Improvement Of Prediction Skill Research Articles

Scalable interpolation of satellite altimetry data with probabilistic machine learning

We present GPSat; an open-source Python programming library for performing efficient interpolation of non-stationary satellite altimetry data, using scalable Gaussian process techniques. We use GPSat to generate complete maps of daily 50 km-gridded Arctic sea ice radar freeboard, and find that, relative to a previous interpolation scheme, GPSat offers a 504 × computational speedup, with less than 4 mm difference on the derived freeboards on average. We then demonstrate the scalability of GPSat through freeboard interpolation at 5 km resolution, and Sea-Level Anomalies (SLA) at the resolution of the altimeter footprint. Interpolated 5 km radar freeboards show strong agreement with airborne data (linear correlation of 0.66). Footprint-level SLA interpolation also shows improvements in predictive skill over linear regression. In this work, we suggest that GPSat could overcome the computational bottlenecks faced in many altimetry-based interpolation routines, and hence advance critical understanding of ocean and sea ice variability over short spatio-temporal scales.

Improving the Predictability of the US Seasonal Surface Temperature With Convolutional Neural Networks Trained on CESM2 LENS

AbstractTo better understand and improve the prediction of the seasonal surface temperature (TS) across the United States, we employed convolutional neural network (CNN) models trained on the Community Earth System Model Version 2 Large Ensemble (CESM2 LENS). We used lagged sea surface temperatures (SST) over the tropical Pacific region, containing the information of the El Niño Southern Oscillation (ENSO), as input for the CNN models. ENSO is the principal driver of variability in seasonal US surface temperatures (TSUS) and employing CNN models allows for spatiotemporal aspects of ENSO to be analyzed to make seasonal TSUS predictions. For predicting TSUS, the CNN models exhibited significantly improved skill over standard statistical multilinear regression (MLR) models and dynamical forecasts across most regions in the US, for lead times ranging from 1 to 6 months. Furthermore, we employed the CNN models to predict seasonal TSUS during extreme ENSO events. For these events, the CNN models outperformed the MLR models in predicting the effects on seasonal TSUS, suggesting that the CNN models are able to capture the ENSO‐TSUS teleconnection more effectively. Results from a heatmap analysis demonstrate that the CNN models utilize spatial features of ENSO rather than solely the magnitude of the ENSO events, indicating that the improved skill of seasonal TSUS is due to analyzing spatial variation in ENSO events. The proposed CNN model demonstrates a promising improvement in prediction skill compared to existing methods, suggesting a potential path forward for enhancing TSUS forecast skill from subseasonal to seasonal timescales.

Improving subseasonal forecasting of East Asian monsoon precipitation with deep learning

Accurate subseasonal forecasting of East Asian summer monsoon (EASM) precipitation is crucial, as it directly impacts the livelihoods of billions. However, the prediction skill of state-of-the-art subseasonal-to-seasonal (S2S) models for precipitation remains limited. In this study, the authors developed a convolutional neural network (CNN) regression model to enhance the prediction skill for weekly EASM precipitation by utilizing the more reliably predicted circulation fields from dynamic models. The outcomes of the CNN model are promising, as it led to a 14 % increase in the anomaly correlation coefficient (ACC), from 0.30 to 0.35, and a 22 % reduction in the root-mean-square error (RMSE), from 3.22 to 2.52, for predicting the weekly EASM precipitation index at a leading time of one week. Among the S2S models, the improvement in prediction skill through CNN correction depends on the model's performance in accurately predicting circulation fields. The CNN correction of EASM precipitation index can only rectify the systematic errors of the model and is independent of whether the each grid point or the entire area-averaged index is corrected. Furthermore, u200 (200-hPa zonal wind) is identified as the most important variable for efficient correction.摘要东亚夏季风(EASM)降水的准确次季节预报至关重要, 因为它直接影响着数十亿人的生计. 然而, 最先进的次季节-季节(S2S)预测模型的预测技巧仍然有限. 本研究开发了一种卷积神经网络(CNN)回归模型, 通过利用动力预测模型预测的更可靠的环流场来提高EASM周降水的预测技巧. 经过CNN模型的订正, 在提前一周预测EASM降水指数时, 11个S2S模式的平均距平相关系数从增加了14 %, 从0.30增加到0.35; 均方根误差减少了22 %, 从3.22减少到2.52. 在这些S2S模式中, 通过CNN订正对预测技巧的提高程度取决于模式在准确预测大气环流变量方面的表现. 对EASM降水指数的CNN订正只能订正模式的系统误差, 与逐个网格订正还是整个区域平均指数订正无关, 并且在不同的提前期内CNN的订正效果基本不变. 此外, 200hPa纬向风被认为是有效订正的最重要变量.

A Simple Statistical Postprocessing Scheme for Enhancing the Skill of Seasonal SST Predictions in the Tropics

Abstract Seasonal prediction systems are subject to systematic errors, including those introduced during the initialization procedure, that may degrade the forecast skill. Here we use a novel statistical postprocessing correction scheme that is based on canonical correlation analysis (CCA) to relate errors in ocean temperature arising during initialization with errors in the predicted sea surface temperature fields at 1–12-month lead time. In addition, the scheme uses CCA of simultaneous SST fields from the prediction and corresponding observations to correct pattern errors. Finally, simple scaling is used to mitigate systematic location and phasing errors as a function of lead time and calendar month. Applying this scheme to an ensemble of seven seasonal prediction models suggests that moderate improvement of prediction skill is achievable in the tropical Atlantic and, to a lesser extent, in the tropical Pacific and Indian Ocean. The scheme possesses several adjustable parameters, including the number of CCA modes retained, and the regions of the left and right CCA patterns. These parameters are selected using a simple tuning procedure based on the average of four skill metrics. The results of the present study indicate that errors in ocean temperature fields due to imperfect initialization and SST variability errors can have a sizable negative impact on SST prediction skill. Further development of prediction systems may be able to remedy these impacts to some extent. Significance Statement The prediction of year-to-year climate variability patterns, such as El Niño, offers potential benefits to society by aiding mitigation and adaptation efforts. Current prediction systems, however, may still have substantial room for improvement due to systematic model errors and due to imperfect initialization of the oceanic state at the start of predictions. Here we develop a statistical correction scheme to improve prediction skill after forecasts have been completed. The scheme shows some moderate success in improving the skill for predicting El Niño and similar climate patterns in seven prediction systems. Our results not only indicate a potential for improving prediction skill after the fact but also point to the importance of improving the way prediction systems are initialized.

Ocean data assimilation for the initialization of seasonal prediction with the Community Earth System Model

Seasonal prediction highly depends on its initialization scheme. Currently, the operational utilization of advanced data assimilation methods in the oceanic initialization is still relatively less and its benefit is far from being fully unveiled. In this study, the Community Earth System Model (CESM) is configured with an Ensemble Adjustment Kalman Filter (EAKF) to initialize the seasonal prediction. The performance of the established prediction system is assessed by comparing it with a benchmark prediction system, which is based on the same model but with a nudging scheme. Results show that the assimilation of only oceanic observations can not only effectively constrain the oceanic variables, but also improves the atmospheric variables through coupling processes, which helps to produce accurate and compatible initial conditions. With the initial conditions, the current prediction system is better than the benchmark prediction system and can produce skillful predictions for sea surface temperature, ocean heat content, air temperature at 2 m, precipitation and major climate variabilities including El Niño–Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD). The current prediction system also shows better ensemble consistency and less prediction drifts, which could contribute to the improvement of prediction skill.

Prediction of plasmaspheric hiss spectral classes

We present a random forests machine learning model for prediction of plasmaspheric hiss spectral classes from the Van Allen Probes dataset. The random forests model provides accurate prediction of plasmaspheric hiss spectral classes obtained by the self organizing map (SOM) unsupervised machine learning classification technique. The high predictive skill of the random forests model is largely determined by the distinct and different locations of a given spectral class (“no hiss”, “regular hiss”, and “low-frequency hiss”) in (MLAT, MLT, L) coordinate space, which are the main predictors of the simplest and most accurate base model. Adding to such a base model any other single predictor among different magnetospheric, geomagnetic, and solar wind conditions provides only minor and similarly incremental improvements in predictive skill, which is comparable to the one obtained when including all possible predictors, and thus confirming major role of spatial location for accurate prediction.

Self-Organizing Maps Identify Windows of Opportunity for Seasonal European Summer Predictions

We combine a machine learning method and ensemble climate predictions to investigate windows of opportunity for seasonal predictability of European summer climate associated with the North Atlantic jet stream. We particularly focus on the impact of North Atlantic spring sea surface temperatures (SST) on the four dominant atmospheric teleconnections associated with the jet stream: the summer North Atlantic Oscillation (NAO) in positive and negative phases, the Atlantic Ridge (At. Ridge), and Atlantic Low (At. Low). We go beyond standard forecast practices by not only identifying these atmospheric teleconnections and their SST precursors but by making use of these identified precursors in the analysis of a dynamical forecast ensemble. Specifically, we train the neural network-based classifier Self-Organizing Maps (SOM) with ERA-20C reanalysis and combine it with model simulations from the Max Planck Institute Earth System Model in mixed resolution (MPI-ESM-MR). We use two different sets of 30-member hindcast ensembles initialized every May, one for training and evaluation between 1902 and 2008, and one for verification between 1980–2016, respectively. Among the four summer atmospheric teleconnections analyzed here, we find that At. Ridge simulated by MPI-ESM-MR shows the best agreement with ERA-20C, thereby representing with its occurrence windows of opportunity for skillful summer predictions. Conversely, At. Low shows the lowest agreement, which might limit the model skill for early warning of warmer than average summers. In summary, we find that spring SST patterns identified with a SOM analysis can be used to guess the dominant summer atmospheric teleconnections at initialization and guide a sub-selection of potential skillful ensemble members. This holds especially true for At. Ridge and At. Low and is unclear for summer NAO. We show that predictive skill in the selected ensemble exceeds that of the full ensemble over regions in the Euro-Atlantic domain where spring SST significantly correlates with summer sea level pressure (SLP). In particular, we find a significant improvement in predictive skill for SLP, geopotential height at 500 hPa, and 2 m temperature at 3–4 months lead time over Scandinavia, which is robust among the two sets of hindcast ensembles.

Trivial improvements in predictive skill due to direct reconstruction of the global carbon cycle

Abstract. State-of-the art climate prediction systems have recently included a carbon component. While physical-state variables are assimilated in reconstruction simulations, land and ocean biogeochemical state variables adjust to the state acquired through this assimilation indirectly instead of being assimilated themselves. In the absence of comprehensive biogeochemical reanalysis products, such an approach is pragmatic. Here we evaluate a potential advantage of having perfect carbon cycle observational products to be used for direct carbon cycle reconstruction. Within an idealized perfect-model framework, we reconstruct a 50-year target period from a control simulation. We nudge variables from this target onto arbitrary initial conditions, mimicking an assimilation simulation generating initial conditions for hindcast experiments of prediction systems. Interested in the ability to reconstruct global atmospheric CO2, we focus on the global carbon cycle reconstruction performance and predictive skill. We find that indirect carbon cycle reconstruction through physical fields reproduces the target variations. While reproducing the large-scale variations, nudging introduces systematic regional biases in the physical-state variables to which biogeochemical cycles react very sensitively. Initial conditions in the oceanic carbon cycle are sufficiently well reconstructed indirectly. Direct reconstruction slightly improves initial conditions. Indirect reconstruction of global terrestrial carbon cycle initial conditions are also sufficiently well reconstructed by the physics reconstruction alone. Direct reconstruction negligibly improves air–land CO2 flux. Atmospheric CO2 is indirectly very well reconstructed. Direct reconstruction of the marine and terrestrial carbon cycles slightly improves reconstruction while establishing persistent biases. We find improvements in global carbon cycle predictive skill from direct reconstruction compared to indirect reconstruction. After correcting for mean bias, indirect and direct reconstruction both predict the target similarly well and only moderately worse than perfect initialization after the first lead year. Our perfect-model study shows that indirect carbon cycle reconstruction yields satisfying initial conditions for global CO2 flux and atmospheric CO2. Direct carbon cycle reconstruction adds little improvement to the global carbon cycle because imperfect reconstruction of the physical climate state impedes better biogeochemical reconstruction. These minor improvements in initial conditions yield little improvement in initialized perfect-model predictive skill. We label these minor improvements due to direct carbon cycle reconstruction “trivial”, as mean bias reduction yields similar improvements. As reconstruction biases in real-world prediction systems are likely stronger, our results add confidence to the current practice of indirect reconstruction in carbon cycle prediction systems.

A wavelet-based tool to modulate variance in predictors: An application to predicting drought anomalies

This work presents an open-source tool to predict natural system responses by transforming the frequency spectrum of predictor variables to create a response that better resembles observations. The R package, namely WAvelet System Prediction (WASP), is based on a discrete wavelet transform (DWT)-based variance transformation method. We further introduce the maximal overlap DWT (MODWT)-based variance transformation which allows the method to be used in forecasting applications. We also develop the method to include an unbiased estimator that mitigates the well-known issue of edge effects in wavelet transforms. The predictive model in the method is a k-nearest neighbor (knn) approach. The main functionalities of the software include: (1) transforming the system predictors, (2) identifying significant predictors corresponding to the response, (3) predicting target response using the knn. Results of predicting sustained drought anomalies across Australia show clear improvements in predictive skill compared to the use of untransformed predictors.

Exploring sensitive area in the tropical Indian Ocean for El Niño prediction: implication for targeted observation

Based on initial errors of sea temperature in the tropical Indian Ocean that are most likely to induce spring predictability barrier (SPB) for the El Nino prediction, the sensitive area of sea temperature in the tropical Indian Ocean for El Nino prediction starting from January is identified using the CESM1.0.3 (Community Earth System Model), a fully coupled global climate model. The sensitive area locates mainly in the subsurface of eastern Indian Ocean. The effectiveness of applying targeted observation in the sensitive area is also evaluated in an attempt to improve the El Nino prediction skill. The results of sensitivity experiments indicate that if initial errors exist only in the tropical Indian Ocean, applying targeted observation in the sensitive area in the Indian Ocean can significantly improve the El Nino prediction. In particular, for SPB-related El Nino events, when initial errors of sea temperature exist both in the tropical Indian Ocean and the Pacific Ocean, which is much closer to the realistic predictions, if targeted observations are conducted in the sensitive area of tropical Pacific, the prediction skills of SPB-related El Nino events can be improved by 20.3% in general. Moreover, if targeted observations are conducted in the sensitive area of tropical Indian Ocean in addition, the improvement of prediction skill can be increased by 25.2%. Considering the volume of sensitive area in the tropical Indian Ocean is about 1/3 of that in the tropical Pacific Ocean, the prediction skill improvement per cubic kilometer in the sensitive area of tropical Indian Ocean is competitive to that of the tropical Pacific Ocean. Additional to the sensitive area of the tropical Pacific Ocean, sensitive area of the tropical Indian Ocean is also a very effective and cost-saving area for the application of targeted observations to improve El Nino forecast skills.

Delivering Sustained, Coordinated, and Integrated Observations of the Southern Ocean for Global Impact

The Southern Ocean is disproportionately important in its effect on the Earth system, impacting climatic, biogeochemical and ecological systems, which makes recent observed changes to this system cause for global concern. The enhanced understanding and improvements in predictive skill needed for understanding and projecting future states of the Southern Ocean require sustained observations. Over the last decade, the Southern Ocean Observing System (SOOS) has established networks for enhancing regional coordination and research community groups to advance development of observing system capabilities. These networks support delivery of the SOOS 20-year vision, which is to develop a circumpolar system that ensures time series of key variables, and deliver the greatest impact from data to all key end-users. Although the Southern Ocean remains one of the least-observed ocean regions, enhanced international coordination and advances in autonomous platforms have resulted in progress towards addressing the need for sustained observations of this region. Since 2009, the Southern Ocean community has deployed over 5700 observational platforms south of 40°S. Large-scale, multi-year or sustained, multidisciplinary efforts have been supported and are now delivering observations of essential variables at space and time scales that enable assessment of changes being observed in Southern Ocean systems. The improved observational coverage, however, is predominantly for the open ocean, encompasses the summer, consists of primarily physical oceanographic variables and covers surface to 2000 m. Significant gaps remain in observations of the ice-impacted ocean, the sea ice, depths more than 2000 m, the air-sea-ice interface, biogeochemical and biological variables, and for seasons other than summer. Addressing these data gaps in a sustained way requires parallel advances in coordination networks, cyberinfrastructure and data management tools, observational platform and sensor technology, platform interrogation and data-transmission technologies, modeling frameworks, and internationally agreed sampling requirements of key variables. This paper presents a community statement on the major scientific and observational progress of the last decade, and importantly, an assessment of key priorities for the coming decade, towards achieving the SOOS vision and delivering essential data to all end users.

Improving Met Office seasonal predictions of Arctic sea ice using assimilation of CryoSat-2 thickness

Abstract. Interest in seasonal predictions of Arctic sea ice has been increasing in recent years owing, primarily, to the sharp reduction in Arctic sea-ice cover observed over the last few decades, a decline that is projected to continue. The prospect of increased human industrial activity in the region, as well as scientific interest in the predictability of sea ice, provides important motivation for understanding, and improving, the skill of Arctic predictions. Several operational forecasting centres now routinely produce seasonal predictions of sea-ice cover using coupled atmosphere–ocean–sea-ice models. Although assimilation of sea-ice concentration into these systems is commonplace, sea-ice thickness observations, being much less mature, are typically not assimilated. However, many studies suggest that initialization of winter sea-ice thickness could lead to improved prediction of Arctic summer sea ice. Here, for the first time, we directly assess the impact of winter sea-ice thickness initialization on the skill of summer seasonal predictions by assimilating CryoSat-2 thickness data into the Met Office's coupled seasonal prediction system (GloSea). We show a significant improvement in predictive skill of Arctic sea-ice extent and ice-edge location for forecasts of September Arctic sea ice made from the beginning of the melt season. The improvements in sea-ice cover lead to further improvement of near-surface air temperature and pressure fields across the region. A clear relationship between modelled winter thickness biases and summer extent errors is identified which supports the theory that Arctic winter thickness provides some predictive capability for summer ice extent, and further highlights the importance that modelled winter thickness biases can have on the evolution of forecast errors through the melt season.

Dynamical prediction of Arctic sea ice modes of variability

This study explores the prediction skill of the northern hemisphere (NH) sea ice thickness (SIT) modes of variability in a state-of-the-art coupled forecast system with respect to two statistical forecast benchmarks. Application of the K-means clustering method on a historical reconstruction of SIT from 1958 to 2013, produced by an ocean-sea-ice general circulation model, identifies three Arctic SIT clusters or modes of climate variability. These SIT modes have consistent patterns in different calendar months and their discrete time series of occurrences show persistence on intraseasonal to interannual time scales. We use the EC-Earth2.3 coupled climate model to produce five-member 12-month-long monthly forecasts of the NH SIT modes initialized on 1 May and 1 November every year from 1979 to 2010. We use a three-state first-order Markov chain and climatological probability forecasts determined from the historical SIT mode reconstruction as two statistical reference forecasts. The analysis of ranked probability skill scores (RPSSs) relating these three forecast systems shows that the dynamical SIT mode forecasts typically have a higher skill than the Markov chain forecasts, which are overall better than climatological forecasts. The evolution of RPSS in forecast time indicates that the transition from the sea-ice melting season to growing season in the EC-Earth2.3 forecasts, with respect to the Markov chain model, typically leads to the improvement of prediction skill. The reliability diagrams overall show better reliability of the dynamical forecasts than that of the Markov chain model, especially for 1 May start dates, while dynamical forecasts with 1 November start dates are overconfident. The relative operating characteristics (ROC) diagrams confirm this hierarchy of forecast skill among these three forecast systems. Furthermore, ROC diagrams stratified in groups of 3 sequential forecast months show that Arctic SIT mode forecasts initialized on 1 November typically lose resolution with forecast time more slowly than forecasts initialized on 1 May.

Assessment of prediction skill in equatorial Pacific Ocean in high resolution model of CFS

The effect of increasing atmospheric resolution on prediction skill of El Nino southern oscillation phenomenon in climate forecast system model is explored in this paper. Improvement in prediction skill for sea surface temperature (SST) and winds at all leads compared to low resolution model in the tropical Indo-Pacific basin is observed. High resolution model is able to capture extreme events reasonably well. As a result, the signal to noise ratio is improved in the high resolution model. However, spring predictability barrier (SPB) for summer months in Nino 3 and Nino 3.4 region is stronger in high resolution model, in spite of improvement in overall prediction skill and dynamics everywhere else. Anomaly correlation coefficient of SST in high resolution model with observations in Nino 3.4 region targeting boreal summer months when predicted at lead times of 3–8 months in advance decreased compared its lower resolution counterpart. It is noted that higher variance of winds predicted in spring season over central equatorial Pacific compared to observed variance of winds results in stronger than normal response on subsurface ocean, hence increases SPB for boreal summer months in high resolution model.

Impact of Land Surface Initialization and Land-Atmosphere Coupling on the Prediction of the Indian Summer Monsoon with the CFSv2

The impact of initial land-surface states on monthly to seasonal prediction skill of the Indian summer monsoon (June- September) is investigated using a suite of hindcasts made with the Climate Forecast System version 2 (CFSv2) operational forecast model. The modern paradigm of land-atmosphere coupling is applied to quantify biases in different components of the land-atmosphere coupled system and their effect on systematic errors. Three sets of hindcasts are performed for the period spanning 1982-2009 initialized at the start of April, May and June. For a particular initial date of a given year, one member (Control run) has the analyzed land initial state consistent with the atmosphere, sea ice and ocean states for that year; the other 27 members have land states taken from each of the remaining 27 years. There is significant improvement in the deterministic prediction skill of near surface temperature and soil moisture on monthly and seasonal time scales due to realistic land initial conditions. The improvement occurs in those areas where the land-atmosphere coupling is strongest. Improvements in the prediction skill of precipitation are confined to relatively small areas. The pattern of skill differences resembles patterns of land-atmosphere coupling strength, while biases in the representation of land-atmosphere coupling affect the skill of temperature and rainfall. The re-emergence of skill in temperature and precipitation towards the end of the season over northwest India within April and June IC hindcasts may be attributed to better simulation of the withdrawal phase of the monsoon as well as increased land-atmosphere coupling. For May IC hindcasts, increased skill in air temperature on the sub-seasonal time scales could be due to other large-scale factors. Errors in the parameterization of radiation, convection, boundary layer processes, surface moisture fluxes and the representation of vegetation contribute to decay in potential predictability and skill attributable to land initial conditions. Furthermore, incorrect representation of daily and sub-daily precipitation statistics over land also likely lead to errors in land-atmosphere coupling. Above all, the importance of accurate land surface initialization and land-atmosphere coupling in improving the Indian summer monsoon prediction on sub-seasonal to seasonal time scales is emphasized.

Prediction of Monthly Summer Monsoon Rainfall Using Global Climate Models Through Artificial Neural Network Technique

The monthly prediction of summer monsoon rainfall is very challenging because of its complex and chaotic nature. In this study, a non-linear technique known as Artificial Neural Network (ANN) has been employed on the outputs of Global Climate Models (GCMs) to bring out the vagaries inherent in monthly rainfall prediction. The GCMs that are considered in the study are from the International Research Institute (IRI) (2-tier CCM3v6) and the National Centre for Environmental Prediction (Coupled-CFSv2). The ANN technique is applied on different ensemble members of the individual GCMs to obtain monthly scale prediction over India as a whole and over its spatial grid points. In the present study, a double-cross-validation and simple randomization technique was used to avoid the over-fitting during training process of the ANN model. The performance of the ANN-predicted rainfall from GCMs is judged by analysing the absolute error, box plots, percentile and difference in linear error in probability space. Results suggest that there is significant improvement in prediction skill of these GCMs after applying the ANN technique. The performance analysis reveals that the ANN model is able to capture the year to year variations in monsoon months with fairly good accuracy in extreme years as well. ANN model is also able to simulate the correct signs of rainfall anomalies over different spatial points of the Indian domain .

Calibrating and assessing uncertainty in coastal numerical models

Advanced numerical models used to predict coastal change at a variety of time and spatial scales often contain many free parameters that require calibration to the available field data. At present, little guidance (beyond the adoption of the default values provided) is available in the field of coastal engineering to inform the selection of best-fit parameter values. Common calibration techniques can often lack a rigorous quantification of model sensitivity to parameters and parameter-induced model uncertainty. Here we employ the Generalised Likelihood Uncertainty Estimation (GLUE) method to address these issues. The GLUE method uses Monte Carlo sampling to assess the skill of many different combinations of model parameters when compared to observational data. As a rigorous modelling framework, the GLUE method provides a series of standard tools that assist the modeller to analyse model sensitivity, undertake parameter optimisation and quantify parameter-induced uncertainty. In addition, new tools are presented here to identify where unique calibrated parameter sets are required for different observational data (e.g., should the calibrated parameter set differ between alongshore locations at a site) and investigate the convergence of GLUE estimated optimum parameter values over increasing numbers of Monte Carlo samples.As the methodology and philosophy of GLUE is well established in other fields, this paper presents a practical case study to explore the strengths and weaknesses of the method when applied to a relatively complex coastal numerical model (XBeach). The results obtained are compared to a previously reported and more ‘standard’ model calibration undertaken within the context of a coastal storm early warning system. While the GLUE method requires orders of magnitude more computational power, it is shown that its use in place of the more common one-at-a-time ‘trial-and-error’ approach to model calibration, provides: a significant improvement in predictive skill; a more rigorous evaluation of the model sensitivity to parameters; the ability to identify distinct differences in the XBeach model performance dependent on dune impact processes; and additional analysis including the quantification of parameter-induced uncertainty.

THORPEX Research and the Science of Prediction

AbstractThe Observing System Research and Predictability Experiment (THORPEX) was a 10-yr, international research program organized by the World Meteorological Organization’s World Weather Research Program. THORPEX was motivated by the need to accelerate the rate of improvement in the accuracy of 1-day to 2-week forecasts of high-impact weather for the benefit of society, the economy, and the environment. THORPEX, which took place from 2005 to 2014, was the first major international program focusing on the advancement of global numerical weather prediction systems since the Global Atmospheric Research Program, which took place almost 40 years earlier, from 1967 through 1982. The scientific achievements of THORPEX were accomplished through bringing together scientists from operational centers, research laboratories, and the academic community to collaborate on research that would ultimately advance operational predictive skill. THORPEX included an unprecedented effort to make operational products readily accessible to the broader academic research community, with community efforts focused on problems where challenging science intersected with the potential to accelerate improvements in predictive skill. THORPEX also collaborated with other major programs to identify research areas of mutual interest, such as topics at the intersection of weather and climate. THORPEX research has 1) increased our knowledge of the global-to-regional influences on the initiation, evolution, and predictability of high-impact weather; 2) provided insight into how predictive skill depends on observing strategies and observing systems; 3) improved data assimilation and ensemble forecast systems; 4) advanced knowledge of high-impact weather associated with tropical and polar circulations and their interactions with midlatitude flows; and 5) expanded society’s use of weather information through applied and social science research.

Effects of air-sea interaction on extended-range prediction of geopotential height at 500 hPa over the northern extratropical region

The contribution of air-sea interaction on the extended-range prediction of geopotential height at 500 hPa in the northern extratropical region has been analyzed with a coupled model form Beijing Climate Center and its atmospheric components. Under the assumption of the perfect model, the extended-range prediction skill was evaluated by anomaly correlation coefficient (ACC), root mean square error (RMSE), and signal-to-noise ratio (SNR). The coupled model has a better prediction skill than its atmospheric model, especially, the air-sea interaction in July made a greater contribution for the improvement of prediction skill than other months. The prediction skill of the extratropical region in the coupled model reaches 16–18 days in all months, while the atmospheric model reaches 10–11 days in January, April, and July and only 7–8 days in October, indicating that the air-sea interaction can extend the prediction skill of the atmospheric model by about 1 week. The errors of both the coupled model and the atmospheric model reach saturation in about 20 days, suggesting that the predictable range is less than 3 weeks.

Ensemble Forecasts: Probabilistic Seasonal Forecasts Based on a Model Ensemble

Ensembles of general circulation model (GCM) integrations yield predictions for meteorological conditions in future months. Such predictions have implicit uncertainty resulting from model structure, parameter uncertainty, and fundamental randomness in the physical system. In this work, we build probabilistic models for long-term forecasts that include the GCM ensemble values as inputs but incorporate statistical correction of GCM biases and different treatments of uncertainty. Specifically, we present, and evaluate against observations, several versions of a probabilistic forecast for gridded air temperature 1 month ahead based on ensemble members of the National Centers for Environmental Prediction (NCEP) Climate Forecast System Version 2 (CFSv2). We compare the forecast performance against a baseline climatology based probabilistic forecast, using average information gain as a skill metric. We find that the error in the CFSv2 output is better represented by the climatological variance than by the distribution of ensemble members because the GCM ensemble sometimes suffers from unrealistically little dispersion. Lack of ensemble spread leads a probabilistic forecast whose variance is based on the ensemble dispersion alone to underperform relative to a baseline probabilistic forecast based only on climatology, even when the ensemble mean is corrected for bias. We also show that a combined regression based model that includes climatology, temperature from recent months, trend, and the GCM ensemble mean yields a probabilistic forecast that outperforms approaches using only past observations or GCM outputs. Improvements in predictive skill from the combined probabilistic forecast vary spatially, with larger gains seen in traditionally hard to predict regions such as the Arctic.