Autocorrelation analysis of high resolution weighing lysimeter time series as a basis for determination of precipitation

Abstract

AbstractThe determination of precipitation (P) is still a challenge, but central for quantifying soil water and element balances. Time series of mass changes (ΔM) from high precision weighing lysimeter may be used to estimate P if deep drainage rates are determined independently. High temporal resolution, however, is accompanied by problems such as correlated data and noise. The objective was to analyze the temporal autocorrelation (AC) in ΔM time series and to identify temporal resolutions for determining uncorrelated P rates. Minute‐based time series of ΔM are analyzed; the data have been recorded at the UMS Science Lysimeters that are located in Dedelow (northeast Germany) as part of the TERENO SoilCan lysimeter network. Periods in 2012 and 2013 were selected in which the wind speed was below 6 ms−1. Data noise‐correction was carried out by using a moving average before the ΔM values cumulated over 60, 30, and 10 min intervals were compared. On a monthly basis, the temporal AC lengths for ΔM were larger in spring (68 min), autumn (62 min), and winter (76 min) than in the summer (23 min). These AC lengths reflected mainly the effect of differences in P‐rates and ‐duration between lower‐intensity rainfall and shorter summer storms. The monthly sums of P based on the 60‐min interval were up to 20% lower than those obtained by using the 10‐min intervals. For P‐values obtained by summing up the ΔM over periods shorter than the autocorrelation length, oscillated fluctuations in ΔM did not cancel out within an interval. The temporal autocorrelation in the highly‐resolved lysimeter data limited the evaluation of ΔM time series. Compared to rain gauge data, the P‐rates obtained from the weighing lysimeters were generally higher. However, this difference decreased when increasing the time interval for cumulating mass changes. Cumulated positive ΔM values based on time intervals larger than the AC length (e.g., 60 min) provided an optimal approximation of the quantity of P, but on the expense of a loss in temporal resolution limited by the AC lengths. Smoothing could reduce noise in the original lysimeter data; however, not the validity problems that are related to the temporal AC.