Quality Of Forecasts Research Articles

Probabilistic Solar Forecasts as a Binary Event Using a Sky Camera

With the fast increase of solar energy plants, a high-quality short-term forecast is required to smoothly integrate their production in the electricity grids. Usually, forecasting systems predict the future solar energy as a continuous variable. But for particular applications, such as concentrated solar plants with tracking devices, the operator needs to anticipate the achievement of a solar irradiance threshold to start or stop their system. In this case, binary forecasts are more relevant. Moreover, while most forecasting systems are deterministic, the probabilistic approach provides additional information about their inherent uncertainty that is essential for decision-making. The objective of this work is to propose a methodology to generate probabilistic solar forecasts as a binary event for very short-term horizons between 1 and 30 min. Among the various techniques developed to predict the solar potential for the next few minutes, sky imagery is one of the most promising. Therefore, we propose in this work to combine a state-of-the-art model based on a sky camera and a discrete choice model to predict the probability of an irradiance threshold suitable for plant operators. Two well-known parametric discrete choice models, logit and probit models, and a machine learning technique, random forest, were tested to post-process the deterministic forecast derived from sky images. All three models significantly improve the quality of the original deterministic forecast. However, random forest gives the best results and especially provides reliable probability predictions.

Machine learning-based predictive modelling for the enhancement of wine quality

The certification of wine quality is essential to the wine industry. The main goal of this work is to develop a machine learning model to forecast wine quality using the dataset. We utilised samples from the red wine dataset (RWD) with eleven distinct physiochemical properties. With the initial RWD, five machine learning (ML) models were trained and put to the test. The most accurate algorithms are Random Forest (RF) and Extreme Gradient Boosting (XGBoost). Using these two ML approaches, the top three features from a total of eleven features are chosen, and ML analysis is performed on the remaining features. Several graphs are employed to demonstrate the feature importance based on the XGBoost model and RF. Wine quality was predicted using relevant characteristics, often referred to as fundamental elements, that were shown to be essential during the feature selection procedure. When trained and tested without feature selection, with feature selection (RF), and with key attributes, the XGBoost classifier displayed 100% accuracy. In the presence of essential variables, the RF classifier performed better. Finally, to assess the precision of their predictions, the authors trained an RF classifier, validated it, and changed its hyperparameters. To address collinearity and decrease the quantity of predictors without sacrificing model accuracy, we have also used cluster analysis.

Using Object-Based Verification to Assess Improvements in Forecasts of Convective Storms between Operational HRRR Versions 3 and 4

Abstract The object-based verification procedure described in a recent paper by Duda and Turner was expanded herein to compare forecasts of composite reflectivity and 6-h precipitation objects between the two most recent operational versions of the High-Resolution Rapid Refresh (HRRR) model, versions 3 and 4, over an expanded set of warm season cases in 2019 and 2020. In addition to analyzing all objects, a reduced set of forecast–observation object pairs was constructed by taking the best forecast match to a given observation object for the purposes of bias-reduction and unequivocal object comparison. Despite the apparent signal of improved scalar metrics such as the object-based threat score in HRRRv4 compared to HRRRv3, no statistically significant differences were found between the models. Nonetheless, many object attribute comparisons revealed indications of improved forecast performance in HRRRv4 compared to HRRRv3. For example, HRRRv4 had a reduced overforecasting bias for medium- and large-sized reflectivity objects, and all objects during the afternoon. HRRRv4 also better replicated the distribution of object complexity and aspect ratio. Results for 6-h precipitation also suggested superior performance in HRRRv4 over HRRRv3. However, HRRRv4 was worse with centroid displacement errors and more severely overforecast objects with a high maximum precipitation amount. Overall, this exercise revealed multiple forecast deficiencies in the HRRR, which enables developers to direct development efforts on detailed and specific endeavors to improve model forecasts. Significance Statement This work builds upon the authors’ prior work in assessing model forecast quality using an alternative verification method—object-based verification. In this paper we verified two versions of the same model (one an upgrade from the other) that were making forecasts covering the same time window, using the object-based verification method. We found that the updated model was not statistically significantly better, although there were indications it performed better in certain aspects such as capturing the change in the number of storms during the daytime. We were able to identify specific problem areas in the models, which helps us direct model developers in their efforts to further improve the model.

Влияние короткопериодных вариаций атмосферных воздействий на крупномасштабную структуру океанографических полей

The dependence of the large-scale structure of oceanographic fields on short-term variations in atmospheric forcing (AF) at the ocean surface has been studied. Estimates of this dependence were obtained by comparing the results of two numerical experiments with the NEMO model in the ORCA1 configuration used as a computational core in the data assimilation system at the Hydrometeorological Center of Russia. The experiments similar in all other respects differed only in the discreteness of the AF data: 3–24 hours in the main experiment and 1 month in the experiment with time-averaged AF. Large-scale characteristics were compared: kinetic energy averaged over large regions, Atlantic Meridional Overturning Circulation, meridional heat transport, mixed layer depth, sea ice volume. It is shown that neglecting high-frequency AF variability can significantly distort the long-term evolution of oceanographic fields reproduced by ocean models and thereby affect the quality of model forecasts. Keywords: oceanographic fields, large-scale structure, NEMO model, atmospheric forcing, short-term variations

Corporate Governance Reforms and Analyst Forecasts: International Evidence

In this study, we examine the effect of worldwide corporate governance reforms on analyst forecasts using data on firms from 41 countries that have implemented such reforms. Employing a difference‐in‐differences design, we find robust evidence of a significant positive effect of these reforms, which mainly promote the independence of the audit committee and auditors, on analyst forecast accuracy. We also find significant improvements in post‐reform corporate governance structure and the quantity and quality of corporate disclosure, which validates the predicted economic mechanism of the impact of these reforms. Moreover, the reforms are effective in reducing analyst forecast bias and dispersion, suggesting an overall improvement in analyst forecast quality after reform implementation. Furthermore, we find a moderating effect of analyst general experience on the relationship of interest, which is consistent with the conjecture that improved corporate disclosures following reforms could be more beneficial for analysts with less experience. Overall, our findings shed new insights into how country‐level corporate governance reforms worldwide shape firms’ information environment.

An Ensemble Approach for Intra-Hour Forecasting of Solar Resource

Solar resource forecasting is an essential step towards smart management of power grids. This study aims to increase the performance of intra-hour forecasts. For this, a novel ensemble model, combining statistical extrapolation of time-series measurements with models based on machine learning and all-sky imagery, is proposed. This study is conducted with high-quality data and high-resolution sky images recorded on the Solar Platform of the West University of Timisoara, Romania. Atmospheric factors that contribute to improving or reducing the quality of forecasts are discussed. Generally, the statistical models gain a small skill score across all forecast horizons (5 to 30 min). The machine-learning-based methods perform best at smaller forecast horizons (less than 15 min), while the all-sky-imagery-based model performs best at larger forecast horizons. Overall, for forecast horizons between 10 and 30 min, the weighted forecast ensemble with frozen coefficients achieves a skill score between 15 and 20%.

Pattern detection and prediction using deep learning for intelligent decision support to identify fish behaviour in aquaculture

The United Nations’ Sustainable Development Goals (SDGs) underscore aquaculture as a sustainable practice. However, many countries rely on ineffective conventional methods to monitor the aquaculture. Fish behaviour monitoring systems lack an integrated strategy, often focusing on underwater footage analysis or water quality data. Additionally, limited research exists on using learning algorithms to identify sensor data anomalies, limiting the effectiveness of such systems. This study proposes a comprehensive and effective aquaculture monitoring system to address these issues. The work introduces a novel method integrating fish motion analysis with water quality metrics. The study also presents a learning system that anticipates potential water quality parameters. The proposed model has the potential to significantly enhance aquaculture management procedures and contribute to the UN’s SDGs through its ability to forecast water quality parameters and detect abnormal behaviour without requiring training detection and tracking algorithms.The study employs a large aquarium to capture water quality parameters and fish movement patterns. The study develops an FPGA-based water quality monitoring system and a prediction model to forecast potential water quality metrics. Among multiple LSTM models, the bi-directional LSTM performs well, achieving an RMSE score of 1.01 for water quality prediction. Subsequently, an outlier detection system is applied to identify unusual water quality parameters, triggering the behaviour analysis model. The behaviour analysis model uses an auto-encoder-based reconstruction approach to detect the frantic patterns in fish. The suggested behaviour analysis model has a f1-score of 0.68, demonstrating its reliability in contrast to conventional methods. The developed system is also tested on aquaculture site videos and includes a novel dataset of Sillago sihama with annotations for frantic behaviour patterns. The system works effectively as a decision support system for aquaculture.

Improving the quality of coffee yield forecasting in Dak Lak Province, Vietnam, through the utilization of remote sensing data

This research aimed to identify sensitive areas for Robusta coffee trees in Dak Lak province, Vietnam, where frequent droughts caused fluctuations in productivity. To improve yield forecasting, a mask was developed to extract potential predictive variables from satellite-derived vegetation indices (VIs). Correlation coefficients between VIs and coffee yield were analyzed to determine sensitive areas, and grid cells with high multiple correlation coefficients and a variable over time were used to build the mask for extracting VIs as predictor variables. The study found that sensitive areas had more challenging farming conditions than long-term crops, and the Vegetation Health Index was the most appropriate index for predicting coffee yield. The forecast quality for 6-8 months in advance was relatively high, with a ‘Willmott’s index of agreement’ ranging from 0.85 to 0.97 and the Mean Absolute Percentage Error ranging from 4.9% to 7.5%. Compared to previous research, the forecast quality has significantly improved. This study provides valuable insights for predicting coffee yield in Dak Lak and highlights the importance of considering sensitive areas and VIs for accurate forecasting.

Information-Based Probabilistic Verification Scores for Two-Dimensional Ensemble Forecast Data: A Madden–Julian Oscillation Index Example

Abstract Probabilistic forecasting is a common activity in many fields of the Earth sciences. Assessing the quality of probabilistic forecasts—probabilistic forecast verification—is therefore an essential task in these activities. Numerous methods and metrics have been proposed for this purpose; however, the probabilistic verification of vector variables of ensemble forecasts has received less attention than others. Here we introduce a new approach that is applicable for verifying ensemble forecasts of continuous, scalar, and two-dimensional vector data. The proposed method uses a fixed-radius near-neighbors search to compute two information-based scores, the ignorance score (the logarithmic score) and the information gain, which quantifies the skill gain from the reference forecast. Basic characteristics of the proposed scores were examined using idealized Monte Carlo simulations. The results indicated that both the continuous ranked probability score (CRPS) and the proposed score with a relatively small ensemble size (<25) are not proper in terms of the forecast dispersion. The proposed verification method was successfully used to verify the Madden–Julian oscillation index, which is a two-dimensional quantity. The proposed method is expected to advance probabilistic ensemble forecasts in various fields. Significance Statement In the Earth sciences, stochastic future states are estimated by solving a large number of forecasts (called ensemble forecasts) based on physical equations with slightly different initial conditions and stochastic parameters. The verification of probabilistic forecasts is an essential part of forecasting and modeling activity in the Earth sciences. However, there is no information-based probabilistic verification score applicable for vector variables of ensemble forecasts. The purpose of this study is to introduce a novel method for verifying scalar and two-dimensional vector variables of ensemble forecasts. The proposed method offers a new approach to probabilistic verification and is expected to advance probabilistic ensemble forecasts in various fields.

Evaluation of an Improved AROME Configuration for Fog Forecasts during the SOFOG3D Campaign

Abstract This paper evaluates fog forecasts of a new AROME configuration dedicated to fog thanks to observations of the recent field campaign SOuth westFOGs 3Dimensions (SOFOG3D). This new configuration takes advantage of an upgraded horizontal and vertical resolution of a two-moment microphysical scheme [Liquid Ice Multiple Aerosols (LIMA)], and of the inclusion of a parameterization of the droplet’s deposition onto vegetation. A statistical study conducted over the 6 months of the SOFOG3D field campaign allowed the evaluation of the quality of fog forecasts produced by this new configuration to compare it to the current operational configuration of AROME. The main findings are as follows: the new configuration forecasts more fog events, with a few more false alarms, but improved the amount of fog events with low top height and with a low water content, underestimated by the reference configuration. The importance of the first level height for a good representation of the first few meters above the ground is crucial to improve the fog formation forecast. A delay of fog dissipation in the morning was highlighted in operational simulations and slightly reduced thanks to LIMA. This two-moment scheme produced thinner fogs, with less water content. These are more realistic, compared with observations, and thinner fog is also easier for solar radiation to dissipate.

Уточнення чисельних метеорологічних прогнозів за допомогою глибокого навчання

This paper briefly describes the history of numerical weather prediction development. The difficulties, which occur in the modelling of atmospheric processes, their nature and possible ways of their mitigation, are described. It also indicates alternative methods of improving the quality of meteorological forecasts. A brief history of deep learning and possible ways of its application to meteorological problems are given. Then, the paper describes the format used to store the 2m temperature forecasts of the COSMO numerical regional model. The proposed neural network architecture enables correcting the forecast errors of the numerical model. We conducted the experiments on the data of eight meteorological stations of the Kyiv region, so we obtained eight trained neural network models. The results showed that the proposed architecture enables obtaining better-quality forecasts in more than 50% of cases. Root-mean-square errors of the resulting forecasts decreased, and it is a widespread skill-score of improved-quality forecasts in meteorological science.

A real-time calibration method for the numerical pollen forecast model COSMO-ART

Technologies for monitoring pollen concentrations in real-time made substantial advances in the past years and become increasingly available. This opens the possibility to calibrate numerical pollen forecast models in real-time and make a significant step forward regarding the quality of pollen forecasts. We present a method to use real-time pollen measurements in numerical pollen forecast models. The main idea is to calibrate model parameterizations and not to assimilate measurements in a nudging sense. This ensures that the positive effect persists throughout the forecast period and does not vanish after a few forecast hours. We propose to adapt in real-time both the model phenology scheme and the overall tuning factor that are present in any numerical pollen forecast model. To test this approach, we used the numerical pollen forecast model COSMO-ART (COnsortium for Small-scale MOdelling-Aerosols and Reactive Trace gases) on a mesh size of 1.1 km covering the greater Alpine domain. Test runs covered two pollen seasons and included Corylus, Alnus, Betula and Poaceae pollen. Comparison with daily measurements from 13 Swiss pollen stations revealed that the model improvements are large, but fine-tuning of the method remains a challenge. We conclude that the presented approach is a first valuable step towards comprehensive real-time calibration of numerical pollen forecast models.

Mixed-Order Relation-Aware Recurrent Neural Networks for Spatio-Temporal Forecasting

Spatio-temporal forecasting has a wide range of applications in smart city efforts, such as traffic forecasting and air quality prediction. Graph Convolutional Recurrent Neural Networks (GCRNN) are the state-of-the-art methods for this problem, which learn temporal dependencies by RNNs and exploit pairwise node proximity to model spatial dependencies. However, the spatial relations in real data are not simply pairwise but sometimes in a higher order among multiple nodes. Moreover, spatio-temporal sequences deriving from nature are often regulated by known or unknown physical laws. GCRNNs rarely take into account the underlying physics in real-world systems, which may result in degenerated performance. To address these issues, we devise a general model called Mixed-Order Relation-Aware RNN (MixRNN+) for spatio-temporal forecasting. Specifically, our MixRNN+ captures the complex mixed-order spatial relations of nodes through a newly proposed building block called Mixer, and simultaneously addressing underlying physics by the integration of a new residual update strategy. Experimental results on three forecasting tasks in smart city applications (i.e., traffic speed, taxi flow, and air quality prediction) demonstrate the superiority of our model against the state-of-the-art methods.

A study into the relationship between the type of Japanese candlesticks of a nationally significant exchange-traded asset and investor sentiment in the national financial market

Subject. This article examines the relationship between the type of Japanese candlesticks of a nationally significant exchange-traded asset and investor sentiment in the national financial market. Objectives. The article aims to determine the relationship between the type of Japanese candlesticks for a nationally significant asset traded on the national stock exchange and the sentiments of financial market participants. Methods. For the study, we used the methods of induction and deduction, statistical methods for processing the results of observations, and the method of physiognomy in determining the emotions of persons making investment decisions. Results. The article determines the type of Japanese candlesticks for a nationally significant asset traded on the national stock exchange and the sentiments of financial market participants. Conclusions. When taking into account new indicators, the analyst has the means to bring the forecast result closer to real conditions, when, in addition to traditional indicators, combinations of technical analysis tools and psychological indicators in the markets give a new interpretation of the events taking place there. This improves the quality of analytical work and its results. The results of the research described in the article can be applied in the work of a market analyst. The use of new indicators complements and expands the classic analytical tools, improves the quality of market forecasts.

Advancing Sugarcane Disease Detection through CNN-Based Deep Learning

Agriculture produce especially sugarcane crop is no exception to diseases as compared to the other crops. Sugarcane, a vital cash crop for the global sugar industry, faces numerous challenges, with the Top Borer disease. Disease prone sugarcane crop directly affects the production quality and quantity. Sugarcane infections are a cause of worry for the farmers because they can wipe out the entire crop field. Researchers are working on applying Artificial Intelligence (AI) techniques, like Machine Learning (ML) and Deep Learning (DL), to analyse the agricultural data (yield prediction, selling price forecasting, climate, and soil quality etc.) and prevent crop damage due to various reasons, diseases being one of them. Deep neural network which includes Convolutional Neural Network (CNN) is a modern technique for agricultural disease detection. Hence, this paper presents the feasibility study and the effectiveness of DL based CNN algorithm in the disease detection of crops with special reference to selective four diseases of sugarcane crop in India. The proposed system integrates state-of-the-art deep learning algorithms, leveraging Convolutional Neural Networks (CNNs) and recurrent models, to analyze high-resolution images captured by unmanned aerial vehicles (UAVs) or ground-based sensors. These images provide a comprehensive view of the sugarcane plantation, allowing for the identification of subtle symptoms and early-stage infections that may go unnoticed by the human eye. The key components of the developed system include a robust image preprocessing pipeline to enhance the quality of input data, a customized deep neural network architecture trained on a diverse dataset of sugarcane images, and a real-time monitoring system for timely intervention. The model's performance is evaluated on a large-scale dataset collected from sugarcane plantations across diverse geographic regions. The results demonstrate the system's high accuracy in detecting and classifying the Top Borer disease, outperforming traditional methods.

Прогнозирование цен на рынке жилья в условиях изменения основной тенденции

The development of digital technologies contributes to the growth of the use of nonparametric methods. The presented study proposes a methodology for the forecast assessment of prices in the residential real estate market, taking into account the possible determination of the direction of dynamics in the anticipation period based on the application of nonparametric Nadaraya – Watson estimation. The forecast model construction in the work is considered on the basis of the historically established tendency of determination of the price level of the primary or secondary housing market of Krasnodar Krai and does not take into account other factors. Par-ticular attention is paid to the application of the Chow test to identify the point in time at which there was a struc-tural shift, which allows to determine the period devoid of structural breaks for modeling the trend in order to determine the confidence interval of the forecast. The existing housing problem reflects the relevance of the development of methods for forecasting price dynamics in the housing market, and the absence of additional factors reduces the error and increases the forecast quality.

Forecasting Inflation in Russia Using a TVP Model with Bayesian Shrinkage

The paper emphasizes the relevance of improving methodological tools for macroeconomic forecasting. In particular, it is pointed out, that models with a large number of explanatory variables on relatively short samples can often overfit in-sample and, thus, forecast poorly. The article reviews studies on forecasting inflation in Russia and explains the applicability of the model with Bayesian shrinkage of time-varying parameters based on hierarchical normal-gamma prior. Models of this type allow for possible nonlinearities in relationships between regressors and inflation and, at the same time, can deal with the problem of overfitting.The choice of a system of statistical indicators used to forecast monthly inflation in Russia during the period 2011–2022 is substantiated. It is shown that at short forecast horizons (of one to three months) Bayesian normal-gamma shrinkage TVP model with a large set of inflation predictors outperforms in forecasting accuracy, measured by mean absolute and squared errors, its linear counterpart, linear and Bayesian autoregression models without predictors, as well as naive models (based on random walk). At the horizon of six months, the autoregression model with Bayesian shrinkage exhibits the best forecast performance. As the forecast horizon rises (up to one year), statistical differences in the quality of forecasts of competing models of Russian inflation decrease.The developed method can be used by the Bank of Russia and executive authorities for rapid assessment of inflation forecasts until the end of the year in order to evaluate risks of inflation deviation from the target level and elaborate preventive economic policy measures.

Inventory control strategy: based on demand forecast error

PurposeThis study examines the relationship between demand forecasting error and retail inventory management in an uncertain supplier yield context. Replenishment is segmented into off-season and peak-season, with the former characterized by longer lead times and higher supply uncertainty. In contrast, the latter incurs higher acquisition costs but ensures certain supply, with the retailer's purchase volume aligning with the acquired volume. Retailers can replenish in both phases, receiving goods before the sales season. This paper focuses on the impact of the retailer's demand forecasting bias on their sales period profits for both phases.Design/methodology/approachThis study adopts a data-driven research approach by drawing inspiration from real data provided by a cooperating enterprise to address research problems. Mathematical modeling is employed to solve the problems, and the resulting optimal strategies are tested and validated in real-world scenarios. Furthermore, the applicability of the optimal strategies is enhanced by incorporating numerical simulations under other general distributions.FindingsThe study's findings reveal that a greater disparity between predicted and actual demand distributions can significantly reduce the profits that a retailer-supplier system can earn, with the optimal purchase volume also being affected. Moreover, the paper shows that the mean of the forecasting error has a more substantial impact on system revenue than the variance of the forecasting error. Specifically, the larger the absolute difference between the predicted and actual means, the lower the system revenue. As a result, managers should focus on improving the quality of demand forecasting, especially the accuracy of mean forecasting, when making replenishment decisions.Practical implicationsThis study established a two-stage inventory optimization model that simultaneously considers random yield and demand forecast quality, and provides explicit expressions for optimal strategies under two specific demand distributions. Furthermore, the authors focused on how forecast error affects the optimal inventory strategy and obtained interesting properties of the optimal solution. In particular, the property that the optimal procurement quantity no longer changes with increasing forecast error under certain conditions is noteworthy, and has not been previously noted by scholars. Therefore, the study fills a gap in the literature.Originality/valueThis study established a two-stage inventory optimization model that simultaneously considers random yield and demand forecast quality, and provides explicit expressions for optimal strategies under two specific demand distributions. Furthermore, the authors focused on how forecast error affects the optimal inventory strategy and obtained interesting properties of the optimal solution. In particular, the property that the optimal procurement quantity no longer changes with increasing forecast error under certain conditions is noteworthy, and has not been previously noted by scholars. Therefore, the study fills a gap in the literature.

Military expenditure and problems of national economic growth: Theory and some macroeconomic conclusions from a special military operation

Subject. The article investigates the systemic crisis of the neoliberal market model and the role of military spending in the Russian economy development. Objectives. The focus is on critical analysis of the current system of national accounts, criteria and parameters for assessing the economic growth. Methods. The study employs methods of statistical and military-economic analysis. Results. The paper shows that the gross domestic product does not reflect the level of economic development of the country, since it does not take into account the results obtained at the intermediate stages of production. Conclusions. If considered, the findings will enable to improve the quality of military-political and macroeconomic forecasting.

Improving the accuracy of multimodel short-to-medium-range precipitation and streamflow forecasts over the Upper Bhima river basin, India

Accurate precipitation forecasting with sufficient lead time is a prerequisite for developing a robust flood warning system (FWS), which is very challenging, particularly in developing countries like India. This study evaluates the utility of the TIGGE multimodel ensemble meteorological forecasts over the Upper Bhima river basin and investigated the hydrological utility of the TIGGE forecasts through a calibrated hydrological (VIC-RAPID) model followed by the postprocessing of streamflow through Bayesian model average (BMA) approach. Results show that the quality of the meteorological forecasts of precipitation, and simulated streamflow deteriorated with increasing lead time, which can be ameliorated with a suitable bias-correction technique. The BMA-based postprocessing further improved the streamflow simulations, especially in case of extreme events, which highlighted its efficacy in flood forecasting. From the results of the study, it is recommended that a compound system of improved precipitation prediction, calibrated VIC-RAPID model and postprocessing of streamflows in an integrated manner would facilitate a reliable FWS for operational purposes.