Sampling and Optimizing Methods of Cultivated Land Quality Based on Spatial Simulated Annealing Algorithm
Yang Jianyu,Yue Yanli,Song Hairong,Tang Sai,Ye Sijing,Xu Fan
DOI: 10.11975/j.issn.1002-6819.2015.20.035http://www.tcsae.org
2015-01-01
Abstract:M Monitoring points in country area are the foundation to reflect changes of cultivated land quality, which directly affect the result of farmland grading and its accuracy. Through the monitoring network for cultivated land quality in county area, the distribution and changing trend of the cultivated land quality can be reflected. Besides, the quality of non-sampled locations should also be estimated with the data of sampling points. Due to the correlation among spatial samples, the traditional methods such as simple random sampling, stratified sampling and systematic sampling are inefficient to accomplish the task above. Thus, we propose a new spatial sampling and optimizing method based on the spatial simulated annealing (SSA). This paper presents a pre-processing method to determine the number of sampling points, including preprocessing the data of cultivated land quality before sampling, exploring the spatial correlation and spatial distribution pattern of cultivated land quality, and computing the appropriate quantity of sampling points by analyzing the change trend of sampling number and sampling precision, and on this basis we propose the extended spatial simulated annealing method to optimize spatial sampling design for obtaining the minimal Kriging variance. The main steps for computing the optimal sampling design can now be summarized as follows: 1) calculate the semi-variogram of cultivated land quality and determine the parameters of ordinary Kriging interpolation; 2) identify the quantity of samples, choose a set of cultivated land map spots randomly as an initial design, and compute the associated fitness function; 3) given one design,construct a candidate new sampling design by random perturbation; 4) compute the fitness function for the new design, and if it is smaller than or equal to that for the original design, accept the original design, or else accept the new design with an acceptance probability. If the new design is accepted, the estimated point (j) is returned to zero, or else increased by 1; 5) ifj is smaller than or equal to a threshold value of continuous rejections, increasei (representing monitoring point) by 1, or else stop the iteration and current design is the best. Designs by simulated annealing that reduce the average Kriging standard error are always accepted, and designs that worsen the interpolation effect are accepted with a certain probability, which decreases to zero as iterations proceed. However, there are integrated factors such as soil organic matter content, topsoil texture, profile pattern, salinization which affect arable land quality change over time and space, and are taken as potential change factors to detect potential change areas. Under the guidance of expert knowledge, the sampling points are set up through spatial simulated annealing algorithm and adjusted based on potential change areas, rivers, roads and abnormal monitoring points. We illustrate this new method using Daxing District, Beijing City as a case study. Spatial overlay analysis of potential change factors and geostatistics method of GIS are employed to test this method. The spatial variability of cultivated land quality is simulated using natural quality indices and a specified number of network locations is defined which can be used to adequately predict the quality of cultivated land. The experimental results of Daxing District, Beijing City show that 55 monitoring reference sample units are finally deployed, and the average ordinary Kriging standard error with this method is 131.78, which is smaller than the simple random sampling (134.97) and stratified sampling (134.93) when the quantity of samples is the same. Besides, sampling accuracy and cost are both considered and reach a certain balance in this method. This method is suited for counties which have carried out several surveys of cultivated land quality, or counties whose grading factors have certain changes. Besides, it is also suitable for counties which have some prior knowledge but never have conducted a survey of cultivated land quality.