Speaker
Description
The Improved Observation Usage in NWP (iOBS) project aims to contribute to improved weather forecast quality from existing and emerging observation systems by combining world-leading NWP with future generation e-infrastructure. The project targets to produce effective assimilation of diverse observations in regional high-resolution NWP models. The observations include, among other sources, massive amounts of data from “Internet of things” (IoT), such as smartphones and personal weather stations (PWS). It is expected that the NWP forecast quality can be improved due to the increased number of observations. For example, it has previously been shown that high-resolution and high-frequency pressure observations from mobile phones can improve the forecast of convective precipitation events. Overall, private observations represent a novel observation type with some unique characteristics and issues.
As part of the iOBS project, improved QC algorithms are developed for pre-processing the private observations. We investigate the suitability of common machine learning techniques, e.g. classification, regression, clustering, and anomaly detection, for these applications. In particular, we assess the feasibility of various machine learning approaches for identifying systematic and random observation errors and for detecting instrument malfunction. It is presumed that due to the different observational characteristics, the required QC for private observations will differ from the operational QC for traditional in-situ observation networks. The applicable machine learning methods will be adopted in the observational data QC pipelines developed in the project.
The iOBS project also addresses issues on how future e-infrastructure for NWP can make use of cloud and high-performance computing (HPC) resources. These resources are needed to tackle the technical challenges coming from both large-scale data ingestion from private observations and from integration of machine learning components to standard IT systems.
