A visual voting framework for weather forecast calibration

Abstract

Numerical weather predictions have been widely used for weather forecasting. Many large meteorological centers are producing highly accurate ensemble forecasts routinely to provide effective weather forecast services. However, biases frequently exist in forecast products because of various reasons, such as the imperfection of the weather forecast models. Failure to identify and neutralize the biases would result in unreliable forecast products that might mislead analysts; consequently, unreliable weather predictions are produced. The analog method has been commonly used to overcome the biases. Nevertheless, this method has some serious limitations including the difficulties in finding effective similar past forecasts, the large search space for proper parameters and the lack of support for interactive, real-time analysis. In this study, we develop a visual analytics system based on a novel voting framework to circumvent the problems. The framework adopts the idea of majority voting to combine judiciously the different variants of analog methods towards effective retrieval of the proper analogs for calibration. The system seamlessly integrates the analog methods into an interactive visualization pipeline with a set of coordinated views that characterizes the different methods. Instant visual hints are provided in the views to guide users in finding and refining analogs. We have worked closely with the domain experts in the meteorological research to develop the system. The effectiveness of the system is demonstrated using two case studies. An informal evaluation with the experts proves the usability and usefulness of the system.

A survey of urban visual analytics: Advances and future directions

“…Among many meteorological issues, weather forecasts are the most familiar. Weather prediction calibration [93] and comparison of prediction outcomes [92] have been studied using visual analytics.…”

“…Representation learning embeds data into a highdimensional space by vectorization [161]. In this Traffic modelling [53,60,64,80] Data cleaning [13,20] Object tracking [114] Heuristic search [65,69,70,93,98] Simulation-based [16,32,87,113,120,124] Mathematical programming [105,116] Indexing [15,17,22 space, adversarial examples can also be generated for adversarial learning [75,76]. The vectors capture the inherent relationships between data.…”

“…They used a heuristic method, Monte Carlo tree search to solve it. Heuristic strategies have also been used for ODflow sampling [40], and majority voting for weather forecast calibration [93].…”

“…Representative examples provide model diagnoses [75][76][77][78][79] and steering [25,80,88,93,96]. For example, autonomous driving models and systems include artificial-intelligence-based decision-making models that run in the urban space.…”

Research on Informatization Construction of the New Generation qwisdom water networkq Based on qWater Profileq

“…The satellite and radar regularly return meteorological information including more than 30 elements such as rainfall, environmental humidity and soil water potential [7]. These data are growing at an annual PB level, so we need higher storage and computational power to process the data [8].…”