condition data, reducing input domain size, and improving storm intensity enhancement. Experiments using the T-SHiELD configuration demonstrate that using VI significantly improves the representation of initial TC intensity and size, enhancing TC predictions, particularly in storm intensity and outer wind forecasts within the first 48 h.

An Interpretable Transformer Model for Operational Flare Forecasting

Interpretable machine learning tools including LIME (Local Interpretable Model-agnostic Explanations) and ALE (Accumulated Local Effects) are incorporated into a transformer-based deep learning model, named SolarFlareNet, to interpret the predictions made by the model. SolarFlareNet is implemented into an operational flare forecasting system to predict whether an active region on the surface of the Sun would produce a >=M class flare within the next 24 hours. LIME determines the ranking of the features used by SolarFlareNet. 2D ALE plots identify the interaction effects of two features on the predictions. Together, these tools help scientists better understand which features are crucial for SolarFlareNet to make its predictions. Experiments show that the tools can explain the internal workings of SolarFlareNet while maintaining its accuracy.

Physics schemes in the first version of NCEP operational hurricane analysis and forecast system (HAFS)

This document summarizes the physics schemes used in two configurations of the first version of the operational Hurricane Analysis and Forecast System (HAFSv1) at NOAA NCEP. The physics package in HAFSv1 is the same as that used in NCEP global forecast system (GFS) version 16 except for an additional microphysics scheme and modifications to sea surface roughness lengths, boundary layer scheme, and the entrainment rate in the deep convection scheme. Those modifications are specifically designed for improving the simulation of tropical cyclones (TCs). The two configurations of HAFSv1 mainly differ in the adopted microphysics schemes and TC-specific modifications in addition to model initialization. Experiments are made to highlight the impacts of TC-specific modifications and different microphysics schemes on HAFSv1 performance. Challenges and developmental plans of physics schemes for future versions of operational HAFS are discussed.

An ensemble-based data assimilation system for forecasting variability of the Northwestern Pacific ocean

An adjoint-free four-dimensional variational (a4dVar) data assimilation (DA) is implemented in an operational ocean forecast system based on an eddy-resolving ocean general circulation model for the Northwestern Pacific. Validation of the system against independent observations demonstrates that fitting the model to time-dependent satellite altimetry during a 10-day DA window leads to substantial skill improvements in the succeeding 60-day hindcast. The a4dVar corrects representation of the Kuroshio path variation south of Japan by adjusting the dynamical balance between amplitude/wavelength of the meander and flow advection. A larger ensemble spread tends to reduce the skill in representing the observed sea surface height anomaly, suggesting that it is possible to use the ensemble information for quantifying the forecast error. The ensemble information is also utilized for modification of the background error covariance (BEC), which improves the accuracy of temperature and salinity distributions. The modified BEC yields the skill decline of the Kuroshio path variation during the 60-day hindcast period, and the ensemble sensitivity analysis shows that changes in the dynamical balance caused by the ensemble BEC result in such skill deterioration.

Deep Learning of a 200-member Ensemble with a Limited Historical Training to Improve the Prediction of Extreme Precipitation Events

Abstract This study introduces a deep learning (DL) scheme to generate reliable and skillful probabilistic quantitative precipitation forecasts (PQPFs) in a postprocessing framework. Enhanced machine learning model architecture and training mechanisms are proposed to improve reliability and skill of PQPFs while permitting computationally efficient model fitting using a short training dataset. The methodology is applied to postprocessing of 24h accumulated PQPFs from an ensemble forecast system recently introduced by the Center for Western Weather and Water Extremes (CW3E) and for lead times from one to six days. The ensemble system was designed based on a high-resolution version of the Weather Research and Forecasting (WRF) model, named West-WRF, to produce a 200-member ensemble in near real-time (NRT) over the Western U.S. during the boreal cool seasons to support forecastinformed reservoir operations (FIRO) and studies of prediction of heavy-to-extreme events. Postprocessed PQPFs are compared with those from the raw West-WRF ensemble, the operational Global Ensemble Forecast System (GEFSv12) as well as the ensemble from the European Centre for Medium-Range Weather Forecasts (ECMWF). As an additional baseline, we provide PQPFs verification metrics from a recently developed neural network postprocessing scheme proposed by Ghazvinian et al. (2022). Results demonstrate that skill of postprocessed forecasts significantly outperform PQPFs and deterministic forecasts from raw ensembles and the recently developed algorithm. The resulting PQPFs broadly improve upon the reliability and skill of baselines in predicting heavy to extreme precipitation (e.g., > 75 mm) across all lead times while maintaining the spatial structure of the high-resolution raw ensemble.

Application of Radar-Based Precipitation Data Improves the Effectiveness of Urban Inundation Forecasting

Using radar to estimate and forecast precipitation as input for hydrological models has become increasingly popular in recent years because of its superior spatial and temporal simulation compared with using rain gauge data. This study used radar-based quantitative precipitation estimation (QPE) to select the optimal parameter set for the MIKE URBAN hydrological model and radar-based quantitative precipitation forecasting (QPF) to simulate inundation in Nam Dinh city, Vietnam. The results show the following: (1) radar has the potential to improve the modeling and provide the data needed for real-time smart control if proper bias adjustment is obtained and the risk of underestimated flows after heavy rain is minimized, and (2) the MIKE URBAN model used to calculate two simulation scenarios with rain gauge data and QPE data showed effectiveness in combining the application of radar-based precipitation for the forecasting and warning of urban floods in Nam Dinh city. The results in Scenario 2 with rainfall forecast data from radar provide better simulation results. The average relative error in Scenario 2 is 9%, while the average relative error in Scenario 1 is 15%. Using the grid radar-based precipitation forecasting as input data for the MIKE URBAN model significantly reduces the error between the observed water depth and the simulated results compared with the case using an input rain gauge measured at Nam Dinh station (the difference in inundation level of Scenario 2 using radar-based precipitation is 0.005 m, and it is 0.03 m in Scenario 1). The results obtained using the QPE and QPF radar as input for the MIKE URBAN model will be the basis for establishing an operational forecasting system for the Northern Delta and Midland Regional Hydro-Meteorological Center, Viet Nam Meteorological and Hydrological Administration.

Generative emulation of weather forecast ensembles with diffusion models.

Uncertainty quantification is crucial to decision-making. A prominent example is probabilistic forecasting in numerical weather prediction. The dominant approach to representing uncertainty in weather forecasting is to generate an ensemble of forecasts by running physics-based simulations under different conditions, which is a computationally costly process. We propose to amortize the computational cost by emulating these forecasts with deep generative diffusion models learned from historical data. The learned models are highly scalable with respect to high-performance computing accelerators and can sample thousands of realistic weather forecasts at low cost. When designed to emulate operational ensemble forecasts, the generated ones are similar to physics-based ensembles in statistical properties and predictive skill. When designed to correct biases present in the operational forecasting system, the generated ensembles show improved probabilistic forecast metrics. They are more reliable and forecast probabilities of extreme weather events more accurately. While we focus on weather forecasting, this methodology may enable creating large climate projection ensembles for climate risk assessment.

An incremental analysis update in the framework of the four‐dimensional variational data assimilation: Description and preliminary tests in the operational China Meteorological Administration Global Forecast System

AbstractInitialization and the spin‐up effect are critical to model performance in the four‐dimensional variational data assimilation (4DVar) system. The incremental analysis update (IAU) technique is able to combat these problems. This study introduces the IAU technique into the 4DVar framework (IAU‐4DVar) of the operational China Meteorological Administration Global Forecast System (CMA‐GFS) and evaluates the performance of the IAU‐4DVar. In this IAU‐4DVar framework, the initial increments optimally estimated are split up into equally weighted forcing terms and integrated into the tangent model and its adjoint model. Then, the optimal estimation of the analysis increment is gradually introduced into the model integration during the assimilation window. A set of two‐month assimilation and forecast experiments were established and executed in this study, and the preliminary results show that the IAU technique can reduce the spin‐up effect of the initial analysis increment from the operational 4DVar system with digital filter initialization (DFI‐4DVar) by retaining the balance in dynamics and thermodynamics. In the optimal minimization process, the IAU‐4DVar converges faster than the DFI‐4DVar with a reduction of the computational cost of up to ca 40%. In addition, the introduction of IAU‐4DVar into the CMA‐GFS model could have a positive impact on the analysis and forecast skills relative to the operational DFI‐4DVar.

Removing the systematic errors of the model in operational oceanography forecasting system using data assimilation method

For the operational oceanography forecast, the synoptic forecast error is partly from the long-term systematic bias of the model, which can be partly counteracted by adjusting the values of the physical parameters. To this end, a four-dimensional optimization system is implemented into the South China Sea operational oceanography forecasting system, to adjust the values of multi-parameters using data assimilation method. By assimilating Argo temperature profiles of 51 days in the model domain, five physical parameters (coefficients of horizontal/vertical diffusion/viscosity and linear bottom drag) of the model have been adjusted simultaneously, and then the optimal values are obtained. The RMSE of temperature simulations in the assimilation window decreases from 1.17 to 0.97 K, when using the optimal values. The validation of the freerun experiments shows that the temperature RMSE decreases from 0.97 to 0.88 K, which indicates that the optimal values are still valid in a longer and independent period. Finally, the validation of the hindcast experiments shows that at the synoptic scale the temperature RMSE decreases from 0.90 to 0.80 K and other variables also present improvements. It hints that it is feasible to reduce the synoptic forecast errors by adjusting the parameter values at the climatological scale to partly counteract the systematic bias of the model. Therefore, it also provides a potential pathway to improve the synoptic forecast skill for the operational oceanography forecasting system.

Predictability of marine heatwaves: assessment based on the ECMWF seasonal forecast system

Abstract. Marine heatwaves (MHWs), defined as prolonged period of extremely warm sea surface temperature (SST), have been receiving a lot of attention in the past decade as their frequency and intensity increase in a warming climate. This paper investigates the extent to which the seasonal occurrence and duration of MHWs can be predicted with the European Centre for Medium-Range Weather Forecast (ECMWF) operational seasonal forecast system. The prediction of the occurrence of MHW events, the number of MHW days per season, and their intensity and spatial extent are derived from seasonal SST forecasts and evaluated against an observation-based SST analysis using both deterministic and probabilistic metrics over the 1982–2021 period. Forecast scores show useful skill in predicting the occurrence of MHWs globally for the two seasons following the starting date. The skill is the highest in the El Niño region, the Caribbean, the wider tropics, the north-eastern extra-tropical Pacific, and southwest of the extra-tropical basins. The skill is not as good for other midlatitude eastern basins nor for the Mediterranean, with the forecast system being able to represent the low-frequency modulation of MHWs but showing poor skill in predicting the interannual variability of the MHW characteristics. Linear trend analysis shows an increase in MHW occurrence at a global scale, which the forecasts capture well.

Evaluating the sensitivity of fine particulate matter (PM2.5) simulations to chemical mechanism in WRF-Chem over Delhi

Accurate prediction of PM2.5, its optical properties and dominant chemical components are essential for air quality studies. In this study, we investigated the effects of two gas phase chemical schemes coupled with three aerosol mechanisms on the simulated PM2.5 mass concentration in Delhi using the Weather Research and Forecasting model with Chemistry module (WRF-Chem). The model was employed to cover the entire northern region of India at 10 km horizontal spacing and results were compared with comprehensive field data set on dominant PM2.5 chemical compounds from the Winter Fog EXperiment (WiFEX) at the Indira-Gandhi International Airport, New Delhi, and surface PM2.5 observations in Delhi (17 sites), Punjab (3 sites), Haryana (4 sites), Uttar Pradesh (7 sites) and Rajasthan (17 sites). The Model for Ozone and related Chemical Tracers (MOZART) gas-phase chemical mechanism coupled with the Goddard Chemistry Aerosol Radiation and Transport (GOCART) aerosol scheme were selected in the first experiment as it is currently employed in the operational air quality forecasting system of Ministry of Earth Sciences (MoES), Government of India. Other two simulations were performed with the MOZART gas phase chemical mechanism coupled with the Model for Simulating Aerosol Interactions and Chemistry (MOZART-MOSAIC), and Carbon Bond 5 (CB-05) gas mechanism coupled with the Modal Aerosol Dynamics Model for Europe/Secondary Organic Aerosol Model (CB05 - MADE/SORGAM) aerosol mechanisms. The evaluation demonstrated that chemical mechanisms affect the evolution of gas-phase precursors and aerosol processes which in turn affect the optical depth and overall performance of the model for PM2.5. All the three chemical schemes underestimate the observed concentrations of major aerosol composition and precursor gases over Delhi. Comparison with observations suggests that, the simulations using MOZART gas-phase chemical mechanism with MOSAIC aerosol scheme performed better in simulating aerosols over Delhi and its optical depth over the IGP.

Intercomparison of Air Quality Models in a Megacity: Toward an Operational Ensemble Forecasting System for São Paulo

AbstractAn intercomparison of four regional air quality models is performed in the tropical megacity of São Paulo with the perspective of developing a forecasting system based on a model ensemble. Modeled concentrations of the main regulated pollutants are compared with combined observations in the megacity center, after analyzing the spatial scale of representativeness of air monitoring stations. During three contrasting periods characterized by different types of pollution events, the hourly concentrations of carbon monoxide (CO), nitrogen oxides (NOx), sulfur dioxide (SO2), and particulate matter (PM2.5 and PM10) modeled by the ensemble are in moderate agreement with observations. The median of the ensemble provides the best performance (R ≈ 0.7 for CO, 0.7 for NOx, 0.5 for SO2, 0.5 for PM2.5, and 0.4 for PM10) because each model has periods and pollutants for which it has the best agreement. NOx concentration is modeled with a large inter‐model variability, highlighting potential for improvement of anthropogenic emissions. Pollutants transported by biomass burning events strongly affect the air quality in São Paulo and are associated with significant inter‐model variability. Modeled hourly concentration of ozone (O3) is overestimated during the day (≈20 ppb) and underestimated at night (≈10 ppb), while nitrogen dioxide (NO2) is overestimated at night (≈20 ppb). The observed O3 concentration is best reproduced by the median of the ensemble (R ≈ 0.8), taking advantage of the variable performance of the models. Therefore, an operational air quality forecast system based on a regional model ensemble is promising for São Paulo.

Predictions of air quality and challenges for eliminating air pollution during the 2022 Olympic Winter Games

Scientifically and efficiently ensuring good air quality for important events is an issue of concern to the government. In addition to analysis based on historical data, advanced prediction before an event is essential for the government having ample time to take effective actions to improve air quality during the event period. Taking “the 2022 Olympic Winter Games (OWG)” as a typical case, a chemical transport model coupled with a tracer-tagged module was used to evaluate the air quality and source apportionment of ambient pollutants in the OWG host cities under historical and predicted meteorological conditions. Driven by the downscaling of meteorological fields from an operational real-time climate forecast system, the potential ability of air quality forecasting three months ahead was investigated, which was meaningful for designing control strategies. Sensitive simulations indicated that under unfavorable meteorological conditions, such as those during February 2014, both Beijing and Zhangjiakou faced a high risk of experiencing haze episodes, even based on current anthropogenic emission intensity. The contribution of the joint prevention and control region to Beijing and Zhangjiakou would become larger under worse meteorological conditions, which favor heavy air pollution. The source apportionment results indicated that strengthened emission control in cities including Beijing, Zhangjiakou and south of Beijing (Baoding, Langfang, Tianjin, and Tangshan) is effective for reducing haze episodes in the host cities. There is still a long way to make accurate daily fine particulate matter predictions on a seasonal-scale in advance; however, it could capture the trends in air quality in host cities around the OWG period three months ahead. The comparison of observations and predictions confirmed and highlighted the role of regional emission controls on the realization of the “OWG blue”.

A case study with extreme low-level jet in southern Brazil

This study aimed to analyze a period of intense Low-Level Jets (LLJ) in southern Brazil that occurred in a year of intense El Nino-South Oscillation signal. Observational data were collected by radiosonde data carried out at Salgado Filho airport, which measured values of the order of 30 ms-1 in intensity at 700/750 meters in height and a predominant direction of 270 degrees. The LLJ’s are horizontal vortices in the low-level turbulence process, optimizing the transport of energy and water vapor at the mesoscale; they are low-cost energy structures with maximum transport efficiency. In general, their modeling by numerical models at high resolution is difficult because they have a thin layer with a small thickness of the order of 50 meters and have a non-linear nature in the turbulence processes. However, global reanalysis models, in general, cannot represent the optimization effects in the turbulent process performed by LLJ. However, there is a tendency to represent the magnitude of the wind modulus, an approximation of this value by the current state of the art in global reanalysis modeling in models, in this case, the maximum wind. This implies that global models generally underestimate flows at low levels, in general, as they do not represent the optimization effects associated with LLJ. In this case study, we tried to use an extreme situation of occurrence of LLJ for analysis, therefore, the regional climate model RegCM-4.9 in high spatial resolution was used to simulate the period from May to June 1997. The analyses were concentrated in June 1997. The initialization and boundary conditions were used from the EC-EARTH historical of the Earth system model based on the operational seasonal forecast system of the European Center for Medium-Range Weather Forecasts (ECMWF). It was observed that the simulation result could significantly underestimate the wind intensity and the processes associated with the LLJ with the initialization model, with low information resolution, the global initialization models do not have adequate resolution for this type of simulation.

A Machine Learning Augmented Data Assimilation Method for High‐Resolution Observations

AbstractThe accuracy of initial conditions is an important driver of the forecast skill of numerical weather prediction models. Increases in the quantity of available measurements, particularly high‐resolution remote sensing observational data products from satellites, are valuable inputs for improving those initial condition estimates. However, the traditional data assimilation methods for integrating observations into forecast models are computationally expensive. This makes incorporating dense observations into operational forecast systems challenging, and it is often prohibitively time‐consuming. Additionally, high‐resolution observations often have correlated observation errors which are difficult to estimate and create problems for assimilation systems. As a result, large quantities of data are discarded and not used for state initialization. Using the Lorenz‐96 system for testing, we demonstrate that a simple machine learning method can be trained to assimilate high‐resolution data. Using it to do so improves both initial conditions and forecast accuracy. Compared to using the Ensemble Kalman Filter with high‐resolution observations ignored, our augmented method has an average root‐mean‐squared error reduced by 37%. Ensemble forecasts using initial conditions generated by the augmented method are more accurate and reliable at up to 10 days of forecast lead time.

Short-lead seasonal precipitation forecast in northeastern Brazil using an ensemble of artificial neural networks

This study assesses the deterministic and probabilistic forecasting skill of a 1-month-lead ensemble of Artificial Neural Networks (EANN) based on low-frequency climate oscillation indices. The predictand is the February-April (FMA) rainfall in the Brazilian state of Ceará, which is a prominent subject in climate forecasting studies due to its high seasonal predictability. Additionally, the study proposes combining the EANN with dynamical models into a hybrid multi-model ensemble (MME). The forecast verification is carried out through a leave-one-out cross-validation based on 40 years of data. The EANN forecasting skill is compared with traditional statistical models and the dynamical models that compose Ceará’s operational seasonal forecasting system. A spatial comparison showed that the EANN was among the models with the smallest Root Mean Squared Error (RMSE) and Ranked Probability Score (RPS) in most regions. Moreover, the analysis of the area-aggregated reliability showed that the EANN is better calibrated than the individual dynamical models and has better resolution than Multinomial Logistic Regression for above-normal (AN) and below-normal (BN) categories. It is also shown that combining the EANN and dynamical models into a hybrid MME reduces the overconfidence of the extreme categories observed in a dynamically-based MME, improving the reliability of the forecasting system.