TY - JOUR

T1 - Fuzzy mapping of momentum fluxes in complex shear flows with limited data.

AU - Hankin, B. G.

AU - Kettle, H.

AU - Beven, Keith J.

PY - 2001

Y1 - 2001

N2 - This study is divided into three parts centred around modelling the complex turbulent fluxes across strong shear layers, such as exist between the channel and floodplain flow in an over bank flood flow. The three stages utilize Adaptive Neuro-Fuzzy Inference Systems (ANFIS) to make fuzzy mappings between the fluxes and different data types. The de-fuzzification stage commonly used in Fuzzy Inference Systems is adapted to avoid the generation of crisp outputs, a process which tends to hide the underlying uncertainty implicit in the fuzzy relationship.Each stage of the study utilizes conditioning data that makes the fuzzy mappings more tenuously linked with what would normally be considered physically based relationships. The need to make such mappings in distributed models of complex systems, such as flood models, stems from the sparsity of available distributed information (e.g. roughness) with which to condition the models. If patterns in distributed observables which clearly affect, or are affected by, the river hydraulics can be linked to the local fluxes, then the conditioning of the model would improve. Mappings such as these often suffer from scaling effects, an issue addressed here through training the fuzzy rules on the basis of both laboratory and field collected data.

AB - This study is divided into three parts centred around modelling the complex turbulent fluxes across strong shear layers, such as exist between the channel and floodplain flow in an over bank flood flow. The three stages utilize Adaptive Neuro-Fuzzy Inference Systems (ANFIS) to make fuzzy mappings between the fluxes and different data types. The de-fuzzification stage commonly used in Fuzzy Inference Systems is adapted to avoid the generation of crisp outputs, a process which tends to hide the underlying uncertainty implicit in the fuzzy relationship.Each stage of the study utilizes conditioning data that makes the fuzzy mappings more tenuously linked with what would normally be considered physically based relationships. The need to make such mappings in distributed models of complex systems, such as flood models, stems from the sparsity of available distributed information (e.g. roughness) with which to condition the models. If patterns in distributed observables which clearly affect, or are affected by, the river hydraulics can be linked to the local fluxes, then the conditioning of the model would improve. Mappings such as these often suffer from scaling effects, an issue addressed here through training the fuzzy rules on the basis of both laboratory and field collected data.

M3 - Journal article

VL - 3

SP - 91

EP - 103

JO - Journal of Hydroinformatics

JF - Journal of Hydroinformatics

SN - 1464-7141

IS - 2

ER -