Date: Mon, 16 Mar 1998 13:57:36 +1000
To: owner-sowacs@aqua.ccwr.ac.za
From: cliff.hignett@adl.soils.csiro.au (Cliff Hignett)
Subject: Re: Neutron probe : calibration curves for all textures?
Reply to Marcus - re 'universal neutron meter calibration'
Clinton Shock is dead right - anyone who uses a NMM without a local
calibration is asking for trouble. HOWEVER, one trick we learned was that
the NMM is VERY good at measuring water content change - and not so good at
measuring water content. This may seem a silly comment - but after 20
years, I assure you it is not. You can ALMOST get away with a 'universal '
calibration for change in water content.
If we are talking about homogenised soil in a calibration vessel then water
content and water content change are indeed, the same. As soon as you go
into the field, water content and water content change are quite different
because (in my experience) all fields have soil variability. The effect
is particularly pronounced in duplex soils or in soils with a varying amount
of clay at a particular depth. This means that the H present in clay which
is NOT H2O is causing the count at a particular depth to vary independently
of water content. (Where I come from, soils are notoriously low in organic
matter, so H in organic matter is not a problem)
We looked at 50 Australian clays and measured the (non H2O) hydrogen by
first heating them to 105 degree C (to drive off 'water') then burnt them in
a tube furnace at 800 degrees C in dry oxygen and used a steam absorbant
material to collect steam given off We called the quantity of steam the
'equivalent water'.
the relationship we found was
We = 0.124 (+-0.012) C + 0.015
where C is clay content in g/g and We is equivalent water also in g/g.-
measured relative to 105 degree soil weight.
This is published in 'Soil Water Assessment by the Neutron method' ed E L
Greacen publisher CSIRO Australia. 1981.
We also published relevant work in Australian Journal of soil Science :
Greacen E L and Hignett C T (1979) 'Sources of Bias in the field calibration
of a NMM' AJSR 17 405-15.
Knowing why we have variability is all very well but how do you USE this
information?
It is my experience that most users of the NMM ultimately want water content
CHANGE from the device as their objective. ie they calibrate for water
content (against count or count fraction), they measure neutron count then
use the calibration to get water content then CALCULATE water content change
between reading times.
We found that not only was this counter productive in error terms
(subtracting one value subject to field and calibration curve error from
another point similarly affected by error doubles the calibration curve
error) but as the above reference showed, it actually produced the WRONG
answer BECAUSE of the field error. The result was biased - ie even with
reduced error, the mean was tending to the wrong answer.
What do you do about it?
a) if you have already calibrated against a field using randomly selected
sites then you must tolerate the bias. However, you can still decrease the
error in estimation of water use down to that inherent in your calibration
by a simple modification of the arithmetic procedure which can be done
retrospectively if required.
Instead of calculating water content at each depth and replicate in the
field for each time and then subtracting water content at one time from
another, calculate the CHANGE in counts at each access tube and depth
between two reading times. THEN use the slope of your calibration only, to
calculate the water used. If you have a field where the clay content
varies with depth or position, then you will see a spectacular reduction in
field error when you analyse the error in change in water content compared
to water content itself.
A full explanation of this phemomena requires too much space to give it
here, but you can get some idea of what is happening if you consider an
extreme case of a field with loam at all depths in all but one replicate
access tube which has clay at all depths. The field error associated with
(any) water content measure in this field will always be large because of
the higher intrinsic water content of the clay - most of which is below
wilting point and will not be used. However, the change in water content in
the loam over a time interval is not that different to change in the clay
over the same interval therefore the error in water content change across
the field is smaller. This effect is further enhanced by the NMM
characteristics which is VERY sensitive to clay content due to its H content.
b) If you have not yet calibrated (or can face that horrible job again)
i) make the decision as to how many sacrificial calibration access tubes
will be used as you would have before, based on the field situation.
ii) Then group the tubes into pairs - and situate the tubes comprising the
pair as close to each other as possible one to be sacrificed in the dry
condition and one in the wet. The assumption will be made that the field
clay content is the SAME for the two tubes in the pair and that field error
will be manifested BETWEEN pairs. (In my experience, this assuption is
reasonable if the tubes are within 2m of each other)
iii) treat each soil horizon in each pair separately when doing the
calibration. - you should end up with a series of parallel lines with the
position of the line depending on the clay content at the site. (You may
like to use the soil clay content and the above equation to see if the
displacement along the water axis is equal to the equivalent water content)
iv) to calculate water content change use the average SLOPE of these lines
and the CHANGE in counts or count fraction at each reading point. You will
find a much reduced error term in any field where clay content varies
v) if you need to know water content then use the same data to plot a
conventional calbration - use it once only (at the driest or wettest time of
the year) then use (iv) to calculate change in water content from this time.
If you really want to be horrified, try using the same field calibration
data to prepare a conventional and a difference calibration curve and then
do the corresponding arithmeticfor each method - you may find up to 30%
difference in water use - this is the bias referred to in the reference
above. The bias results because the slope of the calibration line through
all the points will NOT be the same as the average of the slopes of the
calibrations for each soil location. (If you doubt this then try drawing a
series of parrallel lines on a page , mark a point at the ends of each line
and do a regression on those points. )
For those that are still with me, you might like consider that the basic
regression formulae assume that there is NO error in the independent
variable. Under some calibration procedures, the smallest error may be in
the water content . Specifically those procedures were the calibration
measures ALL the soil counted by the NMM. Therefore regressions should be
done with count vs water content not water content vs counts. (NO the
answer is NOT the same, if there is error then the the two lines are QUITE
different).
Cliff Hignett
Cliff Hignett CPSS CPAg
CSIRO Land and Water
PMB 2 Glen Osmond
South Australia 5064
ph (08)8303 8459
fx (08) 8303 8551
ah(08) 8276 7706
email cliff.hignett@adl.clw.csiro.au
Date: Thu, 19 Mar 1998 11:29:23
To: owner-sowacs@aqua.ccwr.ac.za
From: Trevor Finch
Subject: Re: In response to Cliff's question on neutron meters
I was interested in Cliff Highett's comments with regard to the importance
of *change* of soil moisture.
I have been trying to collect calibration equations for NP's for different
soil types, comparing equations from different probes by normalising to
counts in a 200 l water drum.
(taking standard counts in the shield seems to be pointless...)
Calibrating several hundred different CPN 503's in the same four sealed
drums of soil show a strong linear relationship between the different
probes. Whenever the cross-correlation was non-linear an error was found
in the 'dry' drum, probably because of an external influence on the count.
The conclusion is that the most reliable way to 'cross-calibrate' a new
instrument is to just do a careful count in water. It is also the most
practicable way to field check an instrument. If the water drum count
changes then the instrument is faulty - do not re-calibrate.
However, I couldn't understand the comments...
>We found that not only was this counter productive in error terms
>(subtracting one value subject to field and calibration curve error from
>another point similarly affected by error doubles the calibration curve
>error) but as the above reference showed, it actually produced the WRONG
>answer BECAUSE of the field error. The result was biased - ie even with
>reduced error, the mean was tending to the wrong answer.
Could you explain this a little more ?
VSW1 = Count1 * Slope + Intercept
VSW2 = Count2 * Slope + Intercept
Change = VSW2 - VSW1 = (Count2 - Count1) * Slope + Intercept
I have obviously missed something, because it seems to me that,
arithmeticaly, you get the same result whether you...
scale and then subtract
subtract and then scale
------
In terms of statistical reliabilty of measuring *changes*, it seems to me
that continuous monitoring of a site to measure *changes* is essentially
non-destructive testing and the criteria should be...
'If the total soil water has not changed and we measure again, how much
change do we get in the reading ?'
(and readings taken with a neutron probe down the same tube in stable clay
show remarkably constant counts over many years)
I would suggest that the recommendation of routine field calibration can
lead to errors. Limited soil moisture and bulk desity measurements,
especially when not taken over a wide range of soil moisture, can lead to
more errors than using a 'standard' calibration curve.
-------
Anyway, we are looking for equations for different soil types in the form...
VSW = (Count/WaterDrumCount) * Slope + Intercept
Typically...
Slope = 0.68
Intercept = -0.01
Does anybody have any equations in this form for different soil types ?
----
Trevor Finch
Research Services New England
8/16 Nicholson St, Balmain NSW 2041 Australia
email: trevor@rsne.com.au
tel: +61 (2) 9810 3563
fax: +61 (2) 9810 3323
----
Date: Fri, 20 Mar 1998 17:30:31 +1000
To: owner-sowacs@aqua.ccwr.ac.za
From: cliff.hignett@adl.soils.csiro.au (Cliff Hignett)
Subject: Response to Trevor Finch
Answer to Trevor Finch's question about differences between neutron
moisture menter calibration for water content change and water content -
The concept that I was explaining is not an easy one to grasp - even if you
were in my office and I could sketch the effect we have found. However, I
will try to do it in words.
First SOME DISCLAIMERS
1) If your field is uniform in clay content (ie a pure sand or a clay which
is uniform in type and % at all depths) then my comments DO NOT APPLY.
2) If you are using the NMM to measure water in a field and are not
concerned about errors up to +-20% in water content or +_10% in water
content change then my comments DO NOT APPLY (for many commercial
applications this error level is quite acceptable)
3) if the scatter of the points in a conventional calibration is trivially
small (ie <+-5%)then my comments DO NOT APPLY
WHERE DOES IT MATTER?
My comments only apply where the NMM is being used as a precision instrument
in a research application where it is worth spending effort and money to get
the error down to 5% or better. In these applications, there is usually
some form of replication of access tubes and an 'experimental design' which
will be analysed in a statistical sense - eg a replicated block design
comparing water use of different crops.
In such work, field error is often a major consideration especially in
geographic areas (like mine) where soil variability is a major problem. In
my location we have duplex soils which vary in clay content from 2% in the
surface to 30-50% at 300mm depth then reduce to 20% or so at 800mm depth.
Not only is there this vertical variation in soil type but variation in
depth to the clay horizon can occur across any experimental site along with
variation of clay content at a given depth.
If the variation is associated ONLY with depth (as in a duplex soil) then
it is simple enough to do a separate calibration for each soil horizon.
BUT clay content also varies at a single depth across a field. The worst
case is near the interface between the clay and surface loam horizons when
the depth to clay varies in a random fashion across the field which is seen
at the NMM measuring depth as field variability.
Needless to say, we have been strongly motivated to seek a 'universal
calibration' for the NMM to reduce the effort going into calibration. We
also have been motivated to look closely at the effect of this field
variabilty on the calibrations and the results which we have obtained.
SOME HISTORY AND BACKGROUND (I don't know what your prior knowledge is)
I should point out that this work is not new, it was done by Dr E L Greacen
and myself in the 1970's and (mostly) published in Australian Journal of
Soil Research.
NMM calibrations done in sands and clays have a different slope. In
searching for a universal calibration, we sought a reason for this
difference - preferably something we could measure in the laboratory on a
gravimetric soil sample.
At first we thought it was soil density, and NMM theory supports this -
neutrons are slowed and scattered (ie reflected back to the counter tube) by
H , but they are also scattered (without slowing) by any atom including Si,
Al and oxygen - ie other soil components. So it was conceivable that the
size of the slow neutron 'swarm' around the counter tube was being changed
by density - hence affecting the calibration. We tested this with both a
simulation model of neutron scattering and field calibrations. (AJSR 1979,
17:405-15). We were only partly right.- density has an effect, but that
effect is caused mainly by the non water H content in the clay. (This H
can be associated with organic matter, but we have very little of that
here). As the soil density decreased going from sand to clay, so did the
concentration of this non water derived H, which we were not including in
the calibration. At the time the book 'Soil water assessment by the neutron
method' was published, we were measuring this non water H and multiplying
its mass concentration by 9 and calling it 'equivalent water' ( in H2O, H:0
as 1:9) and adding it into the calibration as if it was water which did not
'move'
This answered a lot of questions, but it still missed the most critical
effect of variation in clay, equivalent water, and density. That was,
the effect when a field calibration includes variation in the clay content
of the soil being calibrated. Normal calibration procedures assume that the
only variables are counts and water (as released from soil at 105 degrees
C). If there are other variables affecting the relation between water and
counts then they show up as error either in the slope or the intercept.
By measuring equivalent water and adding it to the ordinary water we
accounted for a lot of the error. However, to use this information in the
field, we would have to know the density and equivalent water for every site
and depth in the field. (At this stage gravimetric sampling started to look
more attractive than the NMM) . However, we found a more general way to
solve the problem.
FIRST, WHAT IS THE EFFECT?
Consider the following calibration data
count water
fraction cc/cc
, pair 1
0.2 0.2
0.37 0.32
, pair 2
0.3 0.3
0.4 0.37
, pair 3
0.3 0.35
0.4 0.42
The regression of water on count fraction is Vw = 0.044(+-0.03) +
0.85(+-0.21)*Cf
ie a change in counts of 0.1 indicates a water change of 0.85
While I have stretched these points a little for effect, I think you will
agree that the error associated with this calibration is not unduly large
for a field calibration in a highly variable field. The calibration slope
is also typical of many in loam type soils.
Now, consider the additional information that each pair of points came from
a pair of access tubes situated close to each other (one coordinate sampled
wet and one sampled dry) at each of three locations spread across the
field. Under these conditions it is reasonable to assume that the field
variability between the tubes of each pair is negligable (because they are
close together) compared to the error between different pairs (say 100m
apart) . If you plot the points and connect the wet and dry point for each
pair you will see that the slopes for all pairs are identical (no error here
at all - yes, I did cheat - I have never got data this good, but I need to
make a point)
You will also note that the slopes are 0.70, NOT the average slope of 0.85
from the regression done on the SAME set of points. Clearly, if three
individual sites showed that a count change of 0.1 is equivalent to a
moisture change of 0.07 and the combined calibration shows a change of 0.1
is equivalent to a moisture change of 0.085 then something is wrong. The
discrepancy is an unacceptable 0.015/0.07 = 21%
I have emphasised the effect here for demonstration purposes but it is real!
It does happen!
Our explanation of this phemomena is that the three sites had different
amounts of clay H eg pair 1 had 1% volumetric equivalent water, pair 2
had 1.3%, pair 3 had 1.8%. If you plot total water (ordinary water plus
equivalent water) against count, you will see the slope is 0.7 and the error
is near zero.
HOW TO INTERPRET THIS
We interpret this sort of data as meaning that
(a) in a field with high clay variability the statistical uncertainty
associated with measuring, or calibrating for, water content is very high
and unavoidable. Because of displacement of points along the water axis of
the calibration by an unknown amount.
(b) this uncertainty primarily applies to the water content, it does not
affect the change in water content because the clay at a single measurement
point does not vary.
(c) the problem can be avoided by pairing the wet and dry calibration tubes
to get an accurate measure of the calibration slope without the effect of
clay variation.
NOTE
1) that the best measure of field water content still is the conventional
calibration analysis (ie regression of all points in together) You must
live with the statistical error caused by clay H.
2) if this conventional calibration is used to calculate water content
change the result is not only subject to the above statistical error but is
biased. ie even if the error is zero, the answer will be wrong!
3) If water content CHANGE is needed then the best measure comes from the
average of the slopes of the individual pairs of tubes. (in this case you
don't know the actual water content)
WE RECOMMEND
1) pairing calibration tubes (ALWAYS ! - it doesn't cost any more)
2) do a conventional calibration data analysis for use in measuring water
content.
3) calculate the average slope of the pairs of tubes (we call this a
diference calibration) and use it to measure water content change.
4) DO NOT use a conventional calibration to calculate the water content at
two points in time and subtract them to get water use.
Hope this helps describe our findings
Cliff Hignett
Cliff Hignett CPSS CPAg
CSIRO Land and Water
PMB 2 Glen Osmond
South Australia 5064
ph (08)8303 8459
fx (08) 8303 8551
ah(08) 8276 7706
email cliff.hignett@adl.clw.csiro.au
