Slide 1Slide 2Slide 3Slide 4Slide 5Slide 6Slide 7A Neural Network Approach to Estimate Snowfall Parameters from Passive Microwave RadiometerWei HuangClass project presentation for ECE539OverviewKnowing snowfall rate over the ground is an important part of hydrological circle. Snowfall is closely related to our daily life.Snowfall can emit thermal radiation in the microwave band, which can be observed by radiometer(microwave is parallel to light in the visible band ). There are many factors that determine the accuracy of snowfall retrieval. The microwave response to snowfall is nonlinear in general.Passive Microwave Response to Snowfall Snow density and shape, which is subject to temperature, wind speed,etcSurface type: sea surface(simple), land surface type(complex)Cloud liquid water can also affect the observation.Satellite based PM radiometer has a bigger observation range. Active radar observation can provide a more accurate estimation, but to a limited rangeMethodSimulated database, which includes simulated brightness temperature at 89.0, 36.5GHz (V/H)and snowfall rate over the ground (mm/hr)Satellite database ,which includes the brightness temperature at 89.0 and 36.5 GHzNeural network(feed-forward back-prop network 4-4-1)Comparison with available Radar data, which includes the radar derived snowfall rateResults(Feature space)Training set feature space:89.0,36.6GHz, vertically and horizontally polarized.Output: Snowfall rate.ANN retrievalTop row: Left: Model output snowfall rate of 2003.12.19 Wakasabay. Right: Trained ANN output snowfall rate.(correlation: 0.98)Bottom row:Left: ANN output snowfall rate of 2005.12.25. WakasakbyRight: Radar retrieved snowfall rate of 2005.12.25. Wakasabay. (correlation:0.95)ConclusionANN works pretty well, when training data set can cover a wide range of
View Full Document