Algorithm Failure & Mitigation

By bryan Date 2008-08-26 18:48

To algorithm providers:

As discussed today at our telecon, would you please list the failure conditions and mitigation strategies employed within your algorithms, when applied to satellite data. Specifically:

1) input Rrs/nLw ranges for which algorithm is allowed to attempt solution, and which bands are tested?

2) any intermediate tests that are performed within the inversion, to fail the algorithm or adjust the retrieval.

3) output IOP product ranges (or other tests) for which you consider the retrieval valid, which products do you test, and does failure in one constitute failure in all.

4) considering the above, and the fact that we wish to produce global, spatially- and temporally-averaged IOP products, is there anything you might change in the current failure/mitigation strategy.

By EmmanuelBoss Date 2008-08-26 20:40

For the Boss/Roesler algorithm in the IOCCG report we:
1. Did not filter the Rrs/nLw ranges. This should be done with global data.

2. For each Rrs/nLw spectra inverted we generated we generated a whole number of solutions. We required that IOPs (a_ph, a_cdm and bb_p) be above a minimal treshold (e.g. -0.005 for absorption, similar to that based on IOP measurements) and that the reconstructed spectra (based on the inversions) be within a given relative treshold from the inverted Rrs in ALL wavelengths used (varied from 5 to 20%). The solutions which satisfied these criteria were then lumped and their median was considered the most likely solution with the 16 and 84 percentiles providing the uncertainties (equivalent to 1std for normally distributed data). In certain cases no solution was within these criteria and thus no solution was reported.

Cheers,
Emmanuel

By jilly Date 2008-09-03 22:39

For the niwa algorithm:

1) Rrs(any band) < 0 is set to 0, but the algorithm attempts a solution regardless of whether any Rrs values are 0.

2) Only returns a result if
a: a(490)/a(510) is within prescribed range [1.32 : 1.58], and
b: upper and lower estimates of b(490) are both > 0, and
c: retrieved a(490) and a(510) both > 0.001.

Retrieval is adjusted if a,b go out of range of the look-up tables used for calculating F(theta,theta0,dphi,lambda,a(lambda),b(lambda)). In this case, values are reset to their lower prescribed limits.

3) Retrievals are accepted if conditions a-c above are met. Failure of any one condition causes the algorithm to return a,b=0 with a fail flag.

We consider the algorithm valid for the ranges : a<= 20 m-1; bb(490)<=1.05 m-1; sensor theta <= 45; sun theta <= 75.

4) It may be useful to return epsilon_a as an algorithm output. If eps_a goes out of the prescribed range, we could set the failure flag, but use less strict limits for the final 'fail: set to zero' test. The current eps_a range is based on data in the North Atlantic, so probably not truly global. We could also include the option of setting upper and lower limits on eps_a, as well as setting eps_b (=bb(490)/bb(510)) - currently eps_b is set to 1.022.

By zplee Date 2008-09-04 16:04

In case you don't know, there will be no pickup for people arriving before 4th and 5th. Here is some "mitigation" information I got from Trudy:

The best way is to write to ilCiocco at: sales@ciocco.it. Aside
from that, I do believe that the hotel will go to the train station and
pick up guests. Following is the information on Trains:

*TRAIN*

The web site for train information in Italy is www.trenitalia.com
<http://www.trenitalia.com/>. Catch the train from either Pisa
Areoporto, or Pisa Centrale. Pisa Centrale is a 5 minute (3km) cab ride
from the airport, so it is often easier to do this (as trains are more
regular from the Pisa Centrale). There is also another station after
Pisa Centrale called Pisa S. Rossore. The train route you need is "Pisa
- (Lucca) - Barga-Gallicano".

Barga-Gallicano is the station most convenient to Il Ciocco. When
searching for this station using trenitalia
<http://www.trenitalia.com/>, type it exactly as written here.
Barga-Gallicano is about 5km from Il Ciocco. When you reach the station,
you can call the hotel (0583 7191) from the bar next to the station and
ask them to send a taxi for you. A taxi to the hotel is approximately
EUR10.00 - EUR20.00.

The total journey time varies but is approximately 1 hour 30 minutes.
After Lucca, the train is a small slow valley train. To assist you, the
station before Barga-Gallicano is Fornaci di Barga. A word of warning
regarding your ticket: once purchased, you must stamp it using the small
yellow machines which are located on the platforms. This puts a date
stamp on your ticket and validates it. You can be fined if your ticket
is not validated!