Rule-based machine learning for knowledge discovering in weather data

doi:10.1016/j.future.2020.03.012

2020-07-01

PRE

Lassana Coulibaly

Bernard Kamsu-Foguem *

Fana Tangara

DOI：10.1016/j.future.2020.03.012

原文链接

原文求助

摘要

En 中文

The Climate change trains regularly some phenomena threatening directly the environment and the humanity. In this context, meteorology plays a more important role in the control of these phenomena. It is thus important to search resources allowing to contribute to the improvement of the numerical model for the predictions of weather and climate. The objective of this work is to look for the weaknesses of the models in the simulation of exchanges between the surface and the atmosphere. These exchanges are quantified by sensible and latent heat fluxes. The preprocessing is done through the combined use of k-nearest neighbors algorithm (k-NN) and Autoregressive integrated moving average (ARIMA) model in order to estimate missing values. The processing is performed with the learning of the association rules and the knowledge extracted enables us to make some comparisons between observations and simulations by the numerical model. The postprocessing is made by logical and graphical reasoning that facilitates the visualization of links between the obtained rules. This method is deployed on a database containing measured variables (sensible and latent heat flux, temperature and humidity of the air, wind speed and direction, rain, global radiation, etc.) at the experimental site of the Centre de Recherches Atmospheriques (CRA) which is one of the two sites composing the Pyrenean Plateforme for the Observation of the Atmosphere (P2OA) in France. The obtained and expressed results in the form of association rules have made it possible to highlight that the differences between model and observations from a surface flux point of view are often concomitant with an important difference on global radiation. The expected profits are relative to the generation of knowledge useful for the improvement in the quality of the prediction with a better analysis of the important concomitant factors during errors on a weather model. (C) 2020 Elsevier B.V. All rights reserved.

Keyword:

Data mining

Association rule learning

Prediction

Time series

Meteorology

Land-atmosphere exchanges