| Summary: | 碩士 === 國立臺灣大學 === 農業化學系 === 85 === Spatial Distribution Prediction of the Depth Class
of Fe, Mn Concretion and Grey Mottle Using
Indicator Kriging Combined with
Discriminant Analysis Chiang,
Tien-Yin ABSTRACT
Spatial variability of basic soil properties are mainly
dependent on soil pedogenic processes and the environmental
factors in soil survey, especially caused by the different
factors of landscape. In this study, the variables of soil-
landscapes including geography, vegetation, and parent material
obtained from geographic information system (GIS) were
collected. These variables include elevation, kinds of parent
materials, distance to local stream, distance to local
irrigation channel, distance to local irrigation pond, distance
to local road, and distance to living area. Then, the soil-
landscape model were established by multivariate discriminant
analysis (MDA) based on the variables used in soil-landscape
model and basic soil properties, and to predict the spatial
variability of soil properties distributed in another area which
was not sampled. The spatial distribution of soil properties can
be predicted by indicator kriging method of geostatistics
attributed to the spatial dependent of variations of soil
properties in the study area. Two hundred and twelve
soil pedons were sampled from 1774 hectares rural soils located
at Nankang, Taoyuan county in northern Taiwan. The grid sampling
method was used to sample soils from soil surface to 150 cm
depth by auger to identify the depth of redoximorphic features
including iron concretion, manganese concretion and grey mottle
in soil pedons. Indicator kriging method and it was combined
with soil-landscape model were used to predict the spatial
distribution of Fe concretion, Mn concretion and grey mottle
under different sampling strategies including different
densities (50 and 100 pedon numbers) and different methods (grid
and random methods), respectively. The accuracy of
prediction on the depth class of Fe concretion, Mn concretion
and grey mottle by indicator kriging method under the sampling
intervals of distance from 250 to 500 meters (n=100) are much
better than those of the sampling distance of 500 meters (n=50).
The accuracy of prediction by grid sampling method under low
sampling density is better than that of random sampling method,
but the accuracy of prediction by random sampling method under
high sampling density is better than that of grid sampling
method. On the boundary area of different depth classes of
concretions or mottles, better accuracy of prediction can be got
by grid method. There are no significant differences on the
prediction of depth classes of grey mottles in the soil pedon
between grid and random sampling methods. Soil-landscape
model combined with indicator kriging method can increase the
accuracy of prediction on the depth classes of Fe-Mn concretions
and grey mottles in the study area compared that predicted only
by indicator kriging method. This model combined with indicator
kriging method can effectively identify the boundary area of
different depth classes of concretion and mottles and also
increase the accuracy of prediction on the spatial variability
of soil properties. This combined technique can be effectively
predict the spatial distribution of basic soil properties caused
by pedogenic factors, especially on the application of the
geostatistics in soil science. Key Words: Indicator Kriging,
Multivariate discriminant analysis, Soil-landscape
model, Sampling strategy, Pedogenic processes, Fe concretion, Mn
concretion, Grey mottle, Depth classes, GIS, Geostatistics.
|
|---|
