U.S. flag

An official website of the United States government

Dot gov

Official websites use .gov
A .gov website belongs to an official government organization in the United States.


Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.


Main content area

Modified Centroid for Estimating Sand, Silt, and Clay from Soil Texture Class

Matthew R. Levi
Soil Science Society of America journal 2017 v.81 no.3 pp. 578-588
clay, clay fraction, databases, ecological function, field capacity, geometry, land management, models, particle size, sand, silt, soil profiles, soil sampling, soil surveys, soil texture, texture, water content, wilting point
Core Ideas Knowledge of texture class and clay enables continuous estimates of sand and silt. Rosetta under predicted water retention compared with measured values. Rosetta performance differed by soil texture class. Rosetta estimates were similar for modified centroid and measured particle size. Models that require inputs of soil texture are often limited by the availability of soil information. While texture class is easier to obtain than particle size, texture classes do not represent the continuum of soil size fractions. Soil texture class and clay percentage are collected as a standard practice for many land management agencies (e.g., NRCS, BLM, FAO) and clay content is frequently estimated with acceptable accuracy (±5%). When clay and texture class is known, sand and silt can be constrained to a narrow range that may differ from the geometric centroid of a given texture class (gcent). I tested the concept of a modified centroid approach (mcent) using 75,736 soil samples from the National Cooperative Soil Survey pedon database that also had measured hydraulic soil properties. Comparisons were made using the Rosetta pedotransfer function (PTF) to test modeled values from gcent, mcent, and measured data in comparison with measured values of water content at field capacity (θ₃₃₀) and wilting point (θ₁₅₀₀₀). The mcent approach produced a continuous distribution of values for sand and silt whereas gcent produced a stair‐step pattern for each of the 12 texture classes. Rosetta underestimated water content. Error metrics were similar when all texture classes were combined, but differences between models emerged when individual texture classes were compared. Results support the practice of collecting field estimates of clay percentage along with texture class to expand the use of soil profile data for both spatial and non‐spatial modeling of soil processes that will advance our understanding of soil contributions to ecosystem function.