Quantify IOP quality derived from QAA Zhong Ping

Quantify IOP quality derived from QAA Zhong. Ping Lee 1, Robert Arnone 2 1 Northern 2 Naval Gulf Institute, Mississippi State University, Stennis Space Center, MS 39529; zplee@ngi. msstate. edu Research Laboratory, Stennis Space Center, MS 39529

Q: How big is the error (or confidence) of inverted IOPs at each pixel? “Average error of an algorithm is insufficient to describe the error of specific locations. ” -- Boss et al We should not expect the same error bars for the three pixels.

The Quasi-analytical algorithm (QAA) Rrs( ) The three stating points: Y (Lee et al. 2002)

If Rrs is error free, and the relationship between Rrs and {bb, a} is accurate, quality of inverted a&bb by QAA depends on the estimation of a(λ 0) and Y.

A “Hydrolight” data set a 440 distribution a 550 distribution bbp 440 distribution Y distribution

QAA inverted a 440&bbp 440 vs known a 440&bbp 440 a 440 inv [m-1] bbp 440 inv [m-1] -- when a 550 and Y are known exactly. a 440 known [m-1] bbp 440 known [m-1] Indication: the formulation and scheme can get a closure!

QAA inverted a 440 vs known a 440 -- impact of imprecise a 550; Y is known exactly a 550 is 10% higher a 440 inv [m-1] a 550 is 10% lower a 440 known [m-1] Indication: nearly the same impact to all a 440 values.

QAA inverted a 440 vs known a 440 -- impact of imprecise Y; a 550 is known exactly Y is 0. 4 higher a 440 inv [m-1] Y is 0. 4 lower a 440 known [m-1] Indication: nearly the same impact to all a 440 values.

QAA inverted a 440 vs known a 440 -- compounding effects of imprecise Y and imprecise a 550 Y is 0. 4 higher and a 550 is 10% higher a 440 inv [m-1] Y is known; a 550 is 10% higher a 440 known [m-1] Indication: more errors!

QAA inverted a 440 vs known a 440 -- compensation between imprecise Y and imprecise a 550 Y is 0. 4 lower; a 550 is 10% higher a 440 inv [m-1] Both Y and a 550 known a 440 known [m-1] Indication: nearly compensating each other.

Q: 1. How big is the a 550 error for different a 550? 2. How this error propagates to other IOPs? εa 550 Error at a 550 (εa 550) vs estimated a 550 Global ocean a 550 inv [m-1]

Error at a 440 vs compound error of a 550 and Y (± 0. 2, ± 0. 4) δa 440 Error at a 440 εa 440 Error at a 550 (εa 550) δ: percentage error at 84 th percentile or likelihood range

Error at a 410 vs compound error of a 550 and Y (± 0. 2, ± 0. 4) δa 440 Error at a 410 εa 440 Error at a 550 (εa 550) δ: percentage error at 84 th percentile

Data flow to quantify quality of inverted IOP : Rrs aλ 0 εIOP & δIOP Note: No Rrs error considered yet; but can be added when quality measure of Rrs is known.

Conclusion: 1. Qualities of IOPs derived by QAA, in addition to quality of Rrs, rely on the estimation of a(λ 0) and Y. 2. The qualities at each pixel, measured by projected average error and projected likelihood range, can be quantified by evaluating the error propagation in the QAA process. 3. Higher qualities for IOPs of oceanic waters, as both a(λ 0) and Rrs possess higher reliability.

Acknowledgement: The supported from NASA Ocean Biology and Biogeochemistry Program is greatly appreciated.

a 550 [m-1] aw values dominate the longer wavelengths, so a(λ 0) can be estimated reasonably well. a 440 [m-1]