TY - THES
AU - Cahoon, Peter G.
PY - 1978
TI - Application of auto regressive filtering techniques to flood flow prediction
KW - Thesis/Dissertation
LA - eng
M3 - Text
AB - This thesis describes a study of the application of multichannel
deconvolution, an autoregressive filtering technique, to the problems of flood flow prediction. This application is divided into four component segments:
1. The description of the behaviour of a multiple input, multiple output basin using multichannel autoregressive techniques with multiple lags of finite length. These descriptions fall into two catagories:
a) the Weiner autoregressive technique₇.
b) minimum entropy auto regression₁₃.
2. The restatement of the input/output problem as a time varying state-space description with a feedback mechanism for implementation of information having a unit time delay.
3. The analysis and characterization of the state-space and autoregressive methods using standard spectral analysis techniques and statistical confidence limits.
4. The linking of the state-space/autoregression characterizations for snowpack depletion and rainfall/runoff into a finite state machine algorithm coalescing the two processes so they may be linked to satellite information.
Several case studies were used in which the multiple precipitation
records and multiple flow records were characterized. These steady state characteristics were then updated using the Kalman state-space description to provide an "online"₁, information update. The snowpack depletion problem was treated as a multiple input (rainfall, temperature, humidity), single output (snowmelt) phenomena and characterized by a single multichannel autoregression.
A second class of characterizations was employed to cope with the "spike" arrivals caused by rapid snowmelt flowing into a single output. This minimum entropy₁₃ technique is developed for "flash flood" prediction.
Finally, a finite state machine algorithm is developed to link the snowpack depletion to the rainfall runoff problem in such a way that it can be readily linked to satellite produced data streams.
For those unfamiliar with the complex and obtuse language of deconvolution and control theory, the role of this thesis can be described by analogy:
You have arrived in the middle of a party where the participants are all quite intoxicated. They are huddled together in small knots talking excitedly, but without much clarity about decon operators, M.E.D. processes, frames and finite state machines. My role is to catch you up on the conversation and help you find your way to the bar.
N2 - This thesis describes a study of the application of multichannel
deconvolution, an autoregressive filtering technique, to the problems of flood flow prediction. This application is divided into four component segments:
1. The description of the behaviour of a multiple input, multiple output basin using multichannel autoregressive techniques with multiple lags of finite length. These descriptions fall into two catagories:
a) the Weiner autoregressive technique₇.
b) minimum entropy auto regression₁₃.
2. The restatement of the input/output problem as a time varying state-space description with a feedback mechanism for implementation of information having a unit time delay.
3. The analysis and characterization of the state-space and autoregressive methods using standard spectral analysis techniques and statistical confidence limits.
4. The linking of the state-space/autoregression characterizations for snowpack depletion and rainfall/runoff into a finite state machine algorithm coalescing the two processes so they may be linked to satellite information.
Several case studies were used in which the multiple precipitation
records and multiple flow records were characterized. These steady state characteristics were then updated using the Kalman state-space description to provide an "online"₁, information update. The snowpack depletion problem was treated as a multiple input (rainfall, temperature, humidity), single output (snowmelt) phenomena and characterized by a single multichannel autoregression.
A second class of characterizations was employed to cope with the "spike" arrivals caused by rapid snowmelt flowing into a single output. This minimum entropy₁₃ technique is developed for "flash flood" prediction.
Finally, a finite state machine algorithm is developed to link the snowpack depletion to the rainfall runoff problem in such a way that it can be readily linked to satellite produced data streams.
For those unfamiliar with the complex and obtuse language of deconvolution and control theory, the role of this thesis can be described by analogy:
You have arrived in the middle of a party where the participants are all quite intoxicated. They are huddled together in small knots talking excitedly, but without much clarity about decon operators, M.E.D. processes, frames and finite state machines. My role is to catch you up on the conversation and help you find your way to the bar.
UR - https://open.library.ubc.ca/collections/831/items/1.0063016
ER - End of Reference