A synthetic model to downscale and forecast evapotranspiration using wavelets and SVMs
Location
ECC 203
Event Website
http://water.usu.edu/
Start Date
4-5-2007 3:10 PM
End Date
4-5-2007 3:30 PM
Description
Provision of reliable forecasts of evapotranspiration (ET) at the farm level can be a key element in efficient water management in irrigated basins. This paper presents an algorithm that provides a means to downscale and forecast ET images. The key concepts driving the development of this algorithm are building multiple relationships between inputs and outputs at all different spatial scales, and using these relationships to downscale and forecast the output at the finest scale. This downscaling/forecasting algorithm is designed for dependent properties such as ET. Decomposing and reconstructing processes are done using two-dimensional (2D) discrete wavelet decomposition (2D- DWT) with basis functions that suit the physics of the property in question. 2D- DWT, for one level, results in one datum image (Low-Low pass filter image, or LL) and three detailing images (Low-High or LH, High-Low or HL, and HighHigh or HH). The underlying physics between the input variables and the output are learned by using Support Vector Machines (SVMs) at the resolution of the output. The machines are then applied at a higher resolution to produce detailing images to help downscale the output image (e.g., ET). In addition to being downscaled, the output image can be shifted ahead in time, providing a means for the algorithm to be used for forecasting. The algorithm has been applied on two case studies, one in Bondville, Illinois where the results have been validated against Ameriflux observations, and another in the Sevier River Basin, Utah.
A synthetic model to downscale and forecast evapotranspiration using wavelets and SVMs
ECC 203
Provision of reliable forecasts of evapotranspiration (ET) at the farm level can be a key element in efficient water management in irrigated basins. This paper presents an algorithm that provides a means to downscale and forecast ET images. The key concepts driving the development of this algorithm are building multiple relationships between inputs and outputs at all different spatial scales, and using these relationships to downscale and forecast the output at the finest scale. This downscaling/forecasting algorithm is designed for dependent properties such as ET. Decomposing and reconstructing processes are done using two-dimensional (2D) discrete wavelet decomposition (2D- DWT) with basis functions that suit the physics of the property in question. 2D- DWT, for one level, results in one datum image (Low-Low pass filter image, or LL) and three detailing images (Low-High or LH, High-Low or HL, and HighHigh or HH). The underlying physics between the input variables and the output are learned by using Support Vector Machines (SVMs) at the resolution of the output. The machines are then applied at a higher resolution to produce detailing images to help downscale the output image (e.g., ET). In addition to being downscaled, the output image can be shifted ahead in time, providing a means for the algorithm to be used for forecasting. The algorithm has been applied on two case studies, one in Bondville, Illinois where the results have been validated against Ameriflux observations, and another in the Sevier River Basin, Utah.
https://digitalcommons.usu.edu/runoff/2007/AllAbstracts/30