PSI validation results and open issues

Frascati, 2 December 2009

PSI validation results and open issues Michele Crosetto, Oriol Monserrat, Jorge De Arriba, Rubén Iglesias Institute of Geomatics Castelldefels

1

Introduction ‰ What is PSI validation about? ƒ

Not a simple comparison against ground truth, e.g. thermometer: PSI products are “complex” and not easy-to-understand

ƒ

It is a learning process, which involves: ƒ

Product characterization, sometimes “discovering” “new” pros and cons that are not well understood or documented

ƒ

Highlighting pros & cons Î Communicate to end users (a lot to be done)

ƒ

Increasing product acceptability (a lot to be done)

‰ Remark: often multi-purpose PSI products are assumed: some validation

outcomes can be irrelevant for some specific application (I do not care about mm!). But usually are relevant for many applications. ‰ Remark: we do (mainly) “technical validation”; the “user validation” will follow (especially at the beginning of the learning process the user can be “still to be found”) 2

A learning process: examples ‰ PSIC4: ƒ

blind test

ƒ

“big deformation” (C-band viewpoint!) Î aliasing

3

PSIC4 (earth.esa.int/psic4)

Is this still valid? What are the performances of new unwrapping techniques?

4

A learning process: examples ‰ PSIC4: ƒ

blind test (detection not working)

ƒ

“big defo” (C-band viewpoint!) Î aliasing

‰ Jubilee Line: ƒ

Very focused target, both spatially and temporally

ƒ

“Non linear” deformation

5

JLE: deformation of Treasury building

- Out of standard PSI capabilities in 2004 (a posteriori statement!) - Result relevant to key civil engineering applications (e.g. tunnels) - Is this still valid? http://www.terrafirma.eu.com/JLE_intercomparison.htm

6

Another example from tunnel monitoring 20

Deformation [mm]

10

0

-10

-20

-30

-40 14/01/2004

01/08/2004

17/02/2005

05/09/2005

24/03/2006

10/10/2006

28/04/2007

Date Reference

PS estimation

7

TF Validation Project: inter-comparison ‰ Comparison of the outputs from the different PSI processing chains ƒ comparison in the radar space (not affected by geocoding errors) ƒ considering all combinations (4 results, 6 possible pairs) ‰ Three test sites, four teams: ƒ number of datasets:

12

ƒ number of PS:

~ 700´000

• For each PS: ƒ number of PS pairs:

> 39 samples (ASAR) > 83 samples (ERS) ~ 370´000

“Average PS” vs. “the PS champion”

www.terrafirma.eu.com/Terrafirma_validation.htm

8

Inter-comparison of velocities

[mm/yr] Max. Min. Mean Stdev. # PS TA-TB 4.70 -5.20 -0.07 0.61 9683 TA-TC -8.22 0.51 that concern 6257 These values4.40 are representative of-0.04 the PSI studies areas with similar to0.00 those of the0.56 three test5586 sites. TA-TD 4.29 characteristics -5.06 TB-TC 12.88 -8.62 13954 Key: avoid extrapolation to different 0.03 application 0.66 contexts, e.g. mining TB-TD 8.73 -8.62 0.10 0.59 11341 TC-TD 9.06 -8.73 0.06 0.45 17989

Statistics for the velocity differences computed over all the common PS for ERS Alkmaar dataset.

www.terrafirma.eu.com/Terrafirma_validation.htm

9

Inter-comp. of time series (key product) 15

Team A

PS(592, 20)

Team B

Deformation [mm]

10

5

0 15400

15900

16400

16900

17400

17900

18400

-5

-10

Correlation = 0.301 -15 Julian Days

www.terrafirma.eu.com/Terrafirma_validation.htm

10

VHR PSI

11

VHR: Cosmo Skymed

12

13

VHR PSI: time series ‰ X-band: remarkable quality improvement wrt C-band

14

VHR PSI: thermal dilation

1) worth to properly validate 2) Important to document the (remaining) limitations 15