Global Forecasting Model for LED Lumen Degradation: An Optimal Cluster Estimation Method

Improving the accuracy of global forecasting models using time series data augmentation

Global models for time series forecasting: A Simulation study

LoMEF: A framework to produce local explanations for global model time series forecasts

Ensembles of localised models for time series forecasting

Seasonal climate forecasts may be coupled with crop models to provide quantitative forecasts of crop yield, assess sensitivity to farm management decisions and manage risk associated with seasonal climate variability. Today, seasonal climate forecasts are produced by computationally expensive, physically-based global climate models, which capture large-scale climate patterns well. However, their coarse spatial resolution (typically >50km) means they do not reliably depict daily weather at sub-grid locations, limiting their direct use in crop models. Consequently, operational crop forecasting systems in Australia typically use alternative meteorological forcings such as historical climate analogues based on El Nino - Southern Oscillation phases, which may be less skillful than global climate model forecasts. An emerging tactic for coupling global climate model forecasts and crop models is to apply quantile mapping (otherwise known as cumulative distribution function matching) to adjust forecast ensemble members according to the historical distribution of observations. However, quantile mapping assumes the global climate model forecasts are highly skilful and well-behaved (which they are often not). The overly simplistic formulation of quantile-mapping propagates an assortment of model errors. Additionally, quantile-mapping cannot be used for downscaling to multiple sub-grid locations owing to its deterministic nature. Accordingly, an increasing number of studies are reporting negative results arising from coupling global climate model forecasts and crop models using quantile mapping. Hence, the overarching objective of this thesis is to develop more robust, spatially and temporally relevant post-processing methods to harness global climate model forecasts for use in crop models. To this end, I develop a new multivariate forecast post-processing workflow that combines Bayesian parametric methods and non-parametric methods to calibrate and downscale global climate model forecasts for use in crop models. Forecast calibration means to (1) minimise systematic error such as forecast bias, (2) ensure forecast uncertainty is reliably conveyed by ensemble spread, and (3) ensure forecasts are at least as skilful as climatology. Downscaling means, depending on the context, either: (1) producing a revised forecast with the correct local weather variability at a spatial scale smaller than the GCM grid (2) producing a local forecast based on large-scale climate drivers (e.g. sea surface temperature patterns) (this approach is also referred to as bridging), or (3) spatial or temporal disaggregation of a forecast. Crop forecasting models require physically-coherent inputs of rainfall, temperature and solar radiation. Previous research has established the suitability of the Bayesian joint probability modelling approach for calibrating monthly and three-monthly rainfall forecasts from global climate models. The Bayesian joint probability modelling approach has not previously been applied to post-process temperature or solar radiation forecasts or to post-process multivariate forecasts. However, it is formed on the general assumption that the joint distribution of two or more variables can be modelled as a multivariate normal distribution in transformed space. It can theoretically be extended for multivariate forecast post-processing with a relevant transformation for each variable. Thus the first objective of this thesis is to develop and evaluate several strategies for calibrating multivariate global climate model forecasts using the Bayesian joint probability modelling approach. Three strategies are compared: (1) simultaneous calibration of multiple climate variables in a single statistical model, which explicitly models inter-variable dependence via the covariance matrix; (2) univariate calibration coupled with an empirical ensemble reordering method (the Schaake Shuffle) that injects inter-variable dependence from historical data; and (3) quantile-mapping, which borrows inter-variable dependence from the raw forecasts. Applied to Australian seasonal (three-month) forecasts from the European Centre for Medium-range Weather Forecasts System4 model, univariate calibration paired with the Schaake Shuffle performs best in terms of univariate and multivariate forecast verification metrics. Direct multivariate calibration is the second-best method, with its far superior performance in in-sample testing vanishing in cross-validation, likely because of insufficient data to reliably infer the sizeable covariance matrix. Bayesian joint probability post-processing is confirmed to outperform quantile-mapping. Hence the Bayesian joint probability modelling approach and the Schaake Shuffle should, therefore, be preferred to quantile-mapping as a basis for calibrating GCM forecasts for crop forecasting applications. Global climate model forecast skill is best captured by post-processing on seasonal time scales. However, crop models require daily forecast sequences. Also, it is observed that some operational crop forecasting systems run separate crop models for multiple locations within a region and then aggregate the results into a regional forecast. Therefore, spatial forecasts are also needed. Accordingly, the second objective of this thesis is to develop and evaluate downscaling and disaggregation methods for post-processing global climate model forecasts to higher spatial and temporal resolutions. To this end, I develop an empirical multivariate downscaling method that imparts observed spatial, temporal and inter-variable relationships into disaggregated forecasts whilst completely preserving the joint distribution of forecasts post-processed at coarser spatial and/or temporal scales. Specifically, a Euclidean distance metric is devised to identify a nearest-neighbour in historical observations for each forecast ensemble member. The method of fragments is subsequently applied to simultaneously disaggregate the forecast spatial and temporally. The new method is demonstrated to perform well for downscaling skilful forecasts of rainfall, temperature and solar radiation for six locations in northeast Australia. The climatological distributions of the downscaled forecasts mirror observations and the observed frequency of wet days is also reproduced in forecasts. The new downscaling method is a step towards full integration of calibrated seasonal climate forecasts into crop models and has a significant advantage over quantile-mapping in that it can be applied for multiple sub-grid locations. The final objective of this thesis is to feed global climate model forecasts, post-processed using the new methods, to a crop decision support system to demonstrate an end-to-end solution for linking global climate model forecasts with a crop model to produce yield forecasts. The first crop forecasting application of the new methods is for sugarcane yield forecasting in Tully. The region is selected because it is a non-irrigated region, and it is thus suitable for assessing the value of climate forecasts. Two sets of post-processed forecasts are produced for the Tully Mill weather station in North-east Queensland. The first set is obtained by applying the Bayesian joint probability modelling approach to calibrate monthly rainfall, temperature and solar radiation forecasts for the grid cell containing Tully. The second set is obtained by using global climate model forecasts of the Nino 3.4 climate index (commonly associated with the El Nino Southern Oscillation), also using the Bayesian joint probability modelling approach, to produce local forecasts of monthly rainfall, temperature and solar radiation. In both cases, the monthly forecasts are subjected to the Schaake Shuffle and subsequently downscaled to daily sequences using identical methods. The calibration and bridging forecasts are used to drive a sugarcane crop model to generate long-lead forecasts of biomass in north-eastern Australia from 1982-2016. A rigorous probabilistic assessment of forecast attributes suggests that the calibration forecasts provide the most skilful forecasts overall although the bridging forecasts give more skilful yield forecasts at certain times. The biomass forecasts are unbiased and reliable for short to long lead times, suggesting that the new downscaling methods are effective. My end-to-end solution for linking global climate model forecasts and crop models enables quantitative modelling and risk management at the farm level. It has the potential to improve farm productivity and profitability through better decisions. Future research should investigate the value of the post-processing methods for a wide range of crops.

The global model analysis and forecast have a significant impact on the regional model predictions, as global model provides the initial and lateral boundary condition to regional model. This study addresses an important question whether the regional model can improve the short-range weather forecast as compared to the global model. The National Centers for Environmental Prediction (NCEP) Global Forecasting System (GFS) and the Weather Research and Forecasting (WRF) model are used in this study to evaluate the performance of global and regional models over the Indian region. A 24-h temperature and specific humidity forecast from the NCEP GFS model show less error compared to WRF model forecast. Rainfall prediction is improved over the Indian landmass when WRF model is used for rainfall forecast. Moreover, the results showed that high-resolution global model analysis (GFS4) improved the regional model forecast as compared to low-resolution global model analysis (GFS3).

Today the European Centre for Medium Range Weather Forecasts (ECMWF) runs a 16 km global T1279 operational weather forecast model using 1536 cores of an IBM Power7. Following the historical evolution in resolution upgrades, the ECMWF could expect to be running a 2.5 km global forecast model by 2030 on an exascale system that should be available and hopefully affordable by then. To achieve this would require the Integrated Forecasting System (IFS) to run efficiently on about 1000 times the number of cores it uses today. In a step towards this goal, the ECMWF have demonstrated the IFS running a 10 km global model efficiently on over 40,000 cores of HECToR a Cray XE6 at the Edinburgh Parallel Computing Centre. However, getting to over a million cores remains a formidable challenge, and many scalability improvements have yet to be implemented. The ECMWF is exploring the use of Fortran2008 coarrays; in particular, it is possibly the first time that coarrays have been used in a world-leading production application within the context of OpenMP parallel regions. The purpose of these optimisations is primarily to allow the overlap of computation and communication, and further, in the semi-Lagrangian advection scheme, to reduce the volume of data communicated. The importance of this research is such that if these and other planned developments are successful, the IFS model may continue to use the spectral transform method to 2030 and beyond on an exascale-sized system. The current status of the coarray scalability developments within the IFS are described together with a brief outline of future developments.

Today the European Centre for Medium-Range Weather Forecasts (ECMWF) runs a 16 km global T1279 operational weather forecast model using 1,536 cores of an IBM Power7. Following the historical evolution in resolution upgrades, ECMWF could expect to be running a 2.5 km global forecast model by 2030 on an Exascale system that should be available and hopefully affordable by then. To achieve this would require IFS to run efficiently on about 1000 times the number of cores it uses today. This is a significant challenge, one that we are addressing within the CRESTA project. After implementing an initial set of improvements ECMWF has now demonstrated IFS running a 10 km global model efficiently on over 50,000 cores of HECToR, a Cray XE6 at EPCC, Edinburgh. Of course, getting to over a million cores remains a formidable challenge, and many scalability improvements have yet to be implemented. Within CRESTA, ECMWF is exploring the use of Fortran2008 coarrays; in particular it is possibly the first time that coarrays have been used in a world leading production application within the context of OpenMP parallel regions. The purpose of these optimizations is primarily to allow the overlap of computation and communication, and further, in the semi-Lagrangian advection scheme, to reduce the volume of data communicated. The importance of this research is such that if these developments are successful then the IFS model may continue to use the spectral method to 2030 and beyond on an Exascale sized system.

For the past thirty years, the need for ever greater supercomputer performance has driven the development of many computing technologies which have subsequently been exploited in the mass market. Delivering an exaflop (or a million million million calculations per second) by the end of this decade is the challenge that the supercomputing community worldwide has set itself. Developing techniques and solutions which address the most difficult challenges that computing at the exascale can provide is a real challenge. Equipment vendors, programming tools providers, academics, and end users must all work together to build and to develop the development and debugging environment, algorithms and libraries, user tools, and the underpinning and cross-cutting technologies required to support the execution of applications at the exascale. This special issue of the journal is dedicated to advances in high performance computing in engineering and the way to exascale. It contains some papers which have been selected from the Exascale Applications and Software Conference (EASC2013) held on 9–11 April 2013 in Edinburgh, United Kingdom. The issue contains five papers, illustrates the recent advances made in the exascale path and covers algorithms, implementations and applications to solve large scale engineering problems. The first paper, by Reverdy et al., reports the realisation of the first cosmological simulations on the scale of the whole observable universe. The paper first focuses on the numerical aspects of two new simulations. In practice, each one of these simulations has evolved 550 billion dark matter particles in an adaptive mesh refinement grid, and one of the new simulations has pushed back the total number of grid points from 2000 billion for the L Cold Dark Matter (L CDM) model to 2200 billion due to the formation of a larger number of structures. The authors highlight the optimisations and adjustments required to run such a set of simulations and then summarise some important lessons learnt toward future exascale computing projects. Numerical examples illustrate the effectiveness of the procedure on 4752 nodes of the Curie Bull supercomputer. The second paper, by Mozdzynski et al., presents the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) enhanced to use Fortran2008 coarrays to overlap computation and communication in the context of OpenMP parallel regions. Today ECMWF runs a 16 km global T1279 operational weather forecast model using 1536 cores. Following the historical evolution in resolution upgrades, ECMWF could expect to be running a 2.5 km global forecast model by 2030 on an exascale system that should be available and hopefully affordable by then. To achieve this would require IFS to run efficiently on about 1000 times the number of cores it uses today. This is a significant challenge that is addressed in this paper, where, after implementing an initial set of improvements, ECMWF is demonstrated running a 10 km global model efficiently on over 40,000 cores on the HECToR Cray XE6 supercomputer. The third paper, by Gray et al., describes a multiGPU implementation of the Ludwig application, which specialises in simulating a variety of complex fluids via lattice Boltzmann fluid dynamics coupled to additional physics describing complex fluid constituents. The authors describe the methodology in augmenting the original CPU version with GPU functionality in a maintainable fashion. After presenting several optimisations that maximise performance on the GPU architecture through tuning for the GPU memory hierarchy, they describe how to implement particles within the fluid in such a way as to avoid a major diversion of the CPU and GPU codebases, whilst minimising data transfer at each timestep. Numerical results show that the application demonstrates excellent scaling to at least 8192 GPUs in parallel (the largest system tested at the time of writing) on the Titan Cray XK7 supercomputer. Exascale computers are expected to have highly hierarchical architectures with nodes composed of multiple core processors (CPU) and accelerators (GPU). The different programming levels generate new difficulties and algorithms issues. The paper written by Magoules et al., presents Alinea, which stands for Advanced LINEar Algebra, a library well suited for hybrid CPU/

With the advent of Big Data, many application databases have produced large collections of related time series, which may share key time series properties in common. To exploit these similarities, Global Forecasting Models (GFM) that simultaneously learn from a set of time series have been introduced. This research studies aims to develop a series of novel forecasting methodologies, using Recurrent Neural Networks as the principal forecasting architecture to improve the forecasting accuracy of GFMs in different circumstances. Furthermore, this study demonstrates the empirical evidence of the proposed GFM architectures to address real-world forecasting challenges in the retail and health-care industries.

Microstructure observations in the Pacific cold tongue reveal that turbulence often penetrates into the thermocline, producing hundreds of watts per square meter of downward heat transport during nighttime and early morning. However, virtually all observations of this deep-cycle turbulence (DCT) are from 0°, 140°W. Here, a hierarchy of ocean process simulations, including submesoscale-permitting regional models and turbulence-permitting large-eddy simulations (LES) embedded in a regional model, provide insight into mixing and DCT at and beyond 0°, 140°W. A regional hindcast quantifies the spatiotemporal variability of subsurface turbulent heat fluxes throughout the cold tongue from 1999 to 2016. Mean subsurface turbulent fluxes are strongest (∼100 W m−2) within 2° of the equator, slightly (∼10 W m−2) stronger in the northern than Southern Hemisphere throughout the cold tongue, and correlated with surface heat fluxes (r2= 0.7). The seasonal cycle of the subsurface heat flux, which does not covary with the surface heat flux, ranges from 150 W m−2near the equator to 30 and 10 W m−2at 4°N and 4°S, respectively. Aseasonal variability of the subsurface heat flux is logarithmically distributed, covaries spatially with the time-mean flux, and is highlighted in 34-day LES of boreal autumn at 0° and 3°N, 140°W. Intense DCT occurs frequently above the undercurrent at 0° and intermittently at 3°N. Daily mean heat fluxes scale with the bulk vertical shear and the wind stress, which together explain ∼90% of the daily variance across both LES. Observational validation of the scaling at 0°, 140°W is encouraging, but observations beyond 0°, 140°W are needed to facilitate refinement of mixing parameterization in ocean models.Significance StatementThis work is a fundamental contribution to a broad community effort to improve global long-range weather and climate forecast models used for seasonal to longer-term prediction. Much of the predictability on seasonal time scales is derived from the slow evolution of the upper eastern equatorial Pacific Ocean as it varies between El Niño and La Niña conditions. This study presents state-of-the-art high-resolution regional numerical simulations of ocean turbulence and mixing in the eastern equatorial Pacific. The results inform future planning for field work as well as future efforts to refine the representation of ocean mixing in global forecast models.

Despite significant progress made in snowfall estimation from space, methods utilizing passive microwave measurements continue to be plagued by low detectability compared to those that estimate rainfall. This article presents a hybrid snowfall detection algorithm that combines the output from a statistical algorithm utilizing satellite passive microwave measurements with the output from a statistical algorithm trained with in situ data that uses meteorological variables derived from a global forecast model as predictors. The satellite algorithm computes the probability of snowfall over land using logistic regression and the principal components of the high‐frequency brightness‐temperature measurements at AMSU/MHS and ATMS channel frequencies 89 GHz and above. In a separate investigation, analysis of modelled data derived from NOAA's Global Forecast System (GFS) showed that cloud thickness and relative humidity at 1 to 3 km height were the best predictors of snowfall occurrence. A statistical logistical regression model that combined cloud thickness, relative humidity and vertical velocity was selected among statistically significant variants as the one with the highest overall classification accuracy. Next, the weather‐based and satellite model outputs were combined in a weighting scheme to produce a final probability of snowfall output, which was then used to classify a weather event as “snowing” or “not snowing” based on an a priori threshold probability. Statistical analysis indicated that a scheme with equal weights applied to the weather‐based and satellite model significantly improved satellite snowfall detection. Example applications of the hybrid algorithm over continental USA demonstrated the improvement for a major snowfall event and for an event dominated by lighter snowfall.

Accurate electricity demand forecasts play a key role in sustainable power systems. To enable better decision-making especially for demand flexibility of the end-user, it is necessary to provide not only accurate but also understandable and actionable forecasts. To provide accurate forecasts Global Forecasting Models (GFM) that are trained across time series have shown superior results in many demand forecasting competitions and real-world applications recently, compared with univariate forecasting approaches. We aim to fill the gap between the accuracy and the interpretability in global forecasting approaches. In order to explain the global model forecasts, we propose Local Interpretable Model-agnostic Rule-based Explanations for Forecasting (LIMREF), which is a local explainer framework that produces k-optimal impact rules for a particular forecast, considering the global forecasting model as a black-box model, in a model-agnostic way. It provides different types of rules which explain the forecast of the global model and the counterfactual rules, which provide actionable insights for potential changes to obtain different outputs for given instances. We conduct experiments using a large-scale electricity demand dataset with exogenous features such as temperature and calendar effects. Here, we evaluate the quality of the explanations produced by the LIMREF framework in terms of both qualitative and quantitative aspects such as accuracy, fidelity and comprehensibility, and benchmark those against other local explainers.

A spring-runoff forecast model for central Arizona was developed based on multiple discriminant analysis. More than 6500 potential predictor variables were analyzed, including local precipitation and temperature variables, as well as global sea level pressure variables. The forecast model was evaluated on nine years exclusive of the years on which the model was based. Forecasts are provided in the form of a cumulative distribution function (cdf) of the expected runoff, based on analogs. A ranked probability score to evaluate forecast skill for the cdf forecasts was developed. Ranked probability skill scores ranged from 25% to 45%. Local and global forecast models were developed and compared to the combined data source model. The global forecast model was equivalent in skill to the local forecast model. The combined model exhibited a marked improvement in skill over either the local or global models. Three recurrent patterns in the predictor variables used by the forecast model are analyzed in som...

Retailers must have accurate sales forecasts to efficiently and effectively operate their businesses and remain competitive in the marketplace. Global forecasting models like RNNs can be a powerful tool for forecasting in retail settings, where multiple time series are often interrelated and influenced by a variety of external factors. By including covariates in a forecasting model, we can often better capture the various factors that can influence sales in a retail setting. This can help improve the accuracy of our forecasts and enable better decision making for inventory management, purchasing, and other operational decisions. In this study, we investigate how the accuracy of global forecasting models is affected by the inclusion of different potential demand covariates. To ensure the significance of the study’s findings, we used the M5 forecasting competition’s openly accessible and well-established dataset. The results obtained from DeepAR models trained on different combinations of features indicate that the inclusion of time-, event-, and ID-related features consistently enhances the forecast accuracy. The optimal performance is attained when all these covariates are employed together, leading to a 1.8% improvement in RMSSE and a 6.5% improvement in MASE compared to the baseline model without features. It is noteworthy that all DeepAR models, both with and without covariates, exhibit a significantly superior forecasting performance in comparison to the seasonal naïve benchmark.