论文部分内容阅读
收集南阳1:5万土壤类型图、30 m分辨率数字高程模型和TM影像,计算出高程、坡度、坡向、归一化植被指数(NDVI)、穗帽变换的湿度(TCW)参数等,以439个土壤剖面为训练数据,分别按土壤类型连接法(SCLM)、加权最小二乘法(WLS)回归、地理权重(GWR)回归、随机森林(RF)、普通克里格(OK)、回归克里格(RK)进行1m土体土壤有机碳密度(SOCD)制图,其余49个土壤剖面作为验证集。结果表明:(1)对SOCD变异的解释力是影响制图效果的本质因素。土壤类型、土壤表层有机质(OM)是主要预测变量,SCLM、WLS和GWR均只能利用其中一种主要变量,土壤图的详细化和回归模型的复杂化均不能明显改善SOCD制图效果。基于土属和OM变量,RF对SOCD变异的解释力最强,预测效果最优;地统计学空间变异函数对SOCD变异的解释力大于回归模型,小于RF,而与土壤类型相当,其相对制图效果亦如此。(2)预测变量建模和空间相关是两类不同的土壤变异解释机制,RK未必能使它们产生最佳组合:只有WLS回归、GWR回归和缺乏土壤类型信息的RF(OM+TCW)适合RK算法,在原始模型中它们对训练数据的拟合效果依次升高,但其RK结果的优劣排序则相反;所有RK的结果均未达到土属和OM参与下RF制图的精度。
We collected 1: 50000 soil type map, 30m resolution digital elevation model and TM image in Nanyang, and calculated elevation, slope, aspect, normalized NDVI and TCW parameters, A total of 439 soil profiles were selected as training data and analyzed according to SCLM, WLS, GWR, RF, OK, Kriging (RK) plots soil organic carbon density (SOCD) at 1 m soil depth and the remaining 49 soil profiles are validated. The results show that: (1) The explanatory power of variation of SOCD is the essential factor that affects the mapping effect. Soil type and soil surface organic matter (OM) are the main predictors. Only one of SCLM, WLS and GWR can use one of the main variables. Soil mapping and regression model can not significantly improve SOCD mapping. Based on the soil types and OM variables, RF has the strongest explanatory power for SOCD variation and the prediction effect is the best. The explanatory power of spatial variability of geostatistics to SOCD variation is greater than that of regression model, less than RF, but similar to soil type. The effect is also true. (2) Predict variables modeling and spatial correlation are two different mechanisms of interpretation of soil variability, and RK may not give the best combination of them: only RF (OM + TCW) with WLS regression, GWR regression and lack of soil type information is suitable for RK In the original model, their fitting effects on training data increase in turn, but their RK results are in the wrong order. The results of all RKs do not reach the precision of RF mapping under the participation of soil and OM.