Automated Filtering versus Moving Average in the Analysis of Automated Beerkan Infiltrometer Data

The use of the automated Beerkan infiltrometer (10.1016/j.compag.2015.09.022) represents an interesting alternative to the classical Beerkan infiltration test. The device enables the measurement of field infiltration rates at a higher temporal resolution than manually conducted tests. Although the infiltrometer can be easily implemented in the field and provides good measurement reproducibility, the analysis of raw experimental data requires appropriate processing to ensure reliable estimation of soil hydraulic parameters using the BEST algorithms.

The automated infiltrometer operates as follows. Water infiltrates at the soil surface, and when the water depth decreases below a given threshold, the Mariotte bottle of the infiltrometer is activated and allows an air bubble to enter the reservoir. Consequently, the increase in air pressure in the reservoir releases an amount of water. Under ideal conditions, monitoring of the air pressure in the system produces a piecewise-constant step function. Each plateau corresponds to a constant water height in the reservoir and is separated from the next by oscillations caused by the passage of the air bubble. Therefore, the signal must be filtered using appropriate filtering techniques in order to identify the points that properly define the cumulative infiltration curve. In the ideal case, the selected points should correspond to the end of each plateau, which defines the exact time at which the infiltrometer supplies water. However, in many cases the signal is noisy, making filtering a challenging task.

This study compares the performance of two mathematical approaches commonly used to process raw pressure transducer data: (i) an automatic filtering method based on first and second derivatives to detect plateaus (i.e., periods of constant water height in the reservoir), and (ii) the commonly used moving average technique. Experimental data collected on a well-structured, cultivated sandy soil in Senegal were used to assess the impact of the two filtering approaches on the determination of cumulative infiltration and on the estimation of saturated soil hydraulic conductivity (Ks) and soil sorptivity (S), using the three BEST algorithms (BEST-slope, BEST-intercept, and BEST-steady).

We expect that comparison of the cumulative infiltration curves obtained with the two methods will reveal discrepancies, and that the automated filtering approach will better preserve infiltration dynamics, as suggested by preliminary results. In contrast, the moving average method may excessively smooth the data, potentially leading to biased estimates of hydraulic parameters, particularly under conditions of strong capillary effects. A synthesis of the results from both methods will help identify the most appropriate filtering approach.