I wish to produce a range of soil maps for a dataset, ideally focused on the distribution of soil sand, silt and clay within a study area. Unfortunately, the dataset at my disposal was produced using soil texture field tests (according to USDA), and therefore only contains categorical soil texture descriptions (Sand, Sandy-clay, Sandy-loam etc), not quantitative particle size fraction data.
Are there any recommended methods for producing soil mapping from categorical soil texture data, either by interpolation or other means? Is there any information I can salvage from this data for spatial analysis of soil characteristics?
As per my understanding, you can do "Boolean Query" first with sand, silt and clay ranges of your data-set. Then you can do interpolation > Reclassify.
Please check some of my articles:
https://www.researchgate.net/publication/215783330_Appraisal_of_Soil_Fertility_for_Sitespecific_Fertilizer_Management_in_Irrigated_Rice_Domain_by_GIS_and_Multivariate_Analysis?ev=prf_pub
https://www.researchgate.net/publication/230669412_Assessment_of_Influence_and_Inter-Relationships_of_Soil_Properties_in_Irrigated_Rice_Fields_of_Bangladesh_by_GIS_and_Factor_Analysis?ev=prf_pub
Conference Paper Appraisal of Soil Fertility for Sitespecific Fertilizer Mana...
Article Assessment of Influence and Inter-Relationships of Soil Prop...
There are some applicable methods.
1. Having merely texture data you can rely on Indicator Kriging. This is what we did in our work for building a 3D soil hydrophysical dataset (http://www.akademiai.com/content/k357127241236474/)
2. Having spatial auxiliary environmental datasets (based on DDM, remotely sensed data, geological map etc.) you can choose an appropriate DSM data mining method, like classification tree.
Exhaustively available auxiliary data can also be used in combination with an indicator kriging approach. Examples are "Collocated indicator cokriging" (see: https://www.researchgate.net/publication/222676290_Predicting_the_Areal_Extent_of_Land-Cover_Types_Using_Classified_Imagery_and_Geostatistics?ev=prf_pub) or (easier) some indicator variant of regression kriging.
Note that indicator kriging is a pragmatic solution that has several flaws, one of which being the treatment of indicator transformed data as if they are normally distributed.
Article Predicting the Areal Extent of Land-Cover Types Using Classi...
Tricky question. The problem that shuld be addressed at first is the fact that the texture is determined per soil horizon. The texture of surface horizons (e.g. A, Ap...) can significantly differ to the texture of lower e.g. cambic horizons (Bv, Brz...). Thus it's difficult to talk about general soil texture. In the second step the silt/sand/clay distribution maps for a certain depth (e.g. 0-25cm); can be prepared. What can be done is to interpolate the medium values od clay/silt/sand content for each USDA texture class. Medium values can be defined form texture triangle.
Going this was the error is expected to be acceptable in comparison to the case when the depth/horizon issue is ignored.
Congratulations Borut, very good answer. In fact, create spatial distribution from semantic dataset is a great problem, maybe using medium values and focusing in its interest horizon solves Robbi problem.
Ref: Borut, I wouldn't suggest to interpolate the median values; it will associate a lot of unrealistic values while interpolated. I agree with your first point though. I would suggest to create binary maps with each soil type e.g. 1=sand, 0=no sand.
The most available and pragmatic solution to this problem is the Indicator Kriging, though accomplishing this is not an easy task. The most inspiring example I have at hand is: http://www.sciencedirect.com/science/article/pii/S0016706107001218, where the authors dealt with a similar problem. And here is a example script where the authors presented how they dealt with the problem: https://stat.ethz.ch/pipermail/r-sig-geo/2009-November/007076.html.
Hope that helps.
I guess my mistyped words caused some confusion. "What can be done is to interpolate the medium values of clay/silt/sand content for each USDA texture class. Medium values can be defined from the texture triangle. Going this way the error is expected to be acceptable in comparison to the case when the depth/horizon issue is ignored."
Other words turn categorical texture classes to what they actually represent: % of sand, clay, silt by defining the central point of the individual texture class area in the triangle. E.g. define Loam texture class by 42 % sand, 18 % clay and 40% silt. It’s worth trying simple and straightforward approach first.
Find map of you country or county. Clip his map to your study area. Find location coordinates of your sample points if not exact but nearby. Inter this tabular data of each sample point for each class of texture. save this format as point map. Overlay with Area map so that you find your points in specified spatial location. Go to Arc GIS10---- Interpolation technique and choose one of the techniques. Save the out put. Better try atleast 2 techniques. Compare their performance in predicting vales of unsampled points. Compare the accuracy. Choose the best technique with better accuracy. This results in spatial map of your soil texture in your geographical area for further use and application.
- Badges
- Science topic
- Similar topics
- Agriculture
- Soil Science
- Soil