JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 89, NO.

D2, PAGES 2616-2626,APRIL 20, 1984

Correcting
SatelliteDopplerData for Tropospheric
Effects
H. D. BLACK AND A. EISNER

AppliedPhysicsLaboratory,The JohnsHopkinsUniversity
Using a simplegeometricalmodel,and one fitted parameter,troposphericeffectscan be effectively
removedfrom satelliteDoppler data at microwavefrequencies.
Both the wet and the dry parts of the
troposphericrefractioneffectare removed.The techniqueworks best for low (say, 1200 km or less)

altitudesatellites.
For thesesatellites,
the pass(transit)dui'ationlimitsthe requiredatmospheric
correlationtimeto about20 min.Theeffective
thickness
of theneutralamosphere
(10km for thewetand45
km for the dry) limitsthe requiredcorrelationdistanceto a fewdegrees
in ltitude and longitude.These
conditionsare satisfiedoften enoughto make the fitting techniquehighly useful.The fitted parameter
togetherwith minimaldependence
on modelstructureappearsto skirt a difficultproblem,modelingthe
water vapordistributionin a poorlymixedatmosphere.
Experimentalresults(for a limitedtime period)

confirmthat the approachis validon a globalbasis.In anothercontext,thistechnique

canbe usedto
intensivelysamplethe precipitablewatervapor in the atmospherewithout usingballoons.

1.

INTRODUCTION

above 10 is given by

In correctingsatellite Doppler data for the refractiveeffect

of the neutral atmosphereit is necessaryto computethe correction to the instantaneoussatelliteslant range.This is given
by the expression

AS= 10-6 $ Ndp

(T- 4'12)i
a

As-2.343
Ps T

(6)

wherein

(1)

(7)

where N is the refractivity

N 106(n- 1)

(2) Zcis a weak functionof the elevationangleand surfacetem-

andn istheindexofrefraction.
Theintegral
isto beevaluatedperature,Zc- 1/6 for elevationanglesgreaterthan 15, and if
along an extremum path (connectingthe observerand the
satellite)consistentwith Fermat'sprinciple.
2.

INDEX OF REFRACTION

The index of refraction of the troposphereis the sum of two

effects [Smith and Weintraub, 1953; Thayer, 1974], one that
dependson water vapor and one that does not. The Smith/

Weintraub
formula
fortheatmospheric
refractivity
is
N = Na + Nw

neoussatelliteelevationangle;andReis the distancefromthe

earth's center to the observer.

Above15 elevationangletheimportantfunctional
dependence,implicit in (6), is embodiedin

(3)

As=
2.34
PsI.
Ta'12.1csc
E
(8)

wherein the "dry" term is

ma - 77.6 P/T

(4)

and the "wet" term is

Nw = 3.73x 105(e/T2)

Here we seethat the important variablesare surfacepressure

at the observationsite and the satellite elevation angle. The
temperaturedependence(exceptbelow 5 elevation angle)is

(5) soweakthatit isunimportant.

Thedryeffect
is20m orless

whereN _a_
106 x (n- 1), the "refractivity";n = index of refraction; P - total pressure,millibars(1 atm - 1013.25mbar);
T = temperature, Kelvin; e = partial pressureof the water
vapor, millibars.(The Smith/Weintraubexpressionis valid for
frequencieslessthan about 30 GHz.) It is convenientto talk of
the dry and the wet effects.
3.

Zcwere0, then la would reduceto cscE; ha,the "extent"of the

dry troposphere, a linear function of surface temperature
35 _<ha_<45 km (seeeq. (18)); T is the surfacetemperaturein
degreesK; Ps is the surfacepressurein normal atmospheres;
As is the troposphericrangeeffectin meters;E is the instanta-

for elevation angles greater than 7. The Hopfield [1971]

theory on which this analysisis basedis good to about 1%;
therefore,the residual error is lessthan 20 cm, provided we

havesurfacepressure
measurement
at the sitethat is goodto,
say, 1: 1000 or 1 mbar.
4.

WET TERM

The wet term is not easilydescribed,

but progresshas re-

DRY TERM

centlybeenmade [Goldfinger,1980]. When Goldfinger'swork

The dry term, 85-90% of the combined tropospheric
is combined with that of Reitan [1963] we find that the wet
(range) effect, is highly "predictable" and can be accurately
term correction is proportional to the "precipitable water
modeled[Hopfield, 1971;Black, 1978]:
The dry troposphericrange correction,foi' elevationangles vapor" in the atmosphereand that the latter quantity is, on
the average,linear in the surfacedew point. "On the average"
is an important qualifter.It is frequentlytrue that the water
This paper is not subjectto U.S. copyright.Publishedin 1984 by
vapor is poorly mixed in the atmosphere;consequently,surthe AmericanGeophysicalUnion.
face measurementsdo not characterizethe water vapor aloft.
Reitan averagedreadingsfor a month and found that out of a
Paper number 3D 1948.
2616

BLACK AND EISNER: TROPOSPHERICEFFECTS ON MICROWAVE DOPPLER DATA

2617

25-

2o

15

NOTE

THE BOTTOM 4 CURVES ARE

WATER VAPOR CORRECTION

EXTREME
VALUES OF THE
FOR SURFACE TEMPERATURE

OF +40, +30, 0, AND -30:C

2 THE TOP CURVE IS THE DRY CORRECTION
DEPENDENT
ON SURFACE TEMPERATURE
PROPORTIONAL
TO SURFACE
PRESSURE

IT IS WEAKLY
AND DIRECTLY

10

40

20

40

ELEVATION

Fig. l.

rn

0.67

rn

9O

(DEGREES)

Wet and dry troposphericrangecorrection.

total (vertical) correction of 20 cm, the residual error (assumingsurfacedew point measurements)was about one third
of this amount, or 7 cm, and largely random. Moran and
Rosen[1979] give resultsthat are consistentwith these:Along
the zenith,they find that surfacewater vapor densitymeasurementsproducethe path (range)correctionto 5 cm for summer
data and 2 cm for winter data. Microwavebrightnesstemperature (radiometer)measurementsnear 22.235 GHz do about a
factor of 5 better EGuiraudet al., 1979]. This is the state of the
art.

Somesimulatedvaluesof the dry and wet range corrections

are shownin Figure 1. The dry valuesare highly precise;the
wet values are not. They assumeextreme (saturated) conditionsat the specifiedsurfacetemperatureand an exponential
decreaseof water vapor densitywith height,cf. Figure 2. Typical wet values are one half to one third

60

2.32

6. Reitan [1963] gives the scale height of WTRVPR as

2.2-2.86 km. Monthly meansof precipitableWTRVPR correlate well with surfacedew point In W = a + bt.
7. Roosenand Angione[1977]: Relation betweenprecipitable water vapor and surfacehumidity showsa strongpositive correlation,but the varianceis so large that surfacehumidity is not a reliableindicatorof precipitablewater vapor
for any particularday.
8. Pitts et al. [1977] report a casein which the precipitable water vapor changes1.25 cm over a 3-hour period.The
associatedhorizontalscaleis 100 km. (The associatedchange
in verticalrangecorrectionis 7.5 cm.)
9. Goldfinger[1980] showsthat the rangecorrectionalong
the vertical is approximately6 Pwv.The units are the sameas

of those that are

shownin Figure 1.
5.

This

is

THE PROBLEM AND ITS ANALYSIS

the

1980

status

of

WTRVPR

measure-

ments/knowledge:
1. Guiraudet al. 1979] at NOAA, Boulder,get agreement
of 1 cm along the vertical (5.8 cm at 10) between twofrequencyradiometermeasurements
and radiosonde(balloon)
measurements.
WTRVPR correctioncan changea factor of 4

Ill',
lil X

7- \

I!\

heightof 2.2 kmalthough

thehigh
altitudepartdecreases
moreslowly

than
anexponential."

Moran
&Rosen
Rad/o
Sc/ence
1981

in a fewhours.Modelsutilizinggroundmeasurements
agree
with two-frequencyradiometer measurements(rms) 4 cm
along vertical.
2. VLBI, the Aries systemat the Jet Propulsion Laboratory (G. Resch,private communication,1981),currentlylimits
elevationsto anglesabove20to minimizepropagationerrors.
3. G. Resch(private communication,1981), reports that
the WTRVPR

correction

cannot be correlated

with time of

day or season.
4. Hargrave and Shaw [1978] give a vertical correctionfor
WTRVPR that changes2 cm in 10 km along the horizontal.
Measurementswere made in the United Kingdom.
0
2
4
6
8
10
12
14
5. Moran and Rosen[1979]: rms error of zenith path corPv[gm/cm3
x 10-6]
rection from surfacedata was 3.2 cm. On the average,scale
Fig. 2. The mean profile of the water vapor' data from 45 radioheight of WTRVPR is 2.2 km. See also Reber and Swope sondelaunchesat HaystackObservatoryin August1975.The dotted
[1972].
lines denote + 1 standard deviation.

2618

BLACK AND EISNER' TROPOSPHERIC EFFECTS ON MICROWAVE DOPPLER DATA

,

thoseof the precipitableWTRVPR, Pwv.Prabhakaraand Dalu

[1980] give synoptic maps of Pwv from NIMBUS 4 and
NIMBUS

6 data.

We would like to have an expressionsuch as equation(6)

for the wet troposphericrange effectwhich we could use,togetherwith (6) to correctthe instantaneousrange for both the
wet and dry effects.It is clear from the work of Moran and
Rosen(and others,seebelow) that no suchexpressioncan be
written which is more accurate than about 7 cm along the

(lessthan 15) elevation angles.For relatively high elevation

angles,we can use the straight-line path for evaluating the
integrals. For elevation angles above 15, Hopfield's results
show that the bending effect is appreciablyless than 1 cm.
More on this later.

2. Gardner [1977] showedthat the line-of-sighteffect of a

horizontalgradientin the (dry) troposphere

is typically-1

cm (peak value = 1.25 cm) if elevation anglesabove 20 are

utilized. If we include the 10-20 elevation interval, then the
vertical (60 cm at 7 elevation).
gradient effect increasesto 2-3 cm (peak value = 5 cm).
The nonequilibrium and variability of the atmospheric Gardner'sfindingswere for the mid-Atlantic seaboardduring
water vapor deny this possibility.Faced with this barrier, we January-February 1970. See also item 4 in the list concerning
have found that the following approachis a reasonableone.
the statusof WTRVPR measurements/knowledge.
We can write the sum of the wet and dry corrections:
Even though the water vapor is not well mixed in the atmosphere, it is more closely confined to the surface, and the
As= 10-6[Nadp+ Nwdp]
(9) overall effect is 10-15% of the dry effect (integral).We will
wherein the refractivities are [Smith and Weintraub, 1953] assumethat the gradienteffectsof the wet troposphereare no
given in (4) and (5) and repeatedhere
greaterthan thosein the dry (i.e.,-1 cm for all elevation
anglesabove 20. As a result of these two facts/assumptions
P
Na = 77.6 -(10)
we
can use a straight-linethrough the troposphereto evaluate
T
the integral and also ignore gradients.
We can now write (9) as
e

Nw= (3.73
x 105
K2/mbar)

(11)

(e, T, P) are the water vapor pressure,temperature,and atmosphericpressure.All are point functionsof position within the
atmosphere. The geometry and associated nomenclature is
shownin Figure 3.
For our purposes--as will be seen--the exact form of Na

and Nw are relativelyunimportant.We have given (10) and

(11) to emphasizethat both Na and Nw are point functionsof
positionwithin the troposphere.Important for our purposesis
the fact that both the dry and wet troposphereextend to
heightsabovethe earth'ssurfacethat are smallcomparedwith
the earth'sradius; the dry troposphericeffectpetersout below
ha-45 km and the wet below hw- 13 km (see Hopfield
[1971] and Figure 2).
Equation (6) is an evaluationof the first integral of (9) using
(10) for the refractivity.In the evaluation,the temperatureis
assumedto decreaselinearly with height,and the atmosphere
is assumedto be in hydrostaticequilibrium.(For details see
Black [1978] and Hopfield [1969].)
Beforeattacking the combinedeffectwe needtwo facts:
1. Hopfield [1976a] and others have found from raytracing studiesthat ray bendingis unimportant exceptat low

As= 10-6 (Na+ Nw)dp

(12)

where the integration path is now the straight line (slant

range)connectingthe observerand satellite(Figure 3).
From the geometry in the sketch we obtain
dh

dp= {1- [cos

El(1
+h/Re)]
2}1/2
and (12) can be written

As
=10
-6.o
Ihd(Nd
"}Nw)
dh
{1- [cos
E/(1
+h/R)]
2}/2 (13)
Sinceha> hwwe can write
Nw =

3.73 x 105e/T 2
0

if

h < hw
hw<h

(14)

cf. equation(11).
The term h/Re,in (13), is much lessthan 1; ha/Re is lessthan
0.007 for the dry term (seeequation(18)) and hw/Reis lessthan
0.002 for the wet term (see Figure 2). This suggeststhat we
write (13) as

As= 10-6 law

(Na+ Nw)dh

(15)

wherein

Idw
: 1-- '1-Jrddwdd
/ae'

(16)

for some0 < Zaw< 1.

(This is an expansion of the bracketed term in (13) in a
seriesabout a point within the tropospherekeepingonly the
zeroth-order term.) law is a generalizationof the cosecant'if
Zaw= 0, then law reducesto csc E. As law(E= 90) = 1, the
integral in (15) is the zenith correction. It is useful to realize

Fig. 3. Tropospheric
geometry.

thefollowingpoints:
1. Equation (13), the preciseone, and equation (15), the
approximate one, approach the same form as the elevation
angleappro/tches90.
2. Equation (15) removes the apparent singularity at the
lower limit (in equation(13))when E = 0.

BLACK AND EISNER' TROPOSPHERICEFFECTSON MICROWAVE DOPPLER DATA

'7
r
0.15

I.-

2619

-30%
0Oc

30C

II

40C

X.
clw
012

0.10

0.08'
0

20

40

60

ELEVATION

8O

(DEG)

Fig. 4. Dimensionlessparameterin model.

3. Since the satellite usually does not pass through the

zenith, the integral in (15) is the zenith effectonly if the horizontal gradientsare negligible.Since the dry effect is proportional to the surfacepressure,surfacepressurevariations
from a weather map are a convenientway of evaluating the
importanceof gradients.
4. Equation (15) is a convenientfactorizationof the troposphericrangeeffectinto the form

more preciseformulation is possibleif we admit separateZ

valuesfor both the wet and dry parts of the troposphere.Such
a form would presupposethat we have an independenttechnique for determiningthe amplitudeof either the wet or dry
term.

Using the above equationswe can (by equating the right

sidesof (13) and (15))derivethe Zdwwhichmakes(15) exact:

Re

(geometry)x (amplitude)

1/2

cos
E-1} (17)

Lla2

moreover,in the amplitude(the integral)we have "buried"the

difficultproblemof modelingthe atmosphericstate.This form
anticipatesour eventualneedfor a form having a singlefree
quantity which we can, in turn, estimatefrom the Doppler
data. This point will be clarifiedas the analysisproceeds.A

wherein

hd= 148.98(T-

4.12) m
(18)

= 46.041

km

+30C

0Oc
40C

0001

- 30 C

0.0005

0 0006

I
0

20

40

ELEVATION

(DEG)

Fig. 5. Dimensionless
parameterin equation(16).

2620

BLACK AND EISNER' TROPOSPHERICEFFECTSON MICROWAVE DOPPLER DATA

10-

{'Xhd)
ck'--e
'"'1/2THE
TOTAL
VARIATION
TROPORANGECORRECTION
= 2 5m@

ELEVATION
= 90

10

15
ELEVATION

20

(DEG)

Fig. 6. Approximation error in computingtroposphericrange error.

for T = 313 K (40C) [Hopfield, 1971] and IR is the ratio of

two integrals

+Nw)
{1
--[cos
El(1
+
d(Na
dh
IR =
d
(N,
+Nw)
dh

'

wherein

Iaw
=II- (l).01.
cos
E/2]-/2 (21)
(19)

for elevationsbetween7 and 90 and for surfacetemperatures

between -30C and +40 C. Equations (20) and (21) are
Using (17), (18), (19), (4), and (5), numerical integration and certainlypreciseenoughto usefor an error analysis:Figure 6
showsthe error in the troposphericrange correctioncaused
representative(modeled) vertical profiles of the temperature
and water vapor, we have computed values of dimensionless by an error in (Zw'ha/Re)equal to one half of the variation;
(1-0.88)x 10- 3 = 0.06 x 10- 3.The erroris about7 cmat 7
quantity Zdw.For the temperature-heightdistribution,we used
elevation; much lessthan 1% of the correction itself. This is a
a constantlapse rate, and for the water vapor an exponential
bound on the error for higher elevationangles,as is seenfrom
height profile with saturatedconditionsat the surface.Results
Figure 6: Above 13 the error in this formalismis lessthan 1
are shown in Figure 4 for surfacetemperaturesof -30 , 0,
cm. The associatederror in range difference,subsequentlyre+ 30, and 40C. From the figure it is clear that Za is a
constant(here,becauseof the way we parameterizedthe water quired in a surveyingcomputation,will be lessthan this; the
mean value of the "formal" error, sinceit is highly correlated
vapor distribution, characterizedby the surfacetemperature)
over the differencinginterval, will subtractout.
for all elevationanglesexceeding15. For this elevationinterval 0.141 Za 0.16. These are the extreme values for all
6. BENDING EFFECT
conditions' The vapor pressureis so low at -30C that the
To this point, we have given the bending effect, of some
saturated vapor pressurecontributesnegligibly to the upper
concern
at very low elevationangles,only cursoryattention.
bound on Za. We showed[Black, 1978] that Za(E = 90) =
We
do
not
intend to treat this subjectin any depth.
0.166 for the dry effect alone for all temperatures.It seems
Hopfield
[1976b] commentsthat "... Although the georeasonable to conjecture that the above extreme values are
metrical
path
is greater than the slant range between two
extremesover most(all?)water vapor profiles.
points,
the
electromagnetic
rangeis lessalong the curvedpath
Since the dimensionlesscombination Za(ha/R)appears in
than along the slant-rangevector."She includesthe following
(16) (ha given by equation (18)) we have plotted this quantity
(Table 1) model for the bendingeffect,and she includesit in
in Figure 5. Between 7 and 90 elevation and for surface
temperaturesbetween -30C and + 40C the variation in this her 1976report [seealsoHopfield,1979].We havenot usedit
here;but, note the following.
quantity is
We have computedvaluesof the troposphericrange correc0.00088

<

< 0.0010

ej(a tedious computationshowsthat the upper bound is given

by 1/6 ha/R)or to threesignificantdigits
aha
-

0.1

We will tentativelyacceptthat (see(15) and (16))

d0

tion by numericallyintegrating(13) (neglectingbending)and

have compared the correction with the approximate values
givenby (20) and (21). The precisevaluesare givenin Table 2.
The differences(approximateminusprecise)are given in parenthesisabove the correspondingprecisevalues.As is clear,
the error in the approximateform partially compensatesfor
the bending effect.The compensationis not as good as we
would like (cf. Table 1). Nevertheless,it is of some benefit.
Clearly, the way to deal with the bendingeffectis to use the
Hopfield model to correctthe data and simultaneouslyuse a
table to removethe errors(at low elevationangles)in (21). For

BLACKAND EISNER'TROPOSPHERIC
EFFECTSON MICROWAVEDOPPLERDATA

TABLE 1. Effect of Bending

Elevation,deg.

As (Bending),m

t
5
10

3.11
0.19
0.03

15RMS
=126.4
cm
a(Ast,),m
SAT 30140

0.19
0.006
0.0009

DAY 47, 1980

BRUSSELS, BELGIUM
02:50

TIME

elevation angles below 5 there is another problem as the

elevation angle approacheszero: The dry part becomesincreasinglymore sensitiveto variations is temperature lapse
rate [Black, 1978; Maejirna, 1977].
7.

FITTING

2621

(MIN)

UT

MAX ELEVATION

= 35

Fig. 7. Tropospheric signal remaining in Doppler residuals after

three parametersare fitted out.

(along-trackand slant range at closestapproachplus fre-

PROCEDURE

quency bias) that are already being determined?We answer

We intend to use (20) and (21) to model the tropospheric this questionas follows.
rangeeffectAs. All of the geometricaldependenceis contained
Figure 7 showsthe after-fit residualsfrom a passof data.
in ldw.We will considerthe integral(the zenith correction)
(The troposphericmodelwasomittedin the processing.)
Three
parameterswere determined.We compare this figure with a
simulation based on (23) (the best fitting straight line was
jo
removed from the latter). Results are shown in Figure 8.
as a to-be-determinedquantity associatedwith each 15-min Clearly, there is a troposphericsignal in the data residuals.
passof satellite data. We will use the model, from this point The low elevation data is crucial in separating the tropospheric amplitude from the other data biases; the maximum
on, as
elevation of the pass should exceed(approximately) 20 (see
Figure 1) for the troposphericsignal to be fully representedin
(23) the Doppler data.

Asz
=10
-6In(Na
+Nw)
dh

(22)

cosEx2]1/2

8.

The instantaneousslant-rangeto the satelliteis written at the

ith time point

+ AsiTR+ AsiREL

(24)

HORIZONTALCORRELATIONLENGTH

To makethe fittingprocess
tractable,we ignorehorizontal
gradients(in the wet and dry refractivity) within the troposphere.In other words, we are assumingthat the troposphere
is horizontally stratified.This, in turn, imposesthe constraints
thai the surfacepressureand the precipitablewater vapor

wherein?sis the position of the satellite;?Nis the position of

i
pwv= pwvdh

the observei';
Asi"ELis the relativisticcorrectionto the slantrange;A&z"is the tropospheric
correctionto the slant-range

pw

(equation23); t time on the ground of the ith measurement;

wherePwis the densityof liquid wat6r, in the neighborhoodof
and c is the speedof light.
This question immediately arises: Is it possible,without the observing site be constant. Over the time dbration of a
ihat the troposphere
constraints,to determinea parameterin addition to the three satellitetransit,we are alsoassuming
does not change; in particular, that As remains constant.
Since we are utilizing a satellite with a (nominal) altitude of
TABLE

2.

1100 km, the transit duration is less than 20 min. Moreover,

Precise Values

sincethe thickness
of the troposphere
is smallcomparedwith

Elevation

5.2

__ 30

+ 30

+ 40

(- 35.7)

(- 17.4)

(- 18.4)

(- 29.7)

23.411

23.923

27.330

(-6.2)

(- 5.4)

19.941

22.777

(- 3.8)

(- 2.6)

(- 6.5)

17.908

18.354

20.963

22.991

(--3.4)

(-0.1)

(+0.9)

(-0.1)

12.524

12.867

14.698

16.105

6.4

(- ! 7.5)

7.0

(- 12.8)

19.473

10.4
11.8

14.8
17.3

30.014

(- 10.7)
24.989

(-2.2)

(+0.1)

(+ 1.0)

(+0.4)

11.091

11.402

13.024

14.268

(- 1.1)

(+0.2)

(+0.8)

(+0.6)

8.929

9.185

10.493

11.493

(-0.6)

(+ 0.2)

(+ 0.7)

(+ 0.5)

7.726

7.951

9.084

9.948

Errors in approximate form, in centimeters,otherwise in meters.

Values in parenthesisare in centimeters.Values not in parenthesisare
in meters.

the radius of the earth, the "constant" neighborhood is spatially limited. Figures 9 and 10 illustrate this limitation. The
neighborhoodfor the dry troposphereneed not extend more
than 300 km around the observingsite if we usedata no lower
than 7.5 in elevation. The correspondingneighborhoodfor

15

'

RANGE DIFFERENCES
INTERVALS

lO

AT 30 SEC.

MAX ELEVATION = 36
"

mE

rr
-

-5

TIME(MIN)

-10

-15

Fig. 8. Simulatedtroposphericrange-difference
error.

2622

BLACK AND EISNER' TROPOSPHERIC EFFECTS ON MICROWAVE DOPPLER DATA

'EFFECTIVE'

thesepoints, the mean values at each site were computed.

TOP OF

TROPOSPHERE

These mean valuesfor both data setsare shown in Figure 12.

Four stationswere representedin both data sets;Johannesburg, South Africa; Las Cruces,New Mexico; San JoseDos

TRoPoLsp

Campos,Brazil; and Herndon, Virginia. The data at these

four sitesagreewith the (usuallytrue) finding that the atmospherecontainsmore water vapor in the summerthan it does
in the winter. A tabulation of the resultsshownin Figure 12 is
givenin Table 3.
If we fit the troposphericparameter rather than using a
model per se,as indicatedabove,stationpositiondifferences
are a result. These differencesare shown in Figure 13 (1982
data) and indicate that there are important considerations
here in correctingfor troposphericeffectsin satellite-derived
surveyingcomputations.As Figure 13 showsmost of the troposphericbiaswill appearin the slant-rangecoordinatewhich
will in turn bias the longitude and station height.There are,
however,along-track(latitude)effectsas well.
10.

Fig. 9. Troposphericgeometryand neededcorrelationlength.

the wet term is 130 km in radius. A glanceat severalweather

maps indicatesthat the along-the-surface
pressurevariation
doesnot appreciablyaffectthe dry term.
9.

EXPERIMENTAL RESULTS

The Doppler data we usedwerecharacterized

as follows:(1)
obtained at a globally distributedset of sites;(2) two of the
five Transit satellites 30140 and 30480 were utilized; (3) one
2-day data span was obtained during winter in the northern

hemisphere,February 1980 (days48-49) and anotherduring

summer July 1982 (days 197-198); (4) each data span contained approximately 80 passes.Using each data span (and
the Hopfield model)we determinedan ephemerisfor the satellite. Subsequently,we used the ephemerisand processedthe
individual passestwo differentways:
1. The tropospheric model was again used (the troposphericparameterwas not estimated)and the remainingpass-

DIscussioN

We believethat Figures 11 and 12 verify that this simple

geometricaltechniqueworks and, moreover,is a convenient
and accurateone to use when surfacemeteorologicaldata is
not availableand perhapswhen it is available.
We have not as yet answeredan important question' Is
tropospheric
fitting really better than utilizinga modelper se?
An indication from this study is that it shouldbe. Wet-term
modelscannot be made very precise,and, moreover,it is certainly convenientto avoid dealingwith meteorological
data.
To show whether or not the fitted parametergiveshigher
accuracyin station position determination,we shouldrepeat
the experimentcarriedout by Jenkinset al. [1979]. At eachof
severalsiteswe should (1) acquire several(say, 6-10) setsof
passes,(2) executea multi-setsurveyat eachsite,and (3) compare the uncertaintyof the site coordinatesdeterminedwith
tropospheric
fitting and with the modelutilizingan a priori,

computedamplitude.
Jenkinset al. recognizedthat thereare problemswith fitting

a troposphericparameterif there are errors in the satellite

ephemeris.We werecarefulhere to assurethat the ephemeris,
the station coordinatesused in deriving the ephemeris,the
geopotentialmodel, and the associatedsoftwarewere all as
associatedbiases were absorbed in three least squaresdeter- internallyconsistentaswe couldmake them.
We had expectedto see some evidenceof the third-order
mined parameters;two componentsof the station position,
the "along-track" parameter, and the slant-rangeparameter.
These two quantities associatedata biases,from whatever
source,with the station position. The third parameter is the
frequencybias betweenthe satelliteand groundoscillator.
2. The model, equation(23), was included,and the amplitude,Asz,wasleastsquaresfitted togetherwith the other three
4OO
parameters(seeabove).No a priori constraintswereused.
Figure 11 showsthe individual Aszvalues,
3.5 IN LONGITUDE
AT 40 N
z 300

Asz= 10-6

(Na + N,) dh
z

o 200

for days 198-199. Since the dry term dominates(and lower

bounds) this integral and it is, in turn, proportional to the
surface pressureat the ground sites, we have plotted Asz
versusthe surfacepressure.The straightline is the dry term;
2.305Ps,with an associated
uncertaintyof 1-2%, about 3 cm.
Since the dry term provides a lower bound on the troposphericrange correctionand wet valuesabove 60 cm are unrealistic, there are four "wild" data points. After removing

10
ELEVATION

THRESHOLD

15
(DEG)

Fig. 10. Upper boundon necessary

troposphericcorrelation
distance.

BLACK AND EISNER: TROPOSPHERICEFFECTSON MICROWAVE DOPPLER DATA

2623

PHILIPPINES
(420)

3 m

--- o

eo

AUSTRALIA

o ..a

'

,--

(412)

ANCHORAGE (414)

--- -

OTTAWA

- -r

(128)

VIRGINIA (407)-- I

i.u

DRY
TERM
=fNddh
2.305
Ps

O BAD
(SEE
TEXT)
D = 198-199,

0.85

0.9

0.95

1982

74 PASSES

SAT. 30480

Ps- SURFACEPRESSURE
(ATMS)
Fig. 11. Fitted troposphericamplitude.

ionosphericeffectsin the residualsand results.Exceptpossibly

in the Hawaii residuals, we did not see these effects; however,
this subjectdeservesmore and extendedstudy.

We have emphasizedthe use of this techniquefor removing

troposphericeffectsfrom the Doppler data. The resultsmay
have value in another

context:

Since we seem to be determin-

ing realisticvaluesof the wet term which can be interpretedas

precipitablewater vapor [Goldfinger,1980], we possiblyhave

a new technique for intensively sampling (20 times/day) the

atmosphericwater vapor at the particular site. A surfacepressure measurementis required in the data reduction to remove
the dry term. A 2.4-cm error incurred in removing the dry
term will result in 0.4 cm error in the precipitablewater vapor.
This appearsto be practically achievable.
Before we could decide whether or not the technique is
usefuland complementsother (existing)techniques,further ex-

n-

2.5

--

"

mo2
o

DAYS 198 199,1982

SAT. 30480

DAYS 47-48, 1980 SAT 30140

1.5

0.82{

0.85

0.9

SURFACE

0.95

PRESSURE

IN ATMS

Fig. 12. Pass-averaged

valuesof the troposphericamplitude

2624

BLACKANDEISNER'TROPOSPHERIC
EFFECTS
ONMICROWAVE
DOPPLER
DATA
TABLE 3. Troposphere
Fitting Results
SAT 30140

SAT 30480

FittedTropo, Number FittedTropo Number

Latitude,
deg.

Altitude,
m

414 Anchorage,A1.
125 Calgary,Canada
021 Brussels,Belgium

+ 61
+51
+ 51

69
1272
116

128 Ottawa, Canada

320 Minnesota

+45
+45

86
300

641 Florence,Italy

+44

100

313
407

+44
+39

24.7
119

027 Japan

+ 39

83

330
113
413

+ 34
+32
+32

462
1206
1206

+30

245

+ 21

401

Station

Maine
Herndon, Va.
California
New Mexico
New Mexico

192 Austin, Texas

341

Hawaii

422 San Miguel, P.I.

+ 15

12

023

+13

38

Guam

420 SeychellesIsland

-4

593

424
008
408
105
405

American Samoa
Brazil
Brazil
South Africa
South Africa

-14
-23
- 23
-26
-26

12
613
613
1581
1581

412

Smithfield, Australia

- 34

34

019

Antarctica

- 78

38

D = 47-48

of

1980,m

Total

D = 198-199

Passes

2.4O_+O.4O].

2.45 + 0.15].
2.60 _+0.33].
2.43 +_0.12].
2.32 _+0.12]'
2.32 _+0.33].
2.27 _+0.18].
2.00 + 0.22].

5
6
6
8
7
5
5

2.40 + 0.30].
2.29 + 0.24].

6
4

2.48 __.0.07].

2.73 __.0.17'

2.65 + 0.23*

2.23 __.0.21'

2.21 + 0.31'

2.39 + 0.23

84

of

1982,m

Passes

2.34 + 0.16'
1.95 + 0.37*

11
8

2.47 _+0.06*

2.54 _+0.09*

2.10 + 0.08*

2.67 + 0.16'

2.55 _+0.29].

2.30 _+0.08].

1.98 _+0.05].
2.37 _+0.05].

7
8

2.33 + 0.23

75

*Summer.

'Winter.

perimentationis required.Certainly,comparisonof results essary,estimatedquantities,necessary

to absorborbit and stawith balloon(radiosonde)or radiometerdata is calledfor. The

tion position errors. On the other hand, as we reduce the

noiseon thein-orbitoscillators
or the third-orderionospheric elevationangleappreciablybelow 10,the errorsin the model
errorsare possiblylimitingthe currentaccuracyof our results. and, consequently,
the correlatederrors in the residuals,inWe just do not know.
crease.Our current feeling, based on the noise level of the
The reader is perhapsconfusedover the use of data ac- available data, is that the lowest elevation we can use is 5-7 .
quiredat low-elevationangles.The issueis far from clear.On Fittingthe curvesshownin Figure5 and subsequently
using

theonehand,lowelevationdataarenecessary
to separate
the

the resultsas part of the modelshouldslightlyimprovethe

modelat low elevationangles.To take advantageof this im-

refractivityamplitude(Asz,equation(22)) from the other nec-

Along-track
(m)-

vV
v V

-10

--5

Vv

,vWv

Slant-range (m)

-2--

68 PTS, SATELLITE 30480, DAYS 198-199, 1982

Fig. 13. Stationposition

changes
associated
withthefittingthetropospheric
parameter.

BLACKAND EISNER:TROPOSPHERIC
EFFECTS
ON MICROWAVEDOPPLERDATA

= 0, H = 0). The bracketed terms in (25) and (26) do not

superficiallyresembleeach other, but in fact they are quite
closetogether. We computed the fractional differencebetween
the two bracketsfor 0.85 <_Po <- 1 for elevationanglesabove

0.025
0.02

5, and for T = 233, 288; and 313 K. The brackets differ 2.5%
at 5 elevation

o.o

10

15

2625

20

25

30

and 0.5% or less for elevations

above 10 . The

largest fractional difference among these casesis shown in

Figure 14. Above 30, neither bracket differs more than 0.5%
from the cosecantof the elevationangle.
We are not the first to estimate a troposphericzenith parameter. Herring et al. [1981] mention that they include the
zenith troposphericdelay as a parameter and that "estimates
for the propagation delay through the atmosphereand ionospherein the zenith direction ranged from 6.7 ns (NRAO) to
8.2 ns (Onsala)."

Elevation (deg)

12.

A POSSIBLE APPLICATION

Fig. 14. Normalized differencesbetween the Marini-Murray and

Black-Hopfield models.

The motivation for solving this problem came out of our

successfulattempt to design a satellite system to monitor
strain in the earth'scrust [Westerfieldand Potocki, 1982]. So
that
the system would be all-weather and so that it could
provement would, however, require on-site temperature and
simultaneously
positiona large numberof sites,we electedto
humidity measurements.Expanding (13) to higher orders in
h/Re does not seemlike a very promisingapproach,nor does use microwavefrequencies.This choiceaggravatesthe tropotrying to use data below 5, becauseof the sensitivityto tem- sphericrefraction errors, particularly the water vapor probperature lapse rate. If the data quality shouldimprove clearly lem; effects at microwave frequenciesare about 100 times
we should shift to higher elevationcut-off thresholds,consis- largerthan they are at optical wavelengths.
We can deal with all the other error sourcesin the system:
tent with accuracyrequirements.
1. Ionosphericrefraction can be eliminatedby utilizing a
pair of frequenciesin the neighborhoodof 400-1200 MHz.
11. COMPARISON WITH OTHER MODELS
2. Satellite position errors can be effectivelyremoved by
One of the better known modelsof the troposphericrange
simultaneously observing the satellite at several terrestrial
effect is that of Marini and Murray [1973]. Their work is
points. The points are separated by small distances,small
designed to correct laser data; however, there are enough
when compared with the satellite altitude. This correlatesthe
similaritiesbetweenthe dry troposphericeffect at microwave
satelliteerrors in determiningthe relative position of neighfrequenciesand at optical frequenciesto make a usefulcomboring points.
parison:We ignore the effectsof water vapor; the water vapor
3. Errors associatedwith oscillator noise are suppressed
reftactivity at laser wavelengthsis 11.26 e/T, whereasat miby usingclosed-loopDoppler links.
crowaveit is(equation(5))3.73 x l0 s e/T 2.
These three techniques will push their associatederrors
For elevation anglesabove 5 and for all atmosphericcon- down to the centimeter level.
ditions and station heightswe can replace their equation (18)
Troposphericrefractionis anotherstory.The troposphereis
with
not dispersive;consequently,using multiple frequencieswill
not help. The refractivity at a point within the troposphere
dependson the atmosphericstate and temperature, pressure,
f(, H)
sin E + (B/A/(sin E + 0.01))
and water vapor density.
The difference between this form and theirs is at most 1.3
We believethat the approach describedin the body of this
x 10- 5AR.
report either solvesor potentially solvesthis problem.
The term in square bracketsis unity for E = 90. The term
in front, the amplitude, is (in meters)
Acknowledgments.We appreciatedthe effortsof Carolyn Leroy of

AR=
f('k)
A[ l+(B/A/(l+O.01))
] (25)

f(2)

2.305 Po

f(, H)

1 - 2.6 x 10-3 cos2 -- 3.1 x 10-4H

the DefenseMapping Agency/HydrographicTopographic Center in

making Doppler data available to us. We acknowledgethe contribution of George C. Weiffenbach.He insistedagainstour objections
that somethingcould be done about this problem. This work was
supportedby the U.S. Department of the Navy, Task Y related to

where Po is the surfacepressurein normal atmospheres,H is

the stationaltitude in kilometers,and 4 is the stationlatitude. Task S, under contract N00024-83-C-5501.
From [Black, 1978, p. 1828] we obtain a form comparable
REFERENCES
to (25).

Black, H. D., An easily implemented algorithm for the tropospheric

range correction,J. Geophy.Res.,83, 1825-1828, 1978.
Gardner, C. S., Effectsof horizontal refractivitygradientson the accuracy of laserrangingto satellites,Radio Sci.,11, 1037-1044, 1976.
Gardner, C. S., Correction of laser tracking data for the effects of
horizontalrefractivitygradients,Appl. Opt., 16, 2427-2432, 1977.
Goldfinger,A.D., Refraction of microwavesignalsby water vapor, J.
Geophys.Res.,85, 4904-4912, 1980.

AR
=2.343
po(TT4'12)[1(.1+cosE
.id/rsJ
x2]
-1/2(26)

Once again, the term in square bracketshas magnitude unity

for E = 90. The amplitude, the coefficientin front of the
bracket, increasesfrom 2.302 7o to 2.312 7o as the surface Guiraud, F. O., et al., A dual-channel microwave radiometer for
temperatureincreasesfrom -40C to +40C (cf. the Marinimeasurementof precipitable water vapor and liquid, IEEE Trans.
Geosci.Electron., GE-17, 129-136, 1979.
Murray value of 2.305 Po for b = 45,H = 0, and 2.311 Po for

2626

BLACKANDEISNER:TROPOSPHERIC
EFFECTS
ON MICROWAVE
DOPPLERDATA

Hargrave, P. J., and L. J. Shaw, Large-scaletroposphericirregularities

and their effecton radio astronomicalseeing,Mort. Not. R. Astron.
Soc., 182, 233-239, 1978.

Herring, T. A., et al., Geodesyby radio interferometry:Intercontinental distancedeterminations with sub-decimeterprecision,J. Geophys.Res.,86, 1647-1651, 1981.
Hopfield, H. S., Two quartic troposphericrefractivityprofile for correctingsatellitedata, J. Geophys.Res.,74, 4487-4499, 1969.
Hopfield, H. S., Troposphericeffecton electromagnetically
measured
range:Predictionfrom surfaceweatherdata, Radio Sci.,6, 357-367,

2115, pp. 363-376, 1979.

Pitts, D. E., et al., Temporal variations in atmosphericwater vapor
and aerosoloptical depth determinedby remote sensing,J. Appl.
Meterol., 16, 1312-1321, 1977.
Prabhakara, C., and G. Dalu, Passive remote sensingof the water
vapor in the troposphereand its meterologicalsignificance,
in Atmospheric Water Vapor, pp. 355-374, edited by A. Deepak et al.,
Academic, New York, 1980.

Hopfield, H. S., Troposphericeffectson signalsat very low elevation

angles,APL/JHU Rep. TG 1291, JohnsHopkins Univ. Appl. Phys.

Reber, E. E., and J. R. Swope,On the correlation of the total precipitable water in a vertical column and absolutehumidity at the surface, Rep. TR 0172 (2230-20)13, Aerospace Corp., Los Angeles,
Calif., 1972.
Reitan, C. H., Surface dew point and water vapor aloft, J. Appl.

Lab., Laurel, Md., 1976a.

Hopfield, H. S., Tropospheric effects on low-elevation angle signals:
Further studies,APL/JHU Rep. SDO 4588, Johns Hopkins Univ.
Appl. Phys.Lab., Laurel, Md., 1976b.
Hopfield, H. S., Improvementsin the troposphericrefraction correction for range measurement,Phil. Trans. R. Soc. London Ser. A,

Roosen, R. G., and R. J. Angione, Variations in atmosphericwater

vapor: Baseline results from Smithsonian observations,Publ.
Astron.Soc.Pacific,89, 814-822, 1977.
Smith, E. K., Jr., and S. Weintraub, The constantsin the equationfor
atmosphericrefraction index at radio frequencies,Proc. of the

1971.

294, 341-352, 1979.

Jenkins, R. E., B. D. Merritt, D. R. Messent, and J. R. Lucas, Refinement of positioning software (Doppler), in Proceedingsof the
SecondInternational GeodeticSymposiumon Satellite Doppler Positioning,vol. 1, pp. 233-266, University of Texas,Austin, 1979.
Maejima, I., Global pattern of temperaturelapse rate in the lower

tropospherewith special referenceto the altitude of snow line,

Geograph.Rep. Tokyo Metro. Univ., no. 12, 1977.
Marini, J. W., and C. W. Murray, Correction of laser range tracking
data for atmospheric refraction at elevations above 10 degrees,
NASA/GSFC X-591-73-351, 1973.
Moran, J. M., and B. R. Rosen, The estimation of the propagation
delay through the tropospherefrom microwaveradiometerdata, in
Radio Interferometry Techniquesfor Geodesy,NASA Conf. Publ.

Meterol., 2, 776-779, 1963.

I.R.E., 41, 1635-1637, 1953.

Thayer, G. D., An improvedequationfor the radio refractiveindexof

air, Radio Sci, 9, 803-807, 1974.
Westerfield, E. E., and K. A. Potocki, The fault zone monitoring
system,paper presentedat Proceedingsof the Third International
GeodeticSymposiumon Satellite Doppler Positioning,Las Cruces,
N.M., February 1982.

H. D. Black and A. Eisner,Applied PhysicsLaboratory,The Johns

Hopkins University,JohnsHopkins Road, Laurel, MD 20707.
(ReceivedMay 16, 1983;
revised December 8, 1983;

acceptedDecember 12, 1983.)