Event Detail

Event Type: 
Applied Mathematics and Computation Seminar
Date/Time: 
Friday, May 1, 2015 - 12:00 to 13:00
Location: 
GLK 113

Speaker Info

Institution: 
OSU EECE and OSU ECBE
Abstract: 

Weather data is getting much cheaper to collect, which is inspiring
projects such as TAHMO (Trans-Africa Hydro-Meteorological
Observatory), which seeks to deploy a network of 20,000 weather
stations across sub-Saharan Africa. However, big data is not
necessarily good data, so all of this weather data is going to require
careful quality control. This talk will describe our plans for
automating this. We propose to combine spatial and temporal
de-trending and time series analysis with non-parametric anomaly
detection and probabilistic modeling to automatically flag anomalous
data values and identify broken sensors. For temperature, solar
radiation, and relative humidity, this is fairly easy. Wind speed and
direction are harder. And precipitation is the ultimate challenge,
because it is so intermittent and bursty. Unfortunately, rain gauges
are also the least reliable weather sensors. This talk is partly a
proposal and partly a plea for help!