Noise Reduction Methods for Weighing Lysimeters

Mechanical vibration of the grass and crop weighing lysimeters, located at the University of California West Side Field Research and Extension Station at Five Points, Calif. generated noise in lysimeter mass measurements and reduced the quality of evapotranspiration (ET) data. The estimated power spectral density (PSD) for grass lysimeter mass data acquired at $1.3\mathrm{ms}$ intervals contained a large peak at $11\mathrm{Hz}$ . Crop lysimeter data produced similar peaks at frequencies greater than $1\mathrm{Hz}$ . An effective method for eliminating this noise source is arithmetic averaging of the data, which should be acquired sufficiently rapidly to avoid aliasing. The PSD also increased with decreasing frequency in the range $1.0–0.1\mathrm{Hz}$ . This noise was addressed by Savitsky–Golay (SG) filtering using 7-, 11-, and 15-point filters. Each filter was applied to the same data set consisting of 2,560 measurements taken during a $1\text{-}\min$ interval every $10\min$ over a $26.3\text{-}\mathrm{h}$ period. Noise reduction factors, defined as the ratio of standard deviation of filtered lysimeter mass to standard deviation of unfiltered mean values of lysimeter mass for subsequences of the same data, were 0.90, 0.88, and 0.86 for the 7-, 11-, and 15-point filters, respectively. For the daytime data only, the factors were 0.88, 0.85, and 0.83. The SG filters were more effective during daytime when most of the lysimeter ET occurs. These methods are simple enough to be programmed into commercially available dataloggers for real time filtering. Hourly averages of the standard deviations of lysimeter mass measurements bear a distinct nonlinear relationship to hourly mean wind speed confirming earlier suppositions that wind loading causes noise in counterbalanced weighing lysimeters.