CLEAR VISION ON TURBID WATER: THE NAIVASHA LAKE

Transcription

1 CLEAR VISION ON TURBID WATER: THE NAIVASHA LAKE M.S. Salama June 13-14, 2013

2 Outlines Study Area The problem Objectives Field data Hydro optical models Empirical: CDOM, Chl-a and KD Cal/Val of the empirical models Semi analytical including NAP Errors Depth profile of Kd: a new method Preliminary conclusions 2

3 Study area 3

4 Study area Home for over 400 bird species Wild animals Tourist area Horticulture Fishing Fresh water! 1mil$ a day!! 4

5 Problem The Naivasha Lake is becoming an over-enriched muddy pool, which will shortly become unusable. It is only a matter of time before the lake becomes toxic (Harper, 2004) 5

6 Problem as experienced! On march, 2010 after a heavy rains 6

7 Objectives The main objectives here are: Accounting for the NAP and phycocyanin absorptions and derive the spectral slops; Investigate semi empirical models for CDOM, Chl-a and Kd; Investigate the errors of IOP and their sources; Suggest new forward model to get Kd on different depths; 7

8 Field data Optical and biophysical measurements IOPs Max Min Average Stdev a Chl_a (m -1 ) a CDOM (m -1 ) a t (m -1 ) a NAP (m -1 ) SPM(mg/l) S_CDOM(sr -1 ) S_NAP(sr -1 )

9 Absorption coefficients (m-1) Results of field data Total absorptions Particulate absorptions CDOM absorptions 0 Gilgil inlet Hippo pool Crater Lake Flora industry Lake centre Forest side Lake Naivasha Chart of absorptions of water constituents in lake Naivasha 9

10 acdom (m-1) Rrs (sr-1) Results of field data 0.03 lake Naivasha hippo pool flower industry Crater lake gilgil inlet lake center wavelength (nm) y = e-0.012x R² = (N = 93) Wavelength (nm) 10

11 Hydro optical models Empirical algorithms Use statistical relationships to link observed radiometric quantities to measured IOP s. Semi analytical algorithms Use approximations of radiative transfer and empirical relationships to provide invertible linkages between the AOPs and the IOPs Capability to retrieve several parameters simultaneously because they model the optical properties (maybe!) 11

12 a CDOM (440)/a t (440) Results of the empirical model: CDOM [Field] y = x R² = 0.81 (N = 93) RMSE = 0.05 Crater lake [MERIS] y = x R² = 0.73 (N=23) RMSE = 0.05 Crater lake Main lake 0.15 Main lake a CDOM (440)/a t (440) Comparison between retrieved and in situ measured a CDOM /a t at 412 nm The model of Belanger et al. (2008) 12

13 Results of the empirical model: Chl-a 3B Algorithm by (Gitelson et. al 2008) SCI Algorithm by (Shen et al 2010) & FLH Algorithm(Gower1994,1995) Model Calibration Summary Model n r 2 RMSE 3B SCI FLH Model Trios-Validation Performance Model n r 2 RMSE 3B SCI FLH

14 Results of the empirical model: Kd Austin and Petzold (1981) R 2 =0.98 RMSE=0.388 R 2 =0.78 RMSE=

15 Cal/Val GeoCalVal is a novel model that provides the: 1- optimal subdivision of matchup data set into Cal and Val sets; 2- accuracy of calibration coefficients 3- probability distribution of the validation errors. derived probability distributions (PDs) of calibration coefficients (d, e) and validation errors (f) for Chla absorption per unit concentration Determination coefficient, R 2, between measured and observed values of Chla absorption coefficient from many Cal/Val setups. Lightgrey coloured points represent the optimal Cal/Val pairs. Salama et al., 2012, Biogeoscinces, 9,

16 16 One of the most used semi analytical model is the GSM model which was developed by Gordon et al., (1988), improved by Garber-Siegl and Maritorena abbreviated as GSM and recently modified by Salama, et al (2009). Limitations Could not separate CDOM and NAP absorption Is it good for Kd at different depths? i b b i i w rs a b b g n t R ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( adg ph W a a a a Hydro optical models

17 Including NAP and PC NAP phycocyanin Doxaran et al., (2009) nm Salama eta la in preparation 10 nm 17

18 Results Field spectra are inverted to IOP and compared to the measured ones 18

19 Results Matchup MERIS spectra are inverted to IOP and compared to the measured ones 19

20 How much adds up? Errors The total uncertainty is the sum of three error component 2 t 2 atm 2 noise 2 inv Error % model noise atmosphere Biomass CDOM SPM Salama and Stein,2009. Applied Optics, 48,26,

21 Errors: AC The challenge is that the reliability varies with how dirty the air is and how turbid the water is? Salama and Shen (ECSS, 2010) Turbidity is the challenge! 21

22 Errors: model Water turbidity reduces observed radiometric variability: radiometric variability is higher for clear water than for turbid water Water turbidity complicates the correction and retrievals Salama and Su 2011, IEEE 49,

23 Errors: model Inherent error of model parameterization setup Salama et al., 2011, Optics Express, 19, 18 Salama and Shen, 2010, Optics Express,18,

24 Relative contribution [-] Spatial mismatch original 3x3 5x5 7x7 9x9 Filter window Velde, Salama et al., 2012 Hydrometeorology, Bias Spatial scale mismatch Residual Reduced resolution higher errors!!! 24

25 How can we derive Kd(z)? Old but gold!, the Duntley(1942) two steams model: deds dedd keds; s' Eds Edd Eu dz dz deu seds Edd Eu dz Solving and accounting for singularity gives E d ( z) Eds (0) exp( kz) Edd (0) exp( mz) ( s' rsd ) J1( z) Eds (0) this does not describe a purely exponential attenuation of the flux with depth, since two extinction coefficients are involved, k and m. 25

26 Results Versus Hydrolight at different depths 26

27 Results Versus in-situ measured Ed(z= 50 cm) and Ed(z=100 cm) K dd (440) (m -1 ) K dd (490) (m -1 ) K dd (550) (m -1 ) K dd (670) (m -1 ) R Intercept Slope RMSE (m - 1 ) MRE (%)

28 Preliminary conclusions Empirical models are ok but limited; we will apply GeoCalVal; Semi analytical model with NAP and spectral slopes is acceptable; We can derive Kd depth profile; and improve on the forward model. To modify choose 'Insert' then 'Header and footer' 28

29 Thank You!! 29