Integration of EMI sensor data in soil sampling scheme optimization using continuous simulated annealing

Soil survey is generally time-consuming, labour-intensive and costly. Optimization of sampling scheme allows one to reduce the number of sampling points without decreasing or even increasing the accuracy of investigated attribute. Maps of bulk soil electrical conductivity (ECa) recorded with EMI sensors could be effectively used to direct soil sampling design for characterizing spatial variability of soil moisture. A protocol, using a field-scale bulk ECa survey, has been applied to an agricultural field in Apulia region (south-eastern Italy). Continuous spatial simulated annealing was used as a method to optimize spatial soil sampling scheme taking into account sampling constraints, field boundaries and preliminary observations. Three optimization criteria were used: the first criterion (MMSD) optimizes the spreading of the point observations over the entire field by minimizing the expectation of the distance between an arbitrarily chosen point and its nearest observation, the second criterion (MWMSD) is a weighted version of the MMSD, which uses the digital gradient of the grid ECa data as weighting function, and the third criterion (MMKV) minimizes the mean kriging estimation variance of the target variable. The last criterion utilizes the variogram model of soil moisture estimated in a previous sampling. The three modes and a combination of them were separately tested and compared. Simulated annealing was implemented by the software (written by one of the authors) able to define or redesign any sampling scheme by increasing or decreasing the original sampling locations. The output consists of the computed sampling scheme, many graphical representations summarizing all the optimization phases, such as the convergence graph, the cooling law, the variogram model fitting etc., which can be an invaluable support to the process of sampling design. The proposed approach has shown great flexibility in adapting to the large heterogeneity of the field and searching the optimal solution in a reasonable calculation time. The use of bulk ECa as an exhaustive variable, known at any node of an interpolation grid, has allowed the optimization of the sampling scheme, distinguishing among areas with different priority levels. However, a highly erratic spatial variation of ECa may prevent the application of MWMSD criterion. Further optimization criteria should be added to the procedure in the future, as minimization of cokriging variance, in the case of multi-purpose sampling, or maximation of an economic or social objective function.