Item 23. Development of transfer functions – progresses, needs and way forward
1. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Item 23
Transfer functions – progress, needs and way
forward
Mr. Rich Ferguson
KSSL, USDA
2. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
What is a pedotransfer function?
A pedotransfer function – or “transfer function” (TF) for short – estimates an
unmeasured soil property from a measured soil property or measured soil
properties
Johannes Bouma: “translating data we have into what we need”
Examples:
Estimated CEC = f(measured clay, clay mineralogy, SOM)
Estimated 1500 kPa soil water = f(measured sand, silt, clay, SOM)
Estimated organic carbon = f(measured infrared spectrum)
= f(same soil property - measured by legacy
method)
E.g. Estimated GLOSOLAN pH = f(legacy method pH); Estimated DC OC = f(WB OC)
Estimated soil property by
GLOSOLAN method
3. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
2019 | 3rd GLOSOLAN meeting
• GLOSOLAN goal: Harmonization of data from different methods (SOPs) for measuring
the same named property.
• Demonstration of harmonization using a transfer function was made using two methods
for measuring organic carbon: Dry combustion (DC) & Walkely-Black (WB)
WB and DC comparison data were contributed by
GLOSOLAN members from 25 countries
Data were anonymized by GSP and analyzed
(Ferguson)
DC was predicted from WB (as DC is more expensive
to execute than WB)
TF performance was of fair quality
4. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
2019 | 3rd GLOSOLAN meeting
To improve (lower) TF errors:
Go to smaller geo-taxonomic scales:
o less soil variability improved TF performance
Use data from a single laboratory
o eliminates contribution by interlaboratory method
imprecision
Use a single WB method variant
o Improved precision more consistent TF
performance.
Improve laboratory quality control
Example of improved TF performance:
o samples from single county in Nebraska
o data from a single laboratory
o using a single method variants
o with good quality control
o TF performance was of
good quality
5. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Current applications of transfer functions on the
GLOSOLAN agenda:
• Soil spectrometry (see work plan of GLOSOLAN on soil
spectroscopy)
http://www.fao.org/global-soil-partnership/glosolan/soil-analysis/dry-chemistry-spectroscopy/en/
• Harmonization of data from different methods (SOPs) for
measuring the same named property
During 2020
• Tyurin method data was harmonized with Walkley-Black data (Shamrikova)
• Additional GLOSOLAN methods were standardized – sets stage for
harmonization with legacy methods, etc.
• GLOSOLAN worked on quality control to ensure that datasets used by
laboratories to develop transfer functions are of good quality
6. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Why harmonize data sources? Examples:
1. Harmonization promotes comparability between GLOSOLAN and
historical data
2. Harmonization allows laboratories that cannot implement GLOSOLAN
methods (e.g. by law) to promote comparability of their data with those
of laboratories using GLOSOLAN methods
3. Harmonization promotes comparability between alternative GLOSOLAN
methods for the same named property
4. Harmonization allows integration of soil information obtained from
different laboratories, with implications for transregional and
transnational research and mapping activities
7. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Harmonization of GLOSOLAN and
legacy method data
• Is there bias or incongruency between two methods you’d like
to harmonize? E.g. legacy method and the GLOSOLAN method
data sets? (yes?)
• Are you interested in harmonizing data from two methods, to
the extent possible? (yes?)
If your answers to the above questions were “Yes”, please
consider developing a transfer function that relates the two
method data sources.
8. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
To develop a statistical relationship (a “transfer
function”) relating GLOSOLAN and legacy method
data, what’s needed?
• Comparison data on samples from the target area that the transfer function is
intended to serve
• 100+ samples from a target region to develop transfer function
samples should represent the target region that transfer function would serve, in
terms of:
• area
• depth
• range of property values, soil variability
include a few QC samples
samples should be analyzed by both methods being related; e.g. GLOSOLAN and
legacy methods
measured data should be obtained under conditions of best achievable
laboratory quality control, for maximal transfer function reliability (lowest RMSE,
or error)
9. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Once you’ve got the two sets of comparison data,
you can build the transfer function
• Good practices:
• For those 100+ samples
Hold back a representative set (e.g. 20%) of the paired data for TF validation
Build TF from the rest (e.g. 80%) of the paired data; e.g. by linear regression
analysis.
• Building transfer functions in specific input ranges may be necessary to achieve
optimal performance
Process the validation set through the TF to estimate soil property by
whichever method is being estimated/predicted
Calculate the residuals (differences) between measured and estimated soil
property values
10. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Once you’ve got the two sets of comparison data…
you’re ready to build the transfer function.
• Good practices (cont.):
• Calculate and report the correlation and RMSE (error) for the validation
set.
Errors of transfer functions are often not reported, but they should be.
• The RMSE informs of possible distance between estimated value and what
would have otherwise been measured
• Poor (high) RMSE values may mean that the user elects not to use the TF
Possible reasons for poor (high) transfer function errors:
• One or both methods suffer from high imprecision
• One or both methods were conducted with poor laboratory quality control
• Target region might be too variable in terms of a wide range of soil types that
TF would serve
11. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Opinion on terminology
Is the term “transfer function” (Johannes Bouma: “translating
data we have into what we need”) acceptable for describing
the statistical approach for harmonizing incongruent methods
for measuring the same named soil property?
Poll : Yes / No
If your answer is “no”, please consider proposing
alternative terminology.
12. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Proposed work ideas for laboratories to implement
Activity 1. Do a literature research on transfer functions of interest to you that may
have been to soils in your region.
Activity 2. Develop a workplan for building and implementing needed / desired
transfer functions.
Activity 3. Collect comparison dataset for transfer functions derivation.
Activity 4. Use your transfer functions for practical purposes, such as harmonization
of legacy and GLOSOLAN methods.
Activity 5. Share what you’ve learned, successes or roadblocks; e.g. at the next
GLOSOLAN meeting. If possible, please share your data and transfer functions
(including errors!).
NOTE: Transfer functions are the responsibility of individual laboratories to
develop. On request, GLOSOLAN will provide technical guidance to facilitate
these efforts. Please let Lucrezia and Nopmanee know if you desire such
assistance.
13. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Activity 1. On request, assist local GLOSOLAN efforts on inter-method data
harmonization by providing guidance from experts within GLOSOLAN.
Activity 2. Prepare a quick-reference guide to building a transfer function for
harmonization.
Activity 3. Promote and facilitate discussions at RESOLAN meetings around needs
and progress in harmonization.
Activity 4. Develop a repository at FAO-GSP for inter-method comparison data and
transfer functions that local efforts wish to make available to other GLOSOLAN
participants.
Proposed work ideas for GLOSOLAN at large
14. 4th Meeting of the Global Soil Laboratory Network (GLOSOLAN)
Thanks for your attention