Soil conversions

Soil-conversions

Soil texture is reported as a triple (sand, silt, clay), but the definition of sand and silt depends on the standard you use. USDA puts the sand/silt boundary at 0.05 mm; ISSS at 0.02 mm; KA5 at 0.063 mm. The clay/silt boundary is universally 0.002 mm. So a soil classified as "loam" in USDA may classify differently under ISSS, not because the soil changed but because the binning did.

pedotri.psd.convert translates a (sand, silt, clay) triple from one standard to another so you can re-classify your data without re-analysing the sample.

When do you need this?

You need a particle-size standard conversion when:

Situation	Why you need it
Lab uses one standard, the PTF you want to apply uses another	E.g. an EU lab reports ISSS percentages, you want to run Saxton-Rawls (trained on USDA)
Cross-comparing data from labs in different countries	Mixing French (USDA convention) with German (KA5) requires aligning the sand boundary
Driving a published regression / classification that fixes a specific cutoff	E.g. HYPRES is internally consistent — its inputs were re-binned to a single cutoff before fitting
Updating legacy ISSS data to USDA for a modern analysis	Pre-1980 European literature predominantly used ISSS

You don't need a conversion when:

• All your data and all your downstream tools use the same standard.
• You only need the clay fraction (silt/clay boundary doesn't move).
• You're crossing the USDA ↔ FAO line — both use the same 0.05 mm cutoff, so the triples are identical; only the class grouping differs. Pass the same numbers to classify(..., 'FAO') directly.

What is actually being computed?

The conversion is approximate because it requires assuming a distribution within the silt (or sand) fraction. Pedotri uses a log-linear assumption: the cumulative mass-passing curve is treated as a straight line in log-particle-size space inside each source- standard fraction.

Why log-linear? Because that's how a typical soil PSD actually looks on a log scale over the silt range. It's also the simplest reasonable model that gives a closed-form answer from three input points (sand%, silt%, clay%). For higher accuracy you need a full sieve PSD — see interpolate_psd below.

USDA → ISSS (boundary moves finer: 0.05 → 0.02 mm)

ISSS draws the sand boundary at 0.02 mm, so the 0.02-0.05 mm fraction that USDA called "sand" should be called "silt" under ISSS. We need to estimate how much of the USDA sand sits in that 0.02-0.05 mm slice.

Under the log-linear assumption over the USDA sand range [0.05 mm, 2 mm]:

fraction_remaining_as_sand_ISSS = log10(2 / 0.02) / log10(2 / 0.05)
                               = log10(100) / log10(40)
                               ≈ 1.249

That ratio is > 1 because we're stretching the same mass over a wider log-range, but in this direction (finer cutoff → smaller "sand" fraction) the formula is inverted; pedotri uses the symmetric form so the math is consistent in both directions. See pedotri.psd.convert source for the exact derivation.

USDA → ISSS, numerically

import pedotri.psd as psd

# Sample analysed in USDA convention
sand_usda, silt_usda, clay = 40.0, 40.0, 20.0
sand_isss, silt_isss, _ = psd.convert(
    sand_usda, silt_usda, clay, source="USDA", target="ISSS"
)
print(sand_isss[0], silt_isss[0], clay)
# 32.6  47.4  20.0

Roughly 7 percentage points migrate from "sand" to "silt" — that's a typical magnitude. The clay column is unchanged because all standards agree on the 0.002 mm cutoff.

USDA → KA5 (boundary moves coarser: 0.05 → 0.063 mm)

KA5 has a higher sand boundary, so part of the USDA sand gets reclassified as KA5 silt:

sand_ka5, silt_ka5, _ = psd.convert(
    40.0, 40.0, 20.0, source="USDA", target="KA5"
)
print(sand_ka5[0], silt_ka5[0])
# 38.2  41.8

A smaller shift because 0.05 mm and 0.063 mm are close in log-space.

Reversibility

Conversion is not exactly reversible. A round-trip USDA → ISSS → USDA generally differs from the input by 1-2 percentage points:

s, si, c = 40.0, 40.0, 20.0
a, b, _ = psd.convert(s, si, c, source="USDA", target="ISSS")
s2, si2, _ = psd.convert(a[0], b[0], c, source="ISSS", target="USDA")
print(s2[0], si2[0])   # close to 40, 40 but not exactly

This is a property of the single-point log-linear model, not a bug. If you need exact round-tripping, store the original measurements and re-derive from the source standard each time.

Kachinsky — the non-triangle

The Russian Kachinsky classification cannot be derived from (sand, silt, clay) under USDA / ISSS / KA5. Kachinsky keys on physical clay — the mass fraction of particles < 0.01 mm — which sits between the silt/clay cutoff (0.002 mm) and the silt midpoint. You can't reconstruct it from a triple that aggregates everything below 0.05 mm (or 0.02 mm) into one number.

The Russian Soil Institute publishes the official 2004 / 2008 nomenclature and the synonymy with the 1977 / 1997 schemes at infosoil.ru — see infosoil.ru/index.php?pageID=raznovid04 for the variety-level names. The six official 2004 varieties (песчаная / супесчаная / лёгко-, средне-, тяжелосуглинистая / глинистая) are shipped as the RUS2004 classification — a six-class collapse of the nine-class KACHINSKY. Use whichever resolution matches the document you're reproducing.

If you have a full sieve PSD, use interpolate_psd to read off the < 0.01 mm fraction:

import numpy as np
import pedotri.psd as psd

sieves_mm = np.array([0.002, 0.01, 0.02, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0])
passing  = np.array([18.0, 35.0, 47.0, 55.0, 62.0, 72.0, 88.0, 95.0, 100.0])

physical_clay = psd.interpolate_psd(sieves_mm, passing, target_size_mm=0.01)
import pedotri
pedotri.classify(physical_clay, "KACHINSKY")
# e.g. 'medium_loam'

If you only have the (sand, silt, clay) triple under a different standard, you can't classify Kachinsky — say so loudly in your methods section and pick a triangle-based classification instead.

interpolate_psd — arbitrary cutoffs

When you have a richer PSD than three numbers, prefer interpolate_psd directly. It's the building block under psd.convert and avoids the log-linear assumption inside each fraction:

import numpy as np
import pedotri.psd as psd

# Sieves in mm and cumulative % passing
sieves = np.array([0.002, 0.063, 0.2, 2.0])
pct    = np.array([14.0, 51.0, 78.0, 100.0])

# What's the USDA sand fraction (i.e., > 0.05 mm)?
pct_below_50um = psd.interpolate_psd(sieves, pct, 0.05)
sand_usda = 100.0 - pct_below_50um
print(sand_usda)   # ~51%

How big are these corrections?

For "normal" soils, USDA ↔ ISSS conversion moves ~3-8 pp of mass between sand and silt. That's enough to change the classification in a meaningful fraction of cases — especially near boundaries like loamy-sand ↔ sandy-loam or silt-loam ↔ silt. If your downstream analysis is decision-bearing (irrigation scheduling, crop modelling, hydraulic-property prediction), convert first, classify second.

Related

• Classifications-and-when-to-use-them — picking the right target standard
• Pedotransfer-functions — most PTFs are trained against a specific standard, so converting before driving a PTF is usually mandatory
• Units-and-organic-matter — units= keyword on psd.convert for g/kg / g/g inputs