1 satellite geodesy (ge-2112) processing of observations e. schrama

22
1 Satellite geodesy (ge- 2112) Processing of observations E. Schrama

Upload: tracy-pope

Post on 18-Jan-2018

228 views

Category:

Documents


0 download

DESCRIPTION

3 Preprocessing of observations Oftentimes raw observations are NOT suitable for direct application in parameter estimation algorithms Raw observations typically contain non Gaussian errors like outliers greater than 3 sigma Often there are very good reasons to inspect and clean up the data before you put it into an estimation procedure This topic is much depending on the observation technique, we will just show some well known examples.

TRANSCRIPT

Page 1: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

1

Satellite geodesy (ge-2112)

Processing of observationsE. Schrama

Page 2: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

2

Contents• Preprocessing of observations

– example 1: dual frequency ionospheric– example 2: tropospheric range delay– example 3: normal point compression

• Classification of problems, function models– Parameters in function model – Parameter estimation procedure

• Implementation– Precise Orbit Determination– Variational equations– Organisation parameter estimation

Page 3: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

3

Preprocessing of observations

• Oftentimes raw observations are NOT suitable for direct application in parameter estimation algorithms

• Raw observations typically contain non Gaussian errors like outliers greater than 3 sigma

• Often there are very good reasons to inspect and clean up the data before you put it into an estimation procedure

• This topic is much depending on the observation technique, we will just show some well known examples.

Page 4: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

4

Preprocessing example 1The problem is: how do you eliminate the ionospheric delay from dual frequency range data?

22

21

212

22

1

22

212

10

222

211

)( and

)(

)(

ffN

ffffr

fNr

fNr

o

o

Task: compute the factors for GPS L1/L2

Page 5: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

5

Preprocessing example 2

The air pressure is 1000 mbar, the air temperature is 20 degrees centigrade, the relative humidity is 50%, what is the dry+wet tropospheric delay of a radio signal as a function of the elevation angle for a station at MSL and 50 degrees latitude. The answer is:

• Use the Hopfield model (see Seeber p 45 - 49) to calculate the refractive index

• Use the integral over (n-1) ds to compute the path delay

• For the latter integral various mapping functions exist

Page 6: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

6

Dry tropospheric delay example

0 10 20 30 40 50 60 70 80 900

5

10

15

20

25

Elevation angle

Dry

trop

osph

eric

del

ay

Modified Hopfield model This result is entirely depending on the air pressure P, 1% air pressure change (=10 mbar) gives 1% range change. Since air pressure is usually known to within a millibar the dry tropospheric delay error is small. For low elevation angles the delay error increases due to the mapping function uncertainties. Hence elevation cut-off angles are used (typically 10 degrees).

Page 7: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

7

Wet tropospheric delay example

0 10 20 30 40 50 60 70 80 900

0.2

0.4

0.6

0.8

1

1.2

1.4

Elevation angle

Wet

trop

osph

eric

del

ay

Modified Hopfield model The wet tropospheric range depends on the relative humidity which varies more rapidly in time and place compared to air pressure. Variations of the order of 50% are possible. As a result the vertical path delay can vary between 5 and 30 cm. The alternative is the use of a multifrequency radiometer system, see Seeber p 49.

Page 8: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

8

Normal point compression

0 200 400 600 800 1000 1200-0.5

0

0.5

1

1.5

2

2.5

3x 10

4

Method: Use a compression technique (splines, polynomials, etc) that fit the crosses. Evaluation of the model results in the compression points (the circles). This procedure filters out the noise. Horizontal: time, vertical: range

Case: red crosses is SLR data, there are too many of them and there are clear blunders that we don’t accept in the parameter estimation procedure.

Page 9: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

9

Classification of problems• Terminology:

– Here, a problem refers to an interesting case to study.

• Problems in satellite geodesy:– Type of problem

• does it contain orbit parameters?• does it contain gravity field parameters?• does it contain any other geophysical parameters?

– How do you organize parameter estimation?• it is a batch or a sequential least squares problem?• can you solve it from one observation set or are more sets involved?• Is preprocessing of observations involved or is it in the problem?

Page 10: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

10

Function model (1)

• The function model aims to relate observations and parameters to another

• The unknowns are gathered in vector • The observations are in vector • Usually we begin to approximate reality by

a priori estimates and

X

0X

Y

0Y

Page 11: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

11

Function model (2)

Xe

Ye

Ze

BS

Ri

Rj

Rij

jiij RRR

)(XvY

)( 000 XvY

AxvyXXx

YYy

0

0

Page 12: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

12

Function model (3): Examples

• The overdetermined GPS navigation solution for one receiver

• VLBI observations of phase delay• Two GPS receivers: double difference processing• SLR network: station, orbit parameters, earth

rotation parameters • DORIS with orbit and gravity field improvement• Spaceborn GPS receiver on a LEO

Page 13: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

13

Parameter estimation(1): Least squares

rxNyQAxAQA

xAyQA

xAyQAxxAyQy

xAyQxAy

vQvvxAy

vT

vT

vT

vTT

vT

vT

vT

equations) (normal

0

min

min

min : isour task ,

11

1

zero put to topossible

11

1

1

Page 14: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

14

Parameter estimation (2)

• The way the A matrix is computed completely depends on the type of observations and parameters in your problem.

• We will distinguish between problems that contain orbit parameters and problems that do not.

• Our first task will always be to model an orbit in the best possible way given the existing situation

• This task is called orbit prediction

minimum

)(

1

0

vQv

XXA

vT

Page 15: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

15

Implementation

• From our function model we conclude that:– it is by definition a non linear problem– it depends on a priori information – it almost always depends on orbit dynamics – orbit predictions are used to correct the raw

observations and to set-up the design matrix– the orbit prediction model is not necessarily

accurate the first time you apply it

Page 16: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

16

Orbit prediction (1)Part 2 of these lectures mentioned that one needs to integrate the equations of motion. Suitable numerical techniques are used to treat differential equations of the following type:

)( ),(),( 00 tyytgtyFy

There are numerical procedures like the Runge-Kutta single step integrator and Adams-Moulton-Bashforth multi step integrator that allow the state vector y0 to be propagated from y0 till yn. In this case a state vector at index j coincides with the time index t0+(j-1)*h where h is the integrator step size.

Page 17: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

17

Orbit prediction (2)• The orbit prediction problem is entirely driven by the choice

of the initial state vector y0, the definition of F(y,t) and g(t).• The basic question is of course, where does this information

come from?• F(y,t) and g(t) fully depend on the realism of your

mathematical model and its ability to describe reality• However, knowledge of the initial state vector should follow

from 1) earlier computations or 2) launch insertion parameters• The conclusion is that it is desirable to estimate initial state

parameters from observations to the satellite.

Page 18: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

18

Initial state vector estimation in precise orbit determination

Task: determine the size, orientation and position of the arrow, it determines whether you hit the bull’s eye

Page 19: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

19

Summary Precise Orbit Determination (POD)

• In reality orbit parameters are estimated from observations like range, Doppler or camera to the satellite

• Orbit prediction scheme– Numerically stable schemes are used– Choice initial state vector– Definition satellite acceleration model

• Variational scheme– Define parameters that need to be adjusted using least squares– Iterative improvement of these parameters– Use is made of so-called variational equations

Page 20: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

20

Parameters in POD

• Station coordinates • Station related parameters (clock, biases)• Initial state vector elements of satellite orbits• Parameters in acceleration models satellite• Other satellite related parameters (clock, biases, etc)• Signal delay related parameters• Earth rotation related parameters• Gravity field related parameters

Page 21: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

21

Variational equations

parameter nal variatio:1a : sin0a : cos

)(cos,

,1

,,

2

mm

PY

YCra

aGM

rGMV

FX

VX

FVX

nmnma

nmanma

n

amn

e

e

i

i

i

i

i

Example : initial state vector component, terms in force model etc

Page 22: 1 Satellite geodesy (ge-2112) Processing of observations E. Schrama

22

Organisation parameter estimation

• For large scale batch problems: – separation of arc -- and common parameters– combination of normal matrices and right hand sides– choice of optimal weight factors for combination– example: development of earth models like EGM96

• Sequential problems– apart from the adjustment procedure there is a state vector transition

mechanism– During transition state vector and variance matrix are advanced to the

next time step (normally with a Kalman filter)– Example: JPL’s GPS data processing method