Machine-learning regression applied to diagnose horizontal visibility

What is it about?

Weather forecasting has evolved significantly from the late 1800s and the progress continues. Nevertheless, the numerical weather prediction (NWP) has shown some shortcoming to make cost-effective use of the increase in volume, diversity, and capabilities of observations and environmental products. Indeed, machine learning (ML) techniques are emerging as suitable tools that will have to supplement major components of operational systems. Low-visibility conditions are a weather hazard that affects all forms of transport, and accurate forecasting of their spatial coverage is still a challenge for meteorologists, particularly over a large domain. Current predictions of visibility are based on physical parametrizations in numerical weather prediction models and are thus limited with respect to accuracy. This paper examines the use of supervised machine-learning regression techniques (tree-based ensemble, feed-forward neural network and generalized linear methods) to diagnose visibility from operational mesoscale model forecasts over a large domain.

Featured Image

Photo by Annie Spratt on Unsplash

Photo by Annie Spratt on Unsplash

Why is it important?

Our findings show that Extreme gradient boosting (XGBoost) is a suitable tool for estimating visibility over a large domain while generalized linear model has the worst performance due to the complexity of such meteorological conditions and the non-linear interaction between the physical processes. Besides, Machine Learning algorithms perform better when the forecast depends on multiple predictors instead of only a few with very high importance. Regarding the robustness of the algorithms to the features correlation, it is found that their performance decreases when principal components are used instead of raw correlated data. In addition, their performance is very sensitive to disproportionality between daytime and night-time datasets.

Perspectives