Operational Forecasting System Research Articles

Snowmelt characterization from optical and synthetic-aperture radar observations in the La Joie Basin, British Columbia

Abstract. Snowmelt runoff serves both human needs and ecosystem services and is an important parameter in operational forecasting systems. Sentinel-1 synthetic-aperture-radar (SAR) observations can estimate the timing of melt within a snowpack; however, these estimates have not been applied on large spatial scales. Here we present a workflow to combine Sentinel-1 SAR and optical data from Landsat-8 and Sentinel-2 to estimate the onset and duration of snowmelt in the La Joie Basin, a 985 km2 watershed in the southern Coast Mountains of British Columbia. A backscatter threshold is used to infer the point at which snowpack saturation occurs and the snowpack begins to produce runoff. Multispectral imagery is used to estimate snow-free dates across the basin to define the end of the snowmelt period. SAR estimates of snowmelt onset form consistent trends in terms of elevation and aspect on the watershed scale and reflect snowmelt records from continuous snow water equivalence observations. SAR estimates of snowpack saturation are most effective on moderate to low slopes (< 30∘) in open areas. The accuracy of snowmelt duration is reduced due to persistent cloud cover in optical imagery. Despite these challenges, snowmelt duration agrees with trends in snow depths observed in the La Joie Basin. This approach has high potential for adaptability to other alpine regions and can provide estimates of snowmelt timing in ungauged basins.

Improved forecast of 2015/16 El Niño event in an experimental coupled seasonal ensemble forecasting system

To improve NOAA’s seasonal forecasting capabilities, a new coupled system within the Unified Forecast System (UFS) framework is being developed through a community-wide effort led by NOAA’s Environmental Modeling Center targeting the configuration of a future operational Seasonal Forecast System (SFS v1). An experimental version of this ensemble seasonal forecasting system is tested on forecasting the strong El Niño of 2015/16. The then-operational systems and NCEP real-time seasonal forecasts (CFSv2) underestimated its strength towards the end of 2015 and beginning of 2016. In addition to perturbing the atmospheric initial conditions, run-time stochastic physics-based perturbations are applied in both atmosphere and ocean components of this new coupled system to represent the model uncertainties. The UFS ensembles are initialized on June 1st, 2015 and run through a 9-month period. Compared to CFSv2, the forecast of Niño 3.4 SST and intra-seasonal zonal windstress for the 2015/16 El Niño in the UFS system are improved, as is the ensemble spread. A cold SST forecast error develops in the central equatorial Pacific, likely from excess evaporative cooling, shallower thermocline, and an excessively strong vertical current shear driven cooling. Near the eastern equatorial Pacific coast, on the other hand, warm surface and cool subsurface errors persist from initialization until the end of the forecast. The results suggest that further improvement in the seasonal forecast may be achieved by a combination of factors, including, but not limited to, improving the coupled system initialization, along with the atmospheric physics.

Operational forecasting inundation extents using REOF analysis (FIER) over lower Mekong and its potential economic impact on agriculture

In the Lower Mekong River Basin floodplains, rice cultivation is highly crucial for regional and global food security. However, prolonged flooding can pose damage to rice cultivation and other socio-economic aspects. Yet, there is no rapid operational inundation forecasting system that can help decision-makers proactively mitigate flood damages. Here, we integrated the so-called Forecasting Inundation Extents using Rotated empirical orthogonal function analysis (FIER) framework with an altimetry-based operational Mekong River level forecasting system and built an operational web application, FIER-Mekong, (https://fier-mekong.streamlit.app/) that generates daily skillful forecasted inundation extents (>70% of critical success index) and depths in about 3 and 30 seconds, respectively, with up to 18-day lead times. One of its applications, predicting flood-induced rice economic losses, is also presented. Had FIER-Mekong being adopted, we estimated that the rice damages, up to 87 and 53 million US dollars during the 2020 and 2021 harvest time, respectively, could have been avoided.

Spatiotemporal riskforecasting to improve locust management.

Locusts are among the most feared agricultural pests. Spatiotemporal forecasting is a key process in their management. The present review aims to 1) set a common language on the subject,2) evaluate the current methodologies, and 3) identify opportunities to improve forecasting tools. Forecasts can be used to provide reliable predictions onlocust presence, reproduction events, gregarization areas, population outbreaks, and potential impacts on agriculture. Statistical approaches are used for the first four objectives, whereas mechanistic approaches are used for the latter. We advocate 1) to build reliable and reproducible spatiotemporal forecasting systems for the impacts on agriculture, 2) to turn scientific studies into operational forecasting systems, and 3) to evaluate the performance of these systems.

Advances in Seasonal Predictions of Arctic Sea Ice With NOAA UFS

AbstractThe Unified Forecast System (UFS) is the next generation modeling infrastructure under development for NOAA's operational numerical weather/climate predictions. This study is the first attempt with UFS for seasonal predictions application. In particular, 9‐month hindcasts are performed starting from every month during 2007–2020. The UFS performance in predicting Arctic sea ice was compared to hindcasts by (a) the current operational Climate Forecast System version 2 (CFSv2) and (b) an experimental sea ice prediction system (CFSm5) at NOAA. Evaluations indicate that UFS demonstrates consistently improved skills than both CFSv2 and CFSm5 in seasonal sea ice predictions, together with more realistic climatological sea ice distributions. Diagnostics suggest that the better climatological sea ice distributions in UFS is related to its atmospheric states simulated with FV3, the atmospheric component of UFS, and reinforced by associated ocean circulations. In addition, applications of the multi‐model ensemble strategy do not present skill improvements over the UFS forecasts alone.

Prediction of Extremely Severe Cyclonic Storm “Fani” Using Moving Nested Domain

The prediction of extremely severe cyclonic storms has been a long-standing and challenging issue due to their short life period and large variation in intensities over a short time. In this study, we predict the track, intensity, and structure of an extremely severe cyclonic storm (ESCS) named ‘Fani,’ which developed over the Bay of Bengal region from 27 April to 4 May 2019, using the Advanced Research version of the Weather Research and Forecasting (WRF-ARW) model. Two numerical experiments were conducted using the moving nested domain method with a 3 km horizontal resolution, one with the FLUX-1 air-sea flux parameterization scheme and the other with the FLUX-2 air-sea flux parameterization scheme. The NCEP operational Global Forecast System (GFS) analysis and forecast datasets with a 25 km horizontal resolution were used to derive the initial and boundary conditions. The WRF model’s predicted track and intensity were validated with the best-fit track dataset from the India Meteorological Department (IMD), and the structure was validated with different observations. The results showed that the WRF model with the FLUX-1 air-sea parameterization scheme accurately predicted the track, landfall (position and time), and intensity (minimum sea level pressure and maximum surface wind) of the storm. The track errors on days 1 to 4 were approximately 47 km, 123 km, 96 km, and 27 km in the FLUX-1 experiment and approximately 54 km, 142 km, 152 km, and 166 km in the FLUX-2 experiment, respectively. The intensity was better predicted in the FLUX-1 experiment during the first 60 h, while it was better predicted in the FLUX-2 experiment for the remaining period. The structure, in terms of relative humidity, water vapor, maximum reflectivity, and temperature anomaly of the storm, was also discussed and compared with available satellite and Doppler Weather Radar observations.

Operational wave forecasting for extreme conditions in the Arabian Sea – A comparison with buoy and satellite data

An operational wave forecasting system forced with Global Forecast System (GFS) winds was presented and evaluated for the Arabian Sea. WAVEWATCH III (WWIII) model was used for wave forecasting. Several simulations were executed using different input-dissipation parameterizations embedded in WWIII including ST2, ST3, ST4 and ST6 packages for the period of southwest monsoon and two tropical storms in the Arabian Sea. Performance of forecasting system with different source term packages were evaluated by comparing the forecasted significant wave heights (Hs) with the altimeter-derived Hs and buoy data. Results indicated that time series of the forecasted data derived from different source term packages are well adapted to those of the measured ones at buoy location during southwest monsoon. However, ST3 and ST4 packages outperform the others in terms of forecasting Hs in the entire Arabian Sea as well as buoy location. Also, the forecasted Hs are reliable up to 72-hr lead time. Performance evaluation during the studied tropical cyclones showed that it is not possible to make a general conclusion about the best parameterization. However, the forecast system forced with GFS winds was capable of simulate the peak values of Hs - although with time lag - during the tropical cyclones.

Benefits of fully focused SAR altimetry to coastal wave height estimates: A case study in the North Sea

Estimating the three geophysical variables significant wave height (SWH), sea surface height, and wind speed from satellite altimetry continues to be challenging in the coastal zone because the received radar echoes exhibit significant interference from strongly reflective targets such as sandbanks, sheltered bays, ships etc. Fully focused SAR (FF-SAR) processing exhibits a theoretical along-track resolution of up to less than half a metre. This suggests that the application of FF-SAR altimetry might give potential gains over unfocused SAR (UF-SAR) altimetry to resolve and mitigate small-scale interferers in the along-track direction to improve the accuracy and precision of the geophysical estimates.The objective of this study is to assess the applicability of FF-SAR-processed Sentinel-6 Michael Freilich (S6-MF) coastal altimetry data to obtain SWH estimates as close as possible to the coast.We have developed a multi-mission FF-SAR processor and applied the coastal retracking algorithm CORALv2 to estimate SWH. We assess different FF-SAR and UF-SAR processing configurations, as well as the baseline Level-2 product from EUMETSAT, by comparison with the coastal, high-resolution SWAN-Kuststrook wave model from the Deltares RWsOS North Sea operational forecasting system. This includes the evaluation of the correlation, the median offset, and the percentage of cycles with high correlation as a function of distance to the nearest coastline. Moreover, we analyse the number of valid records and the L2 noise of the records. The case study comprises five coastal crossings of S6-MF that are located along the Dutch coast and the German coast along the East Frisian Islands in the North Sea.We observe that accurate and precise SWH records can be estimated in the nearshore zone within 1–3 km from the coast using satellite SAR altimetry. We find that the FF-SAR-processed dataset with a Level-1b posting rate of 140 Hz shows the greatest similarity with the wave model. We achieve a correlation of ∼0.8 at 80% of valid records and a gain in precision of up to 29% of FF-SAR vs UF-SAR for 1–3 km from the coast. FF-SAR shows, for all cycles, a high correlation of greater than or equal to 0.8 for 1–3 km from the coast. We estimate the decay of SWH from offshore at 30 km to up to 1 km from the coast to amount to 26.4% ± 3.1%.

Downscaled subseasonal fire danger forecast skill across the contiguous United States

AbstractThe increasing complexity and impacts of fire seasons in the United States have prompted efforts to improve early warning systems for wildland fire management. Outlooks of potential fire activity at lead‐times of several weeks can help in wildland fire resource allocation as well as complement short‐term meteorological forecasts for ongoing fire events. Here, we describe an experimental system for developing downscaled ensemble‐based subseasonal forecasts for the contiguous US using NCEP's operational Climate Forecast System version 2 model. These forecasts are used to calculate forecasted fire danger indices from the United States (US) National Fire Danger Rating System in addition to forecasts of evaporative demand. We further illustrate the skill of subseasonal forecasts on weekly timescales using hindcasts from 2011 to 2021. Results show that while forecast skill degrades with time, statistically significant week 3 correlative skill was found for 76% and 30% of the contiguous US for Energy Release Component and evaporative demand, respectively. These results highlight the potential value of experimental subseasonal forecasts in complementing existing information streams in weekly‐to‐monthly fire business decision making for suppression‐based decisions and geographic reallocation of resources during the fire season, as well for proactive fire management actions outside of the core fire season.

Enhancing Forecast Skill of Winter Temperature of East Asia Using Teleconnection Patterns Simulated by GloSea5 Seasonal Forecast Model

GloSea5, a seasonal forecast system of the UK Met Office, shows reasonable skill among state-of-the-art operational seasonal forecast systems. However, the average surface temperature (T2m) in winter (December–February) of GloSea5 is particularly low in East Asia. To improve the seasonal forecast skill over East Asia, we focused on the high skill score of global teleconnection patterns simulated by GloSea5. Among the well-predicted teleconnection patterns, we selected those highly correlated with the East Asian T2m: East Atlantic (EA), Polar/Eurasia (PE), East Atlantic/Western Russia (EAWR), and West Pacific (WP) patterns. A multiple linear regression model was constructed using the selected teleconnection indices as predictors. These results are promising. The statistical skill-score evaluation of the constructed linear regression model using the anomaly correlation coefficient (ACC), root mean squared error (RMSE), and mean-squared skill score (MSSS) showed an improvement in the predicted T2m of East Asia, where the values of ACC and MSSS increased by 0.25 and 0.37, respectively, and the RMSE decreased by 0.63 compared to the dynamic forecast model results. These results suggest that a well-designed combined statistical and dynamical approach for seasonal prediction can be beneficial for some regions where the predictability of the dynamic model exhibits a low value.

Greenlandic sea ice products with a focus on an updated operational forecast system

Sea ice information has traditionally been associated with Manual Ice Charts, however the demand for accurate forecasts is increasing. This study presents an improved operational forecast system for the Arctic sea ice focusing on the Greenlandic waters. In addition, we present different observational sea ice products and conduct inter-comparisons. First, a re-analysis forced by ERA5 from 2000 to 2021 is evaluated to ensure that the forecast system is stable over time and to provide statistics for the users. The output is similar to the initial conditions for a forecast. Secondly, the sea ice forecast system is tested and evaluated based on two re-forecasts forced by the high resolution ECMWF-HRES forecast for the period from January 2019 to September 2021. Both the re-analysis and the re-forecasts include assimilation of sea surface temperatures and sea ice concentrations. We validate the re-analysis and the re-forecast systems for sea ice concentration against different remotely sensed observational products by computing the Integrated Ice Edge Error metric at the initial conditions of each system. The results reveal that the re-analysis and the re-forecast perform well. However, the summertime retreat of sea ice near the western Greenlandic coast seems to be delayed a few days compared with the observations. Importantly, part of the bias associated with the model representation of the sea ice edge is associated with the observational errors due to limitations in the passive microwave product in summertime and also near the coast. An inter-comparison of the observational sea ice products suggests that the model performance could be improved by assimilation of sea ice concentrations derived from a newly-developed automated sea ice product. In addition, analysis of persistence shows that the re-forecast has better skill than the persistence forecast for the vast majority of the time.

Maximizing the forecasting skill of an ensemble model

SUMMARYAn ensemble model integrates forecasts of different models (or different parametrizations of the same model) into one single ensemble forecast. This procedure has different names in the literature and is approached through different philosophies in theory and practice. Previous approaches often weighted forecasts equally or according to their individual skill. Here we present a more meaningful strategy by obtaining weights that maximize the skill of the ensemble. The procedure is based on a multivariate logistic regression and exposes some level of flexibility to emphasize different aspects of seismicity and address different end users. We apply the ensemble strategy to the operational earthquake forecasting system in Italy and demonstrate its superior skill over the best individual forecast model with statistical significance. In particular, we highlight that the skill improves when exploiting the flexibility of fitting the ensemble, for example using only recent and not the entire historical data.

Daily ensemble river discharge reforecasts and real-time forecasts from the operational Global Flood Awareness System

Abstract. Operational global-scale hydrological forecasting systems are used to help manage hydrological extremes such as floods and droughts. The vast amounts of raw data that underpin forecast systems and the ability to generate information on forecast skill have, until now, not been publicly available. As part of the Global Flood Awareness System (GloFAS; https://www.globalfloods.eu/, last access: 3 December 2022) service evolution, in this paper daily ensemble river discharge reforecasts and real-time forecast datasets are made free and openly available through the Copernicus Climate Change Service (C3S) Climate Data Store (CDS). They include real-time forecast data starting on 1 January 2020 updated operationally every day and a 20-year set of reforecasts and associated metadata. This paper describes the model components and configuration used to generate the real-time river discharge forecasts and the reforecasts. An evaluation of ensemble forecast skill using the continuous ranked probability skill score (CRPSS) was also undertaken for river points around the globe. Results show that GloFAS is skilful in over 93 % of catchments in the short (1 to 3 d) and medium range (5 to 15 d) against a persistence benchmark forecast and skilful in over 80 % of catchments out to the extended range (16 to 30 d) against a climatological benchmark forecast. However, the strength of skill varies considerably by location with GloFAS found to have no or negative skill at longer lead times in broad hydroclimatic regions in tropical Africa, western coast of South America, and catchments dominated by snow and ice in high northern latitudes. Forecast skill is summarised as a new headline skill score available as a new layer on the GloFAS forecast Web Map Viewer to aid user interpretation and understanding of forecast quality.

Diagnosing Near-Surface Model Errors with Candidate Physics Parameterization Schemes for the Multiphysics Rapid Refresh Forecast System (RRFS) Ensemble during Winter over the Northeastern United States and Southern Great Plains

Abstract During the winter of 2020/21 an ensemble of FV3-LAM forecasts was produced over the contiguous United States for the Winter Weather Experiment using five physics suites. These forecasts are evaluated with the goal of optimizing physics parameterizations within the future operational Rapid Refresh Forecast System (RRFS) in the Unified Forecast System (UFS) realm and for selecting suitable physics suites for a multiphysics RRFS ensemble. The five physics suites have different combinations of land surface models (LSMs), planetary boundary layer (PBL) parameterizations, and surface layer schemes, chosen from those used in current and possible future operational systems and likely to be supported in the operational UFS. Full-season evaluation reveals a persistent near-surface cold bias in the U.S. Northeast from one suite and a nighttime warm bias in the southern Great Plains in another suite, while other suites have smaller biases. A representative case is chosen to diagnose the cause for each of these biases using sensitivity simulations with different physics combinations or modified parameters and verified with additional mesonet observations. The cold bias in the Northeast is attributed to aspects of the Noah-MP LSM over snow cover, where Noah-MP simulates lower soil water content, and thus lower thermal conductivity than other LSMs, leading to less upward ground heat flux during nighttime and consequently lower surface temperature. The nighttime warm bias found in the southern Great Plains is attributed to overestimation of vertical mixing in the K-profile-based eddy-diffusivity mass-flux (K-EDMF) PBL scheme and insufficient land–atmospheric coupling from the GFS surface layer scheme over short vegetation. A few key parameters driving these systematic biases are identified.

Toward a Deep-Learning-Network-Based Convective Weather Initiation Algorithm From the Joint Observations of Fengyun-4A Geostationary Satellite and Radar for 0–1h Nowcasting

Nowcasting of convective weather is a challenging and significant task in operational weather forecasting system. In this study, a new convolution recurrent neural network (ConvRNN)-based regression network for convective weather prediction is proposed, which is named as the Convective Weather Nowcasting Net (CWNNet). The CWNNet adopts the joint observations of Fengyun-4A geostationary satellite and the ground-based Doppler weather radar data of the last 0–1-hour as the inputs of the model to predict the radar reflectivity factor maps of next 0-1 hour. The statistical validating results clearly demonstrate that the mean values of the probability of detection, false alarm ratio, threat score, root mean square error and mean absolute error evaluating the performance of CWNNet for 1-h nowcasting reach 0.87, 0.137, 0.71, 3.365 dBZ and 1.038 dBZ, respectively. Due to that the geostationary meteorological satellite is capable of capturing the features of convective initiation, the CWNNet shows a good performance in convective initiation nowcasting. Besides, several case studies also further indicate that the CWNNet can predict convective initiation more than 30 minutes in advance by monitoring the convective clouds. The CWNNet based on the joint satellite and radar data shows a better nowcasting performance than that only employing single data source. Thus, it can effectively produce more reliable nowcasting for convective weather events.

Japan Meteorological Agency/Meteorological Research Institute Coupled Prediction System Version 3 (JMA/MRI–CPS3)

A new operational seasonal forecast system, Japan Meteorological Agency (JMA)/Meteorological Research Institute (MRI) Coupled Prediction System (CPS) version 3 (JMA/MRI–CPS3), has been developed. This system represents a major upgrade of the former system, CPS2. CPS3 comprises atmosphere, land, ocean, and sea ice forecast models and the necessary initialization systems for these models. For historical reforecasts, the atmospheric reanalysis dataset JRA-3Q provides initial conditions for the atmosphere and the external forcings for land, ocean, and sea ice analysis. In the operational forecast, JMA's operational atmospheric analysis is used in conjunction with JRA-3Q to initialize the system in near-real time. The land surface model is initialized using an uncoupled free simulation, forced by the atmospheric analysis. The ocean and sea ice models are initialized with the global ocean data assimilation system MOVE-G3, which incorporates a newly developed four-dimensional variational method for temperature, salinity, and sea surface height and a three-dimensional method for sea ice concentration. Compared with the previous system, the CPS3 forecast model components have approximately 2-4 times higher resolution: the atmosphere and land models are configured with ∼ 55 km horizontal resolution, with 100 vertical atmosphere layers; and the ocean and sea ice models have a resolution of 0.25° × 0.25°, with 60 vertical ocean layers. The physical processes of the atmosphere are greatly refined in CPS3 relative to CPS2, resulting in improved representation of sub-seasonal to seasonal scale variability, including the eastward propagation of the Madden–Julian Oscillation, winter blocking highs in the North Atlantic, and coupled atmosphere–ocean variability during El Niño–Southern Oscillation events. Our historical reforecast experiment for 1991-2020 suggests that CPS3 has greater forecast skill than CPS2. The usability of the model output has been improved in CPS3 by reorganizing the operation schedule to provide daily updates of five-member ensemble forecasts.

Comparative Analysis of Forecasting Accuracy and Model Performance for Development of Coastal Wave Forecasting System Based on Unstructured Grid

We develop a coastal wave forecasting system by using the unstructured grid based on sea wind data of Global Data Assimilation and Prediction System. The verification is performed to examine the performance and accuracy of the wave model. Since the conventional grid has limited wave forecasting on complex coastlines and bathymetry, the unstructured grid system is applied for precise numerical simulation, and applicability for operational support is evaluated. Both grid systems show similar prediction trends in offshore and coastal areas, and the difference in prediction errors according to the grid system is not large. In addition, the applicability of the operational wave forecasting system is confirmed by dramatically reducing the model execution time of the unstructured grid under the same conditions.

Seasonal prediction of Horn of Africa long rains using machine learning: The pitfalls of preselecting correlated predictors

The Horn of Africa is highly vulnerable to droughts and floods, and reliable long-term forecasting is a key part of building resilience. However, the prediction of the “long rains” season (March–May) is particularly challenging for dynamical climate prediction models. Meanwhile, the potential for machine learning to improve seasonal precipitation forecasts in the region has yet to be uncovered. Here, we implement and evaluate four data-driven models for prediction of long rains rainfall: ridge and lasso linear regressions, random forests and a single-layer neural network. Predictors are based on SSTs, zonal winds, land state, and climate indices, and the target variables are precipitation totals for each separate month (March, April, and May) in the Horn of Africa drylands, with separate predictions made for lead-times of 1–3 months. Results reveal a tendency for overfitting when predictors are preselected based on correlations to the target variable over the entire historical period, a frequent practice in machine learning-based seasonal forecasting. Using this conventional approach, the data-driven methods—and particularly the lasso and ridge regressions—often outperform dynamical seasonal hindcasts. However, when the selection of predictors is done independently of both the train and test data, by performing this predictor selection within the cross-validation loop, the performance of all four data-driven models is poorer than that of the dynamical hindcasts. These findings should not discourage future applications of machine learning for rainfall forecasting in the region. Yet, they should be seen as a note of caution to prevent optimistically biased results that are not indicative of the true power in operational forecast systems.

An Optimal Interpolation–Based Snow Data Assimilation for NOAA’s Unified Forecast System (UFS)

Abstract Within the National Weather Service’s Unified Forecast System (UFS), snow depth and snow cover observations are assimilated once daily using a rule-based method designed to correct for gross errors. While this approach improved the forecasts over its predecessors, it is now quite outdated and is likely to result in suboptimal analysis. We have then implemented and evaluated a snow data assimilation using the 2D optimal interpolation (OI) method, which accounts for model and observation errors and their spatial correlations as a function of distances between the observations and model grid cells. The performance of the OI was evaluated by assimilating daily snow depth observations from the Global Historical Climatology Network (GHCN) and the Interactive Multisensor Snow and Ice Mapping System (IMS) snow cover data into the UFS, from October 2019 to March 2020. Compared to the control analysis, which is very similar to the method currently in operational use, the OI improves the forecast snow depth and snow cover. For instance, the unbiased snow depth root-mean-squared error (ubRMSE) was reduced by 45 mm and the snow cover hit rate increased by 4%. This leads to modest improvements to globally averaged near-surface temperature (an average reduction of 0.23 K in temperature bias), with significant local improvements in some regions (much of Asia, the central United States). The reduction in near-surface temperature error was primarily caused by improved snow cover fraction from the data assimilation. Based on these results, the OI DA is currently being transitioned into operational use for the UFS. Significance Statement Weather and climate forecasting systems rely on accurate modeling of the evolution of atmospheric, oceanic, and land processes. In addition, model forecasts are substantially improved by continuous incorporation of observations to models, through a process called data assimilation. In this work, we upgraded the snow data assimilation used in the U.S. National Weather Service (NWS) global weather prediction system. Compared to the method currently in operational use, the new snow data assimilation improves both the forecasted snow quantity and near-surface air temperatures over snowy regions. Based on the positive results obtained in the experiments presented here, the new snow data assimilation method is being implemented in the NWS operational forecast system.

Technical note: Data assimilation and autoregression for using near-real-time streamflow observations in long short-term memory networks

Abstract. Ingesting near-real-time observation data is a critical component of many operational hydrological forecasting systems. In this paper, we compare two strategies for ingesting near-real-time streamflow observations into long short-term memory (LSTM) rainfall–runoff models: autoregression (a forward method) and variational data assimilation. Autoregression is both more accurate and more computationally efficient than data assimilation. Autoregression is sensitive to missing data, however an appropriate (and simple) training strategy mitigates this problem. We introduce a data assimilation procedure for recurrent deep learning models that uses backpropagation to make the state updates.