@MASTERSTHESIS\{IMM2011-06072,
    author       = "J. S. Vestergaard",
    title        = "Improved nowcasting of heavy precipitation using satellite and weather radar data",
    year         = "2011",
    school       = "Technical University of Denmark, {DTU} Informatics, {E-}mail: reception@imm.dtu.dk",
    address      = "Asmussens Alle, Building 305, {DK-}2800 Kgs. Lyngby, Denmark",
    type         = "",
    note         = "Supervised by Associate Professor Allan Aa. Nielsen, aa@space.dtu.dk, and Professor Rasmus Larsen, rl@imm.dtu.dk, {DTU} Informatics",
    url          = "http://www.imm.dtu.dk/documents/ftp/ep2011/ep11_38-rev.pdf",
    abstract     = "Global climate changes in recent years have caused a higher frequency of heavy precipitative events in Denmark, due to the increase in the atmospheric temperature. Therefore, a desire to nowcast these events has emerged. Nowcasting is the discipline of short term forecasting (0–3 hours) meteorological events and works on a smaller scale than the numerical weather models typically used for forecasting.
Six dates over the last few years (2007–2010) exhibiting extreme weather phenomena in Denmark have been selected, ranging from heavy snow fall to extreme downpour. Data used in the nowcasting come from the Meteosat-8 satellite and from weather radars operated by the Danish Meteorological Institute (DMI).
The supplied data are used for development of a nowcast system specifically designed for heavy precipitative events in Denmark. This includes a statistical approach to identification of ground truth using linear multivariate statistical methods, such as canonical correlation analysis. A method for learning a discriminative dictionary of satellite image patches is applied for classification and prediction of heavy precipitation.
An operational setting is simulated by use of leave-one-out cross validation, where the nowcast model is built on data from five dates and evaluated on the sixth. While the nowcasting abilities degrade when increasing the nowcast length above 0.5 hours, probably due to the diversity of the six weather situations, the method proves successful in classifying heavy precipitative events as they occur."
}