EcoNet Weather Forecasting

Abstract

As we are all probably familiar with, weather-related predictions have notoriously unreliable, especially when the predictions are made far into the future. Although there are a variety of factors that might impact this lack of reliability issue, one of them is the collection of quality data from sensors. Weather stations and sensors do a relatively good job at collecting data so that it can be reported, however, sometimes these sensors might make erroneous readings which is an issue. Such is the problem that the NC Climate Office is faced with. They have 50+ weather stations spread across the state of North Carolina which relay weather information every minute. Sometimes a reading will be sent that may not look correct, and the NC Climate Office has an automated quality assurance program in place to flag these readings. However, these flagged readings still need to be looked at by a human to make sure that the reading is erroneous and not a weather anomaly. The issue here is that it takes many hours to analyze the data by hand to make sure the readings are legitimate. If there was a way to further automate the QA, it could save hundreds of hours of work while providing a more reliable and consistent way to detect erroneous data. To accomplish this level of automation, there would need to be a robust decision model which would take the weather station data, along with the automated QA flags, and be able to confirm if a reading might be erroneous or not. Such a model would need data to be trained on, which the NC Climate Office has provided.