Ground sampling methods for surface soil moisture in heterogeneous pixels

Reasonable ground sampling methods for the development of soil moisture remote sensing retrieval algorithm models and product validation are very important. Soil moisture has obvious variability at different spatial and temporal scales, especially for surface soil moisture. It makes the surface soil moisture sampling at medium or coarse resolution even more challenging. The study reviewed the current common sampling methods and further described the calculation process and limitation in application. Field experiments were carried out in the sampling domain of 165 x 165 m(2) to compare different sampling methods. The results indicated that the corresponding values of necessary sampling size under a standard error of 0.0076 cm(3) cm(-3) (equivalent to relative error = 5 % and confidence level = 95 %) were about 55, 52, 37, 35 and 26 for the conventional statistical sampling, bootstrap sampling, random combination sampling, geo-statistical sampling and stratified sampling, respectively. However, they all required data collection of soil moisture observation in advance to generating an optimal sampling strategy. Even the temporal stability sampling method also needed more numbers of long-term observational data to obtain less representative stable points. When the area of sampling pixels increases in practice, especially when the sub-pixel variation is larger, it is particularly difficult to conduct pre-sampling or real-time sampling at a medium or coarse spatial resolution because of the limited manpower. Therefore, the soil moisture stratified sampling based on vegetation temperature condition index prior knowledge was put forward and applied in the regional scale. With the rapid development of global change research, many kinds of data can become the prior knowledge, including land use data, soil texture data, DEM meteorological data, etc.