基于模拟退火算法的土壤样点布局和土壤景观模型优化  

Optimization of Soil Sampling Schemes and Soil Landscape Model based on Simulate Annealing Algorithm

作　　者：钟克强 郭澎涛[2] 武伟[3] 刘洪斌[1] 

机构地区：[1]西南大学资源环境学院,重庆400715 [2]中国热带农业科学院橡胶研究所,海南儋州571737 [3]西南大学计算机与信息科学学院,重庆400715

出　　处：《土壤通报》2017年第1期14-21,共8页Chinese Journal of Soil Science

基　　金：国家科技支撑计划课题(2008BADA4B10);中央高校基本科研业务费专项(XDJK2016D041)资助

摘　　要：本研究利用多重线性回归方程,以地形因子为预测变量,构建关于土壤有机质的土壤景观模型,并以西南山地丘陵区的一块面积为2 km2的汇水盆地为研究区,对该区域的土壤有机质空间分布进行预测。在此基础之上,探讨最少可用多少个点来预测土壤有机质的空间分布,并使之预测精度不低于原始集合的精度;同时,找出最优土壤样点布局,确定不同地形部位的取样单元,使之预测精度最高。研究结果表明:在预测误差最小化的情况下,最少可用7个优化的样点就可以代替原始200个采样点,且优化的样点数为124时,模型预测土壤有机质空间分布的精度最高。优化后的土壤景观模型的拟合度比原始模型提高了3.28%,MAE降低了5.3%,RMSE降低了3.94%。A soil landscape model was developed to predict the spatial distribution of soil organic matter using multiple linear regression and terrain factors in this paper. The study was conducted in a catchment located in hilly areas of southwest China. The area of the catchment was about 2 km^2. In order to obtain the minimal number of samples and an appropriate soil sampling density, the developed model was optimized using simulate annealing algorithm considering topographic position without decreasing or even increasing the accuracy of soil organic matter. According to the minimization of mean squared error, the original soil samples （n = 200） were replaced with the optimized soil samples （n = 7） and the model performed best with 124 samples. Compared with the original model, the adjusted coefficient of determination in the optimized model was improved by 3.28% and the mean absolute error and root mean squared error were reduced by 5.3% and 3.94%. respectively.

关 键 词：土壤景观模型 模拟退火 采样点布局 优化 

分 类 号：S151.9[农业科学—土壤学]