PCA （principal component analysis）, CCA （canonical correlation analysis） and PLSR （partial least squares regression） are powerful water quality modeling methods that provide better results than other classical o...PCA （principal component analysis）, CCA （canonical correlation analysis） and PLSR （partial least squares regression） are powerful water quality modeling methods that provide better results than other classical ones such as multiple lineal regression. In this work they were used to model four water quality parameters at the Amadorio Reservoir （Alicante, Spain）, namely： water temperature, dissolved oxygen, pH and conductivity. The main purpose of this study was to predict the future quality of the water and, thus improve its management. Raw data correspond to daily values of mean wind speed, mean wind direction, maximum wind speed, mean, minimum and maximum air temperature, number of hours below 7 ~C, relative humidity, global solar radiation, total precipitation, evapotranspiration, exploitation volume, inflow, outflow, filtration, depth and Julian day. Two years were considered （2004-2005） to get the calibration （186 days, 4,401 registrations） and validation （185 days, 4,573 registrations） datasets. Models were developed using either all the variables or a reduced subset; furthermore, PLSR yielded the best results.展开更多
文摘PCA （principal component analysis）, CCA （canonical correlation analysis） and PLSR （partial least squares regression） are powerful water quality modeling methods that provide better results than other classical ones such as multiple lineal regression. In this work they were used to model four water quality parameters at the Amadorio Reservoir （Alicante, Spain）, namely： water temperature, dissolved oxygen, pH and conductivity. The main purpose of this study was to predict the future quality of the water and, thus improve its management. Raw data correspond to daily values of mean wind speed, mean wind direction, maximum wind speed, mean, minimum and maximum air temperature, number of hours below 7 ~C, relative humidity, global solar radiation, total precipitation, evapotranspiration, exploitation volume, inflow, outflow, filtration, depth and Julian day. Two years were considered （2004-2005） to get the calibration （186 days, 4,401 registrations） and validation （185 days, 4,573 registrations） datasets. Models were developed using either all the variables or a reduced subset; furthermore, PLSR yielded the best results.
