High CO2 world

23
Towards Remotely-Sensed Estimation of Alkalinity in Australian Coastal Waters Kimberlee Baldry, Nick Hardman-Mountford, Jim Greenwood, Francois Dufois, Bronte Tillbrook Presented by Kimberlee Baldry BSc Chemistry and Mathematics and Statistics (UWA) CSIRO Vacation Scholar (2014-2016) [email protected] [email protected]

Transcript of High CO2 world

Page 1: High CO2 world

Towards Remotely-Sensed Estimation of Alkalinity in Australian Coastal Waters

Kimberlee Baldry, Nick Hardman-Mountford, Jim Greenwood, Francois Dufois, Bronte Tillbrook

Presented by Kimberlee Baldry BSc Chemistry and Mathematics and Statistics (UWA)

CSIRO Vacation Scholar (2014-2016)[email protected]

[email protected]

Page 2: High CO2 world

Motivation

Page 3: High CO2 world

Data: IMOS National Reference Stations

IMOS Ocean Data Portal: https://imos.aodn.org.au

Chl-a

DIC/TCO2

NO3

Temp Sal

ALK

Phytoplankton

O2

Page 4: High CO2 world

Background• Look at what effects TA -> Build model

Sal Temp Chl-aIntra-watermass mixing

Freshwater inputs/outputs

Inter-watermass mixing

Nutrient Changes

Primary Productivity

AUSTRALIAN WATERSOpen Ocean Model

Coastal Models

?Other processes that affect TA

Lee et al. (2006)SSS + SST + SSS^2 + SST^2 +c

Page 5: High CO2 world

Methods: ModelsAim Assess predictions of TA from its proxy variables in Australian coastal waters

Multiple Linear Regression (MLR) Analysis(1) TA = aSal + d(2) TA = aSal + bTemp + d(3) TA = aSal + bTemp + cChl-a + d(4) TA = aSal + bTemp + clog[Chl-a] + d

Coastal ModelsAlgorithms calculated from ALL NRS dataRegional ModelsAlgorithms calculated from INDIVIDUAL NRS data

Page 6: High CO2 world

Methods: Statistical AnalysisMethod Pros Cons

Kolmogorov–Smirnov (K-S) Tests

Method of comparing observations to models

Binary95% Confidence Level

Residual Standard Error (RSE)

Model error in standard units

Doesn’t consider number of variables in model, or number of observations

Akaike Information Criterion (AIC)

Combined measure of complexity, and RSE

Sensitive to number of observationsHard to compare differences

Relative Probability of Minimising Information Loss

IntuitiveIn terms of probabilitiesDoesn’t rely on “eyeballing”

Very sensitive

Page 7: High CO2 world

Results: Lee et al. (2006) Open Ocean Model

Page 8: High CO2 world

NRS Model 1Sal

Model 2Sal-Temp

Model 3Sal-Temp-Chl-a

Model 4Sal-Temp-log(Chl-a)

Open OceanLee et al.

(2006)Regional Coastal Regional Coastal Regional Coastal Regional Coastal

1.Darwin ü û ü û ü û ü û û2.Esperance ü ü ü ü ü ü ü ü û3.Kangaroo Island ü û ü û ü û ü û ü4.Maria Island ü û ü ü ü ü ü ü û5.Ningaloo ü û ü ü ü ü ü ü ü6.North Stradbroke Island û ü ü û ü û ü û û7.Port Hacking Bay ü û ü ü ü ü ü ü û8.Rottnest Island ü ü ü û ü û ü û û9.Yongala û û ü û ü û û û û

Results: K-S Tests- ü Drawn from the same distribution- û Not drawn from the same distribution

Page 9: High CO2 world

Results: 95% Confidence Error* - Model Error

Model

* 1.95 x RSE

Page 10: High CO2 world

Results: AIC- Combined measure of goodness of fit (RSE) and complexity (number of parameters) of model

Model

Page 11: High CO2 world

Results: Minimum Model- Relative Probability of Minimising Information Loss- Compared in terms of probabilities, rather than just “eyeballing”

Model

Page 12: High CO2 world

Implications: Modelling TA Sal-TempSal-Temp-log(Chl-a)

Sal

Page 13: High CO2 world

Implications: The Bigger Picture

Model 2 vs Model 4

Model 1 vs Model 2

% difference

% difference

Page 14: High CO2 world

Implications: Modelling pHRegional

Sal-Temp-log(Chl-a)

Coastal

Page 15: High CO2 world

Conclusions• Model 4 -> Minimum model• Chl-a influence generally small but may be important in

some areas• Regional models are better than General Coastal or Open

Ocean Models

Further Work • Application to ship data -> Spatially continuous model• Investigate robustness of Earth Observation application• Temporal robustness of algorithm• Application to Australian-wide carbonate models

Page 16: High CO2 world

Thankyou and Acknowledgements

Page 17: High CO2 world

MLR Results: Model 1NRS Correlation

Coefficient Slope Intercept n RSE AIC

General 0.94 53.69 420.98 1213 10.50 9150.8

Darwin 0.96 54.58 407.94 60 9.49 444.21

Esperance 0.84 64.83 27.87 48 6.02 312.53

Kangaroo Island

0.84 46.25 696.8 110 5.55 693.22

Maria Island

0.85 46.61 678.44 230 3.76 1266.02

Ningaloo 0.62 36.05 1025.43 29 5.82 188.41

North Stradbroke Island

0.94 58.83 236.1 168 4.5 968.27

Port Hacking Bay

0.94 61.67 138.99 194 2.83 957.42

Rottnest Island

0.93 58.44 252.78 167 4.68 993.41

Yongala 0.97 50.84 505.24 207 8.64 1484.12

Page 18: High CO2 world

NRS Correlation Coefficient

Intercept SAL SST n RSE AIC

General 0.95 620.14 48.78 -1.28 826 8.87 5955.05

Darwin 0.96 543.2 51.32 -0.91 39 9.15 288.2

Esperance 0.87 51.17 64.75 -1.15 36 5.52 230.09

Kangaroo Island

0.86 732 45.31 -0.12 61 5.54 386.96

Maria Island

0.9 486.92 52.21 -0.45 142 3.44 759.17

Ningaloo 0.91 -84.86 69.29 -1.68 18 3.09 96.41

North Stradbroke Island

0.92 291.82 57.68 -0.66 133 4.17 762.02

Port Hacking Bay

0.93 190.02 60.5 -0.5 120 2.61 575.31

Rottnest Island

0.94 90.58 63.55 -0.91 112 3.98 631.96

Yongala 0.97 447.78 51.74 1.03 165 8.24 1169.29

MLR Results: Model 2

Page 19: High CO2 world

NRS Correlation Coefficient

Intercept SAL SST Chl-a n RSE AIC

General 0.95 583.85 49.68 -1.17 4.85 801 8.82 5766.72

Darwin 0.96 541.62 51.66 -1.44 6.16 39 8.79 285.98

Esperance 0.87 20.01 65.61 -1.25 6.01 36 5.51 230.85

Kangaroo Island

0.86 764.92 44.52 -0.3 -5.58 56 5.7 359.62

Maria Island

0.91 290.05 57.91 -0.86 2.28 132 3.37 701.47

Ningaloo 0.95 -392.98 78.5 -2.43 21.37 18 2.44 88.62

North Stradbroke Island

0.92 294.77 57.57 -0.64 1.87 133 4.17 762.89

Port Hacking Bay

0.94 184.22 60.58 -0.4 1.1 110 2.59 527.61

Rottnest Island

0.94 83.07 63.74 -0.9 2.29 112 3.99 633.44

Yongala 0.97 448.94 51.74 1.00 -2.08 165 8.23 1169.71

MLR Results: Model 3

Page 20: High CO2 world

NRS Correlation Coefficient

Intercept SAL SST logChl-a n RSE AIC

General 0.95 570.95 50.16 -1.08 3.21 801 8.75 5753.43

Darwin 0.96 566.45 51.19 -1.5 6.92 39 8.81 286.15 Esperance 0.87 25.14 65.62 -1.26 2.95 36 5.48 230.35 Kangaroo Island

0.86 763.08 44.44 -0.28 -2.02 56 5.67 359.05

Maria Island

0.91 284.14 58.19 -0.94 1.91 132 3.32 697.17

Ningaloo 0.94 -342.81 77.49 -2.43 6.86 18 2.56 90.32

North Stradbroke Island

0.92 294.29 57.62 -0.62 0.94 133 4.15 761.94

Port Hacking Bay

0.94 190.09 60.44 -0.37 0.84 110 2.58 526.96

Rottnest Island

0.94 86.86 63.67 -0.9 0.42 112 3.99 633.75

Yongala 0.97 460.53 51.21 1.04 -3.44 165 7.95 1158.29

MLR Results: Model 4

Page 21: High CO2 world

Methods: Statistical Analysis

Kolmogorov–Smirnov (K-S) Test- H0: Two sets of data are drawn from the same distribution- Two parameter test that tests mean and spread- Bootstrapped

Akaike’s information criterion (AIC)- Measures relative quality of statistical models- Combined measure of goodness of fit (RSE) and complexity (number of parameters) of

modelRelative Probability of Minimising Information Loss- Application of AIC values - exp( (AICj – AICmin)/2 ) - Allows differences in AIC to be quantified and compared in terms of probabilities,

rather than just “eyeballing”

Residual Standard Error (RSE)- Measure of the error of a model- Is in absolute units- Multiply by 1.645 to get an error corresponding to a 95% confidence level

Page 22: High CO2 world

K-S Tests - pvalues

NRSSSS SSS-SST SSS-SST-Chl-a SSS-SST-log(Chl-a) Open Ocean

Regional Coastal Regional Coastal Regional Coastal Regional Coastal Lee et al.

(2006)

Darwin 0.9757 0 0.1345 0.0003 0.0799 0.0001 0.1366 0.0003 0.0002Esperance 0.0858 0.0885 0.1081 0.0578 0.6632 0.0654 0.1945 0.0636 0.0337Kangaroo Island 0.6219 0 0.9082 0 0.2536 0 0.748 0 0.9078Maria Island 0.3863 0 0.659 0.0843 0.2409 0.0617 0.5181 0.0658 0Ningaloo 0.7166 0 0.9416 0.4407 0.939 0.232 0.9451 0.2328 0.1105North Stradbroke Island 0.0131 0.0791 0.7328 0.0009 0.1682 0.0007 0.6348 0.0006 0Port Hacking Bay 0.6665 0.0242 0.3613 0.2925 0.1639 0.9949 0.612 0.5038 0Rottnest Island 0.4865 0.2648 0.3284 0.0148 0.6301 0.007 0.8487 0.0146 0.002Yongala 0.0209 0 0.0559 0 0.2176 0.0002 0.0331 0.0007 0

Page 23: High CO2 world

Cross et al. 2013