Sampling number and design for measurements of infiltration rates into puddled rice fields

Abstract

Fifty-six infiltration rate (IR) measurements were made in puddled clay soil in a 0.05 ha experimental rice field consisting of 16 subplots, following a standard stratified sampling scheme. Log-transformed IR measurements were approximately normally distributed. Analysis of stationarity revealed the absence of a trend, whereas estimating the coefficients of the covariance function revealed a pure nugget effect, indicating spatial independence. A sequential t-test was used to determine the number of measurements that is sufficient to characterize the average rate of infiltration in the experimental plot under study. Results showed that 8 measurements yielded mean estimates that were not significantly different with 95% confidence from results obtained using all 56 measurements. Application of the test thus resulted in an 85% saving of measurement effort as compared with the original sampling scheme. The question of the design of the measurement scheme, i.e., which of the 8 measurements to select is addressed by means of selecting all possible combinations of the 8 observations among the 16 subplot centres, and calculating the mean squared error (MSE) and the mean variance of the prediction error (MVP) in a test set consisting of the 40 other measurements. The risk of taking an inappropriate configuration is relatively small as compared to the actual mean value to obtained.