Soil maps-based on the catena approach: All catena-based soil maps showed a
very high correspondence with the respective reference soil maps, with validation
of the independent sampling points yielding the following matches: 80 % in Mae Sa
Mai and Bor Krai, and 83 % in Huai Bong. In all cases the major soil types were
classified correctly.
Randomized grid-based soil maps: The randomized grid-based soil mapping for
Bor Krai revealed 97 % correspondence with the reference soil map and a positive
match for 76 % of all independent sampling points. As seen in the transect
approach, minor units (Chernozems, Ferralsols, Fluvisols, Gleysols) located
between the randomized sampling points were not detected (Fig. 2.7).
Maximum likelihood-based soil maps: The level of correspondence with the
reference map was high throughout, showing an 81 % match in Mae Sa Mai and Bor
Krai, and a 71 % match in Huai Bong, while validation with the independent
sampling points revealed a match of 70–71 % at all the sites. Validation of the
maximum likelihood-based soil map in the Mae Sa watershed, but outside the Mae
Sa Mai sub-area, using 30 independent sampling points, revealed a positive match
in 87 % of all the 30 cases. In the pilot study area of Bor Krai, where only 25
sampling points had been distributed among the local soil units (using a minimal
sampling point approach), the classification accuracy was 80 %, while during
validation 71 % of soils were classified correctly. Due to their limited spatial
coverage and poor definition, Chernozems were not predicted for Mae Sa and
Fluvisols not for Huai Bong. The preselection of major variables using principal
component analysis was essential for the maximum likelihood classification and
greatly improved the results. For all three areas, principal component analysis
identified aspect, petrography, slope and the SPOT 5 bands 2 and 4 as being the
most important predictors – explaining most of the observed variance, such as 82 %
in Mae Sa Mai, 78 % in Huai Bong and 92 % in Bor Krai. In addition, in Mae Sa
Mai, SPOT 5 band 3 and in Huai Bong, LANDSAT 7 bands 5 and 7 were essential.
In Bor Krai meanwhile, LANDSAT 7 bands 5 and 7 as well as elevations were also
important.
Classification tree-based soil maps: The level of correspondence between the
classification tree-based soil map (1:50 k scale) and the site specific reference maps
was satisfactory throughout, with a 79 % match for Mae Sa Mai, 70 % for Huai
Bong and 84 % for Bor Krai, while validation of the independent sampling points
showed a matching range of 73–77 % for all the sites. The advantage of the
classification tree approach is the ability to develop rules for the occurrence of
predicted units; for example, in Mae Sa water bodies split off from other mapping
units due to high values in SPOT band 3 and low values in SPOT band 4, whereas
Technosols were identified via a combination of high values in SPOT bands 3 and 4,
and LANDSAT band 1, and low values in LANDSAT band 4. In Huai Bong,
Alisols prevailed on sloping land under forest cover (low SPOT band 8 values),
while Cambisols were found mostly on flat areas with sandstone, claystone or
alluvial deposits; or on cultivated (high SPOT band 8 values) south to west facing
sloping land. Derived mapping rules for the most important soil types in Bor Krai
were as follows: Acrisols dominate between elevations of 834–886 m.a.s.l., with
2 Beyond the Horizons: Challenges and Prospects for Soil Science and Soil. . .
57
very high correspondence with the respective reference soil maps, with validation
of the independent sampling points yielding the following matches: 80 % in Mae Sa
Mai and Bor Krai, and 83 % in Huai Bong. In all cases the major soil types were
classified correctly.
Randomized grid-based soil maps: The randomized grid-based soil mapping for
Bor Krai revealed 97 % correspondence with the reference soil map and a positive
match for 76 % of all independent sampling points. As seen in the transect
approach, minor units (Chernozems, Ferralsols, Fluvisols, Gleysols) located
between the randomized sampling points were not detected (Fig. 2.7).
Maximum likelihood-based soil maps: The level of correspondence with the
reference map was high throughout, showing an 81 % match in Mae Sa Mai and Bor
Krai, and a 71 % match in Huai Bong, while validation with the independent
sampling points revealed a match of 70–71 % at all the sites. Validation of the
maximum likelihood-based soil map in the Mae Sa watershed, but outside the Mae
Sa Mai sub-area, using 30 independent sampling points, revealed a positive match
in 87 % of all the 30 cases. In the pilot study area of Bor Krai, where only 25
sampling points had been distributed among the local soil units (using a minimal
sampling point approach), the classification accuracy was 80 %, while during
validation 71 % of soils were classified correctly. Due to their limited spatial
coverage and poor definition, Chernozems were not predicted for Mae Sa and
Fluvisols not for Huai Bong. The preselection of major variables using principal
component analysis was essential for the maximum likelihood classification and
greatly improved the results. For all three areas, principal component analysis
identified aspect, petrography, slope and the SPOT 5 bands 2 and 4 as being the
most important predictors – explaining most of the observed variance, such as 82 %
in Mae Sa Mai, 78 % in Huai Bong and 92 % in Bor Krai. In addition, in Mae Sa
Mai, SPOT 5 band 3 and in Huai Bong, LANDSAT 7 bands 5 and 7 were essential.
In Bor Krai meanwhile, LANDSAT 7 bands 5 and 7 as well as elevations were also
important.
Classification tree-based soil maps: The level of correspondence between the
classification tree-based soil map (1:50 k scale) and the site specific reference maps
was satisfactory throughout, with a 79 % match for Mae Sa Mai, 70 % for Huai
Bong and 84 % for Bor Krai, while validation of the independent sampling points
showed a matching range of 73–77 % for all the sites. The advantage of the
classification tree approach is the ability to develop rules for the occurrence of
predicted units; for example, in Mae Sa water bodies split off from other mapping
units due to high values in SPOT band 3 and low values in SPOT band 4, whereas
Technosols were identified via a combination of high values in SPOT bands 3 and 4,
and LANDSAT band 1, and low values in LANDSAT band 4. In Huai Bong,
Alisols prevailed on sloping land under forest cover (low SPOT band 8 values),
while Cambisols were found mostly on flat areas with sandstone, claystone or
alluvial deposits; or on cultivated (high SPOT band 8 values) south to west facing
sloping land. Derived mapping rules for the most important soil types in Bor Krai
were as follows: Acrisols dominate between elevations of 834–886 m.a.s.l., with
2 Beyond the Horizons: Challenges and Prospects for Soil Science and Soil. . .
57