Forecast Verification and Visualization based on Gaussian Mixture Model Co‐estimation

Wang, Y. H.; Fan, Chaoran; Zhang, J.; Niu, T.; Zhang, S.; Jiang, Junguang

doi:10.1111/cgf.12520

Cited by 9 publications

(7 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other task categories such as ‘ automatic detection of patterns, trends, and anomalies ’ relates to R6 [LPCRH18], ‘ prediction/forecasting future trends ’ corresponds to R8 [WFZ∗15], and ‘ identifying relationships between different variables ’ partially corresponds to R7 [WSL∗14]. The remaining categories do not have limited coverage in the visualization research but some of them do not correspond to any task requirements of domain experts.…”

Section: Taxonomiesmentioning

confidence: 99%

See 1 more Smart Citation

The State of the Art in Visual Analysis Approaches for Ocean and Atmospheric Datasets

Afzal

Hittawe

Ghani

et al. 2019

Computer Graphics Forum

View full text Add to dashboard Cite

The analysis of ocean and atmospheric datasets offers a unique set of challenges to scientists working in different application areas. These challenges include dealing with extremely large volumes of multidimensional data, supporting interactive visual analysis, ensembles exploration and visualization, exploring model sensitivities to inputs, mesoscale ocean features analysis, predictive analytics, heterogeneity and complexity of observational data, representing uncertainty, and many more. Researchers across disciplines collaborate to address such challenges, which led to significant research and development advances in ocean and atmospheric sciences, and also in several relevant areas such as visualization and visual analytics, big data analytics, machine learning and statistics. In this report, we perform an extensive survey of research advances in the visual analysis of ocean and atmospheric datasets. First, we survey the task requirements by conducting interviews with researchers, domain experts, and end users working with these datasets on a spectrum of analytics problems in the domain of ocean and atmospheric sciences. We then discuss existing models and frameworks related to data analysis, sense‐making, and knowledge discovery for visual analytics applications. We categorize the techniques, systems, and tools presented in the literature based on the taxonomies of task requirements, interaction methods, visualization techniques, machine learning and statistical methods, evaluation methods, data types, data dimensions and size, spatial scale and application areas. We then evaluate the task requirements identified based on our interviews with domain experts in the context of categorized research based on our taxonomies, and existing models and frameworks of visual analytics to determine the extent to which they fulfill these task requirements, and identify the gaps in current research. In the last part of this report, we summarize the trends, challenges, and opportunities for future research in this area. (see http://www.acm.org/about/class/class/2012)

show abstract

Section: Taxonomiesmentioning

confidence: 99%

“…Clustering is known as a prediction task; many applications used clustering in case there was no available labeled data [WFZ∗15]. The models created by clustering cannot be generalized, so, selecting a suitable similarity metric and validation are considered as the main challenges in clustering.…”

Section: Taxonomiesmentioning

confidence: 99%

The State of the Art in Visual Analysis Approaches for Ocean and Atmospheric Datasets

Afzal

Hittawe

Ghani

et al. 2019

Computer Graphics Forum

View full text Add to dashboard Cite

show abstract

“…The candidate marginal distribution models considered in this study are the Gaussian mixture model (GMM), gamma, lognormal, Pearson type III, Log Pearson type III, generalized extreme value, and Weibull distributions. These models are selected because they are widely used for fitting the distributions of precipitation, drought, and high temperature indicators (Angelidis et al, ; Hao et al, ; Wang et al, ; Yusof et al, ). The Expectation‐Maximization algorithm (EM; Sondergaard & Lermusiaux, ) is applied to generate parameters for the GMM distribution, and the parameters for other marginal distributions and copulas are obtained through a maximum likelihood estimation (Fan et al, ; Shih & Louis, ).…”

Section: Analysis Of Resultsmentioning

confidence: 99%

“…GCMs are commonly used to generate future climate projections under different Representative Concentration Pathway (RCPs; Wang et al, ). However, as GCMs are too coarse to assess regional or local scale climatic characteristics and may have large systematic biases, RCMs are usually employed to transform GCM outputs to a higher resolution prior to subsequent bias corrections (Wang et al, ; Zhou et al, ). Consequently, this study aims to establish a modeling system for investigating the potential future climate change impacts on hot droughts in the Loess Plateau.…”

Section: Introductionmentioning

confidence: 99%

Drought Occurring With Hot Extremes: Changes Under Future Climate Change on Loess Plateau, China

et al. 2019

View full text Add to dashboard Cite

Drought is one of the most widespread and destructive hazards over the Loess Plateau (LP) of China. Due to climate change, extremely high temperature accompanied with drought (expressed as hot drought) may lead to intensive losses of both properties and human deaths in future. A hot drought probabilistic recognition system is developed to investigate how potential future climate changes will impact the simultaneous occurrence of drought and hot extremes (hot days exceeding certain values) on the LP. Two regional climate models, coupled with multiple bias‐correction techniques and multivariate probabilistic inference, are innovative integrated into the hot drought probabilistic recognition system to reveal the concurrence risk of droughts and hot extremes under different Representative Concentration Pathway (RCP) scenarios. The hot‐day index, TX90p, indicating the number of days with daily maximum temperature (Tmax) exceeding the 90th percentile threshold, and the Standardized Precipitation Index are applied to identify the joint risks on the LP using copula‐based methods. The results show that precipitation will increase throughout most of the LP under both RCP4.5 and RCP8.5 scenarios of 2036–2095, while Tmax may increase significantly all over the LP (1.8–2.7 °C for RCP4.5 and 2.7–3.6 °C for RCP8.5). The joint return periods of Standardized Precipitation Index and TX90p show that fewer stations will experience severe drought with long‐term hot extremes in two future scenarios. However, some stations may experience hot droughts that are more frequent and extreme, particularly certain stations in the southwest and south‐central regions of the LP with recurrence period less than 10 years.

show abstract

“…In other applications statistical methods for clustering (Everitt, 1980) are used to identify and define objects within the field (Marzban andSandgathe, 2006, 2008). This clustering approach, which has been reexamined by Lakshmanan and Kain (2010) and more recently by Wang et al (2015), is the basis of the object-identification procedure used in the present work.…”

Section: Introductionmentioning

confidence: 99%

On the effect of model parameters on forecast objects

Marzban

Jones

et al. 2018

Geosci. Model Dev.

View full text Add to dashboard Cite

Abstract. Many physics-based numerical models produce a gridded, spatial field of forecasts, e.g., a temperature "map". The field for some quantities generally consists of spatially coherent and disconnected "objects". Such objects arise in many problems, including precipitation forecasts in atmospheric models, eddy currents in ocean models, and models of forest fires. Certain features of these objects (e.g., location, size, intensity, and shape) are generally of interest. Here, a methodology is developed for assessing the impact of model parameters on the features of forecast objects. The main ingredients of the methodology include the use of (1) Latin hypercube sampling for varying the values of the model parameters, (2) statistical clustering algorithms for identifying objects, (3) multivariate multiple regression for assessing the impact of multiple model parameters on the distribution (across the forecast domain) of object features, and (4) methods for reducing the number of hypothesis tests and controlling the resulting errors. The final "output" of the methodology is a series of box plots and confidence intervals that visually display the sensitivities. The methodology is demonstrated on precipitation forecasts from a mesoscale numerical weather prediction model.

show abstract

Forecast Verification and Visualization based on Gaussian Mixture Model Co‐estimation

Cited by 9 publications

References 19 publications

The State of the Art in Visual Analysis Approaches for Ocean and Atmospheric Datasets

The State of the Art in Visual Analysis Approaches for Ocean and Atmospheric Datasets

Drought Occurring With Hot Extremes: Changes Under Future Climate Change on Loess Plateau, China

On the effect of model parameters on forecast objects

Contact Info

Product

Resources

About