农业工程学报
農業工程學報
농업공정학보
2010年
1期
87-93,前插4
,共8页
李启权%王昌全%岳天祥%李冰%杨娟
李啟權%王昌全%嶽天祥%李冰%楊娟
리계권%왕창전%악천상%리빙%양연
径向基函数网络%误差分析%土壤%有机质%空间异质性%普通克里格
徑嚮基函數網絡%誤差分析%土壤%有機質%空間異質性%普通剋裏格
경향기함수망락%오차분석%토양%유궤질%공간이질성%보통극리격
radial basis function networks%error analysis%soils%organic matter%spatial heterogeneity%ordinary kriging
通过研究土壤性质的空间变异和空间插值方法,快速准确获取土壤性质的空间分布是精确农业和环境保护的基础.该文以四川眉山一块约40km~2的区域为试验区,采集表层土壤(0~20cm)样点80个,利用径向基函数(RBF)神经网络建立空间坐标和邻近样点与土壤有机质间的非线性映射关系(RBF2),模拟土壤有机质的窄间分布.与普通克里法(OK)和仅以坐标为网络输入的神经网络方法(RBF1)相比,RBF2的插值精度有显著的提高;相同样点密度下其相对预测误差分别较OK和RBF1减小了9.87%、1.97%(样本A)和13.09%、2.36%(样本B);即使样点数减半的情况下RBF2的相对预测误差也分别较OK和RBF1减小了10.23%和2.33%,并且插值图差异相对较小,可以更好地反映土壤有机质空间分布的异质性.因此,利用以坐标和邻近样点为输入的神经网络方法可以相对准确、快速地获取区域土壤性质空间分布的异质性信息.
通過研究土壤性質的空間變異和空間插值方法,快速準確穫取土壤性質的空間分佈是精確農業和環境保護的基礎.該文以四川眉山一塊約40km~2的區域為試驗區,採集錶層土壤(0~20cm)樣點80箇,利用徑嚮基函數(RBF)神經網絡建立空間坐標和鄰近樣點與土壤有機質間的非線性映射關繫(RBF2),模擬土壤有機質的窄間分佈.與普通剋裏法(OK)和僅以坐標為網絡輸入的神經網絡方法(RBF1)相比,RBF2的插值精度有顯著的提高;相同樣點密度下其相對預測誤差分彆較OK和RBF1減小瞭9.87%、1.97%(樣本A)和13.09%、2.36%(樣本B);即使樣點數減半的情況下RBF2的相對預測誤差也分彆較OK和RBF1減小瞭10.23%和2.33%,併且插值圖差異相對較小,可以更好地反映土壤有機質空間分佈的異質性.因此,利用以坐標和鄰近樣點為輸入的神經網絡方法可以相對準確、快速地穫取區域土壤性質空間分佈的異質性信息.
통과연구토양성질적공간변이화공간삽치방법,쾌속준학획취토양성질적공간분포시정학농업화배경보호적기출.해문이사천미산일괴약40km~2적구역위시험구,채집표층토양(0~20cm)양점80개,이용경향기함수(RBF)신경망락건립공간좌표화린근양점여토양유궤질간적비선성영사관계(RBF2),모의토양유궤질적착간분포.여보통극리법(OK)화부이좌표위망락수입적신경망락방법(RBF1)상비,RBF2적삽치정도유현저적제고;상동양점밀도하기상대예측오차분별교OK화RBF1감소료9.87%、1.97%(양본A)화13.09%、2.36%(양본B);즉사양점수감반적정황하RBF2적상대예측오차야분별교OK화RBF1감소료10.23%화2.33%,병차삽치도차이상대교소,가이경호지반영토양유궤질공간분포적이질성.인차,이용이좌표화린근양점위수입적신경망락방법가이상대준학、쾌속지획취구역토양성질공간분포적이질성신식.
Fast and accurate simulation of the spatial distribution of soil properties from the study on soil spatial variability and spatial interpolation was the basis for precision agriculture and environmental protection. In this paper, 80 topsoil samples were collected in a 40 km~2 test area in Meishan, Sichuan Province. Nonlinear mapped relations between spatial coordinates and neighbor samples and the content of soil organic matter were established based on radial basis function neural network (RBF2) to simulate the distribution of the content of soil organic matter in the test area. Compared with ordinary kriging method (OK) and radial basis function neural network method only using spatial coordinates as inputs of net (RBF1), the predicted errors achieved by RBF2 were much smaller, which were reduced by 9.87%, 13.09% and 1.97%, 2.36%, respectively;even samples were cut in half, the predicted error was still reduced by 10.23% and 2.33%, respectively, compared with OK and RBF1 which used in all samples. Besides, RBF2, which was able to make the interpolation maps and had smaller difference comparatively in different samples, could express the spatial heterogeneity of soil organic matter well. Thus, the spatial heterogeneity information of soil properties could be achieved exactly and quickly by the method of radial basis function neural network which used spatial coordinates and neighbor samples information as inputs of net.