Abstract: Abstract: Soil moisture has been a key limiting factor for crop growth during the surface process in many lands. It is very necessary to establish the placement criteria of monitoring sites for the soil moisture in optimum irrigation. The spatial and temporal distribution of Available Water Capacity (AWC) was strongly correlated with soil moisture. The AWC spatial distribution pattern was also related to soil characteristics, but it can be more stable than that of soil moisture. In this study, a spatially stratified sampling was proposed to relieve the strong temporal and spatial variability, when the soil moisture was used as a variable. The Stratified Sampling method based on spatial autocorrelation of Auxiliary Variables (SSAV) was also used to combine the classical statistics and geo-statistics, where the AWC was introduced as an auxiliary variable. The experiments were then carried out to verify at a field scale. The results showed that the AWC in the 0-40 and 0-80 cm soil layers followed the normal distribution, indicating a moderate variation. In the 90% confidence interval, the number of monitoring points in the 0-40 and 0-80 cm soil layers in the study area was 7 and 6, respectively, where the sampling error was 10%, indicating that the reducing number of monitoring points, and cost-saving monitoring of soil moisture. The geostatistical analysis demonstrated that the range of two soil layers (0-40 and 0-80 cm) was both 366 m in the semi-variance function of AWC. The relative errors of soil moisture estimated by the Simple Random Sampling (SRS) and SSAV were –27.03%-52.38%, and –23.23%-35.15%, respectively. The relative error of soil moisture estimated by the SSAV was reduced by 26.48%, compared with the SRS. The mean standard deviation was 4.78%, 17.30% lower than that of SRS. A paired t-test indicated that the relative error and the mean standard deviation of soil moisture were 9 times in the two soil layers under two monitoring during 2016-2018. Thus, there were significant differences between the relative error range and the mean standard deviation under two monitoring (P<0.001). Moreover, the uncertainty of SSAV was reduced significantly, whereas, the estimation accuracy was improved significantly, compared with the SRS. Among them, the uncertainty of SRS was attributed to the independent samples that followed the normal distribution. There was also a certain spatial change of soil characteristics at a certain scale, indicating a spatial correlation. Correspondingly, the larger deviation of estimation accuracy was attributed to the SRS model without considering the spatial autocorrelation of soil moisture. The Root Mean Square Error (RMSE) between the observed and estimated values was only 0.010 4 cm3/cm3, indicating significantly lower than that of the SRS (0.012 0 cm3/cm3). As such, the uncertainty of sampling was reduced according to the value range of AWC. The new sampling was fully considered the influence of the spatial structure of the target variable on the layout of monitoring points. Specifically, the distance between any two monitoring points was required to be greater than the range, where the monitoring points were independent of each other. Therefore, the estimation accuracy and precision of soil moisture were improved, compared with the SRS. Consequently, the SSAV can be widely expected to serve a time-, labor- and cost-saving monitoring scheme for the average soil moisture. The finding can provide a promising guideline for water resources management and water use efficiency in modern agriculture.