Oct 5, 2016 - regresi secara lelaran dengan fungsi pemberat spatial (WFd) iaitu ...... To establish the empirical model, the nonlinear regression with reduced.
PSZ 19:16 (Pind. 1/13)
UNIVERSITI TEKNOLOGI MALAYSIA DECLARATION OF THESIS / UNDERGRADUATE PROJECT PAPER Author’s full name : NOR ZAFIRAH BINTI AB.LAH Date of Birth
: 11 MARCH 1989
Title
: Spatial Variability Assessment of Local Chlorophyll-a Estimation Using Satellite Data
Academic Session : 2015/2016 I declare that this thesis is classified as:
CONFIDENTIAL
(Contains confidential information under the Official Secret Act 1972)*
RESTRICTED
(Contains restricted information as specified by the organization where research was done)*
OPEN ACCESS
I agree that my thesis to be published as online open access (full text)
I acknowledged that Universiti Teknologi Malaysia reserves the right as follows: 1. The thesis is the property of Universiti Teknologi Malaysia 2. The Library of Universiti Teknologi Malaysia has the right to make copies for the purpose of research only. 3. The Library has the right to make copies of the thesis for academic exchange. Certified by:
SIGNATURE 890311-11-5356
NOTES:
SIGNATURE OF SUPERVISOR
(NEW IC NO/PASSPORT)
DR. MOHD NADZRI MD REBA NAME OF SUPERVISOR
Date: 5 OCTOBER 2016
Date: 5 OCTOBER 2016
*
If the thesis is CONFIDENTAL or RESTRICTED, please attach with the letter from the organization with period and reasons for confidentiality or restriction.
ii
“I hereby declare that I have read this thesis and in my opinion this thesis is sufficient in terms of scope and quality for the award of the degree of Master of Science (Remote Sensing)”
Signature
: ………………………….........
Name of Supervisor
: DR. MOHD NADZRI MD REBA
Date
: 5 OCTOBER 2016
iii
SPATIAL VARIABILITY ASSESSMENT OF LOCAL CHLOROPHYLL-A ESTIMATION USING SATELLITE DATA
NOR ZAFIRAH AB.LAH
A thesis submitted in fulfilment of the requirements for the award of the degree of Master of Science (Remote Sensing)
Faculty of Geoinformation and Real Estate Universiti Teknologi Malaysia
OCTOBER 2016
iv
I declare that this thesis entitled “Spatial Variability Assessment of Local Chlorophyll-a Estimation Using Satellite Data” is the result of my own research except as cited in the references. The thesis has not been accepted for any degree and is not concurrently submitted in candidature of any other degree.
Signature
:
....................................................
Name
:
NOR ZAFIRAH AB.LAH
Date
:
5 OCTOBER 2016
v
To my family, especially my beloved husband and son
vi
ACKNOWLEDGEMENT
Firstly, I would like to express my sincere gratitude to my supervisor Dr. Mohd Nadzri Md Reba, for his patience and immense knowledge. His guidance helped me throughout the time of research and writing of this thesis. Besides, I would like to thank my former supervisor Dr. Eko Siswanto for his time to answer all my doubt and questions regarding this research. The research of this thesis was financially supported by a grant from the Ministry of Higher Education flagship with Universiti Teknologi Malaysia under project vote number Q.J130000.2527.03H21 and many thanks to Dr. Katsuhisa Tanaka from Japan International Research Centre for Agriculture Science (JIRCAS) with collaboration of Penang Fish Research Institute (FRI) for sharing the in-situ data for my study.
My sincere thanks also go to my fellow friends whom always cheer me and giving me moral support and especially my bestfriend Mimi, who always accompany me during the sleepless night in the laboratory and during the hard time completing this thesis. Last but not least, I’m thankful to my family for supporting me directly or indirectly and financially throughout all my studies at University and as their prayers always with me. Also, thanks to my husband Mohd Firdaus Abdullah for giving me all the support and understanding throughout writing this thesis and my pregnancy.
vii
ABSTRACT
The estimation of Chlorophyll-a (Chl-a) for optically complex water from satellite is challenging. Moderate Resolution Imaging Spectroradiometer (MODIS) is an ocean colour satellite which has low spatial resolution and this has led to bias estimate and scale effect that eventually induced errors in Chl-a retrieval using local ocean colour algorithm. Studies on Chl-a variation, assessment of MODIS data and development of local ocean colour algorithm are less for Malacca Straits water. The aim of this study is to locally calibrate and validate the Chl-a derived from MODIS standard Chl-a algorithm (OC3M) on the latest R2013 data within the acceptable error tolerance at the Absolute Percentage Difference (APD) below 35% and to test the algorithm’s applicability. Iterative regression method with weighted function (WFd) namely Iterative Conditional Regression Model (ICRM) is introduced to reduce the spatial bias in the Chl-a estimate. Locally calibrated OC3M algorithm with in-situ data taken at two static stations and kernel 7×7 size named as OCms1 (calibrated with in-situ Case-1 water) and OCms2 (calibrated with in-situ Case-2 water) remarkably reduced the Chl-a bias with APD of 37% and 30% from 54% and 116% respectively. Then, using the ICRM, the APD of OCms1 WFd and OCms2 WFd is 26% and 29% respectively. Results of OCms WFd and OCms (with and without weighted function respectively) are combined for mapping the Chl-a in Case-1 and Case-2 waters. Result of applicability test and statistical analysis shows that OCms WFd ocean colour algorithm provides statistically highest accuracy for Chl-a estimation. The development of local Chl-a algorithm is essential for accurate Chl-a retrieval and it is significant to other marine studies such as in primary production and algal bloom in Malacca Strait water.
viii
ABSTRAK
Anggaran klorofil-a (Chl-a) untuk perairan yang kompleks secara optikal daripada satelit adalah mencabar. Pengimejan spectroradiometer resolusi sederhana (MODIS) adalah satelit warna lautan yang mempunyai resolusi spatial yang rendah dan membawa kepada anggaran bias dan kesan skala yang akan memberi ralat dalam dapatan Chl-a dengan menggunakan algoritma warna lautan tempatan. Kajian mengenai variasi Chl-a, penilaian data MODIS dan pembangunan algoritma warna lautan tempatan adalah kurang untuk kawasan perairan Selat Melaka. Tujuan kajian ini adalah untuk membuat kalibrasi tempatan dan pengesahsahihan Chl-a
yang
diperolehi daripada algoritma Chl-a piawaian MODIS (OC3M) ke atas data R2013 yang terkini dengan toleransi ralat yang diterima pada perbezaan peratusan mutlak (APD) di bawah 35% dan untuk menguji kebolehgunaan algoritma tersebut. Kaedah regresi secara lelaran dengan fungsi pemberat spatial (WFd) iaitu Model Regresi Lelaran Bersyarat (ICRM) diperkenalkan untuk mengurangkan bias spatial dalam anggaran Chl-a. Algoritma OC3M yang dikalibrasi secara tempatan dengan data lapangan yang diambil pada dua stesen cerapan statik dan saiz tetingkap 7x7 yang dinamakan sebagai OCms1 (dikalibrasi dengan data lapangan untuk perairan Kes-1) dan
OCms2 (dikalibrasi dengan data lapangan untuk perairan Kes-2) telah
mengurangkan bias Chl-a dengan ketara sebanyak 37% dan 30% daripada 54% dan 116%. Seterusnya dengan menggunakan ICRM, APD untuk OCms1 WFd dan OCms2 WFd adalah masing-masing 26% dan 29%. Keputusan OCms WFd dan OCms (dengan fungsi pemberat dan sebaliknya) digabungkan untuk memetakan Chla bagi perairan Kes-1 dan Kes-2. Keputusan untuk ujian kebolehgunaan dan analisis statistik menunjukkan algoritma warna lautan OCms WFd memberi ketepatan yang tinggi secara statistik untuk penganggaran Chl-a. Pembangunan algoritma Chl-a tempatan adalah penting untuk memperoleh Chl-a yang tepat dan boleh digunakan dalam kajian lautan yang lain seperti produktiviti primer dan letusan alga di perairan Selat Melaka.
ix
TABLE OF CONTENTS
ACKNOWLEDGEMENT ABSTRACT ABSTRAK TABLE OF CONTENTS
VI VII VIII IX
LIST OF TABLES
XII
LIST OF FIGURES
XIV
LIST OF ABBREVIATION LIST OF SYMBOLS LIST OF APPENDICES
XVII XIX XX
CHAPTER 1
1
INTRODUCTION
1
1.1 Introduction
1
1.2 Background of Study
3
1.3 Problem Statement
5
1.4 Research Objectives
6
x 1.5 Significant of Study
7
1.6 Study Area
8
1.7 Scope of Study
10
CHAPTER 2
13
LITERATURE REVIEW
13
2.1 Introduction
13
2.2 Remote Sensing Ocean Colour for Chlorophyll-a Estimation
14
2.2.1
Chlorophyll-a Estimation in Case-1 and Case-2 Water
15
2.3 Calibration And Validation Methods For Chlorophyll-A Algorithm
16
2.4 Spatial Variability Bias In Chlorophyll-A Estimation
22
2.4.1
MODIS-Retrieved Chl-a Empirical Algorithm
23
2.4.2
MODISA R2010 and MODISA R2013 Data
24
2.5 Chlorophyll-A Variability Study In The Malacca Strait
25
CHAPTER 3
28
RESEARCH METHODOLOGY
28
3.1 Introduction
28
3.2 Data Acquisition
29
3.2.1
Remote Sensing Data
30
3.2.2
In-situ Data
33
3.3 Establishing Match-up Samples
34
3.4 Calibration and Validation Exercise
36
3.4.1
Iterative Conditional Regression Model (ICRM)
37
3.4.2
Spatial Weight Function
41
3.5 Accuracy Assessment
43
xi 3.5.1
Local OC3M Algorithm Performance Evaluation
44
3.5.2
Local Algorithm Applicability Test
45
CHAPTER 4
48
RESULTS AND DISCUSSION
48
4.1 Introduction
48
4.2 Preliminary Comparison between OC3M Chl-a and In-situ Chl-a
48
4.3 MODIS R2010 versus MODIS R2013
50
4.4 Impact of Spatial Bias
53
4.5 Local-tuned OC3M with spatial Weight Function (WFd)
60
4.6 Local Algorithm Applicability
64
4.7 Chlorophyll-a variation in the Malacca Strait
73
CHAPTER 5
78
CONCLUSIONS AND RECOMMENDATIONS
78
5.1 Conclusion
78
5.2 Limitation of the study
80
5.3 Future Recommendations
80
REFERENCES
81
APPENDIX A
87
APPENDIX B
99
APPENDIX C
103
xii
LIST OF TABLES
Table 2.1 Summary of the previous studies on the standard NASA Chl-a algorithm development and evaluation in local region. ............................ 21 Table 2.2 List of previous study of Chl-a retrieval in the marginal sea. .................... 26 Table 3.1 Data specification of Level 2 product used in this study. ......................... 31 Table 3.2 Description of l2_flags of MODISA level 2 data, the list can be accessed from the data product through SeaDAS. .................................... 31 Table 3.3 Summary of Cal/Val exercise set-up for this study. ................................. 46 Table 4.1 Statistical analysis results for Chl-a obtained by standard OC3M algorithm for different MODISA data processing versions and type of waters. ........................................................................................... 51 Table 4.2 The statistical results for Chl-a obtained by locally-tuned OC3M algorithm for different MODISA data processing versions and type of waters. ........................................................................................... 52 Table 4.3 Statistics of Chl-a retrieved by in-situ and standard OC3M algorithm using k3 and k7 during the period of study at ST1 station (Case-1 water). The standard deviation of Chl-a for each kernel size is also presented. ..................................................................... 56 Table 4.4 Statistics of Chl-a retrieved by in-situ and standard OC3M algorithm using k3 and k7 during the period of study at ST2 station (Case-2 water). The standard deviation of Chl-a for each kernel size is also presented. ..................................................................... 56
xiii Table 4.5 Result of regression and statistics of standard, local and weighted function OC3M algorithm for dual-case, Case-1 and Case-2 water in different kernel size of 3x3 (k3) and 7x7 (k7). OC3M, OC3M 1 and OC3M 2 are the Chl-a extracted using standard OC3M algorithm for dual-case, Case-1 and Case-2 respectively. OCms WFd, OCms1 WFd and OCms2 WFd are the Chl-a estimated by using locally-tuned algorithm with weight function distance (WFd) for dual-case, Case-1 and Case-2 respectively. ............................. 58
xiv
LIST OF FIGURES
Figure 1.1 Map of Malacca Strait showing two stations, Station 1, ST.1 (northern part) and Station 2, ST.2 (southern part) where the in-situ Chl-a data were taken, three section of Malacca Strait (1-North, 2Middle, 3-South) and 4 main rivers along the coast of the strait. ........... 10 Figure 2.1 Spectra profile of remote sensing reflectance (based on Bricaud et al., 2004) for phytoplankton in the range of blue to green. For Case-1 profile the lower blue-green ratio presenting the higher Chla retrival, while for Case-2 profile the higher blue-green ratio reflecting the overestimation of Chl concentration in the algorithm. ..... 18 Figure 3.1 The main flowchart of the study consist of several sub-flowcharts that shows detail process of the method involved, (a) Figure 3.6; (b) Figure 3.7; and (c), (d), (e) in Figure 3.8. .......................................... 29 Figure 3.2 Graphic user interface of SeaDAS 6.3 with features for data display and extraction. ......................................................................................... 32 Figure 3.3 Chl-a measurement by fluorometer. Left: A schematic diagram shows the instrument installed in 1 meter depth from the center of loop buoy, Right: Buoy on the ocean surface with fluorometer mounted at the center. ............................................................................. 34 Figure 3.4 Illustration of establishment of match-up sample by applying the position of the pixel located at the center (shown in dot at ix and iy) and the position of in-situ point (shown in cross at jx and jy) to determine the relative distance, Δd. Different kernel size of 3×3 and 5×5 resulting in 3km × 3km and 5km × 5km respectively was highlighted. Pixels contaminated by cloud features are in black and the valuable ones are in red and blue. ..................................................... 35
xv Figure 3.5 Graphic user interface of the Solver tool in Microsoft Excel 2010 with the input parameters. ....................................................................... 39 Figure 3.6 Flowchart of the iterative fitting routine to locally-tuned the OC3M algorithm. ................................................................................................ 40 Figure 3.7 The sub-flowchart of iterative regression with weight function applied in ICRM. ..................................................................................... 42 Figure 3.8 Sub-flowchart of accuracy assessment. The OCms consists of the locally-tuned Chl-a algorithm, and OCmsWFd consists of the locally-tuned Chl-a algorithm. Both variants are applied for Case-1 and Case-2 water of the Malacca Strait................................................... 43 Figure 4.1 Plot of Chl-a estimated by standard OC3M and measures on seatruth at ST1 (Oct-Jan 2011) and ST2 (Feb-July 2012) stations. The plot also supported by the water leaving radiance at 554 nm for distinguishing the types of water. Chl-a estimation different by using R2010 and R2013 version of data were addressed. ....................... 49 Figure 4.2 Plot of in-situ and locally tuned Chl-a for ST1 and ST2 stations. Both data versions were applied in local OC3M algorithm (OCms). ..... 50 Figure 4.3 Plot of Chl-a derived by standard OC3M algorithm in different kernel sizes (3x3, 5x5 and 7x7) and comparing with the in-situ Chla observations in Case-1 water (ST1). .................................................... 54 Figure 4.4 Plot of Chl-a derived by (a) standard OC3M algorithm and (b) by local OC3M algorithm using k3 and k7 collected at both in-situ stations during the period of study. ......................................................... 55 Figure 4.5 Plot of regression between MODIS derived Chl-a and in-situ Chl-a using standard, local and local with weight function distance in dual-case, Case-1 and Case-2 water respectively. The red, blue and dashed line indicates unity, Type-II regression fit and general regression fit, respectively. Error bars are designed by the standard deviation and the average Chl-a of the sample points. The result is basically produced by the kernel size of 3×3. ......................................... 63 Figure 4.6 Map of climatology OC3M-retrieved Chl-a from 2009 until 2013. ........ 65 Figure 4.7 Map of climatology Chl-a retrieved from locally-tuned algorithm, OCms from 2009 until 2013. .................................................................. 66
xvi Figure 4.8 Map of climatology Chl-a retrieved from applying locally-tuned algorithm, OCms WFd from 2009 until 2013. ........................................ 67 Figure 4.9 The p-value map from the climatology Chl-a retrieved by using OCms algorithm. The black colour in the water area of the map is for significance level of p ≤ 0.05 and white for p ≥ 0.05. ....................... 69 Figure 4.10 The p-value map from the climatology Chl-a retrieved by using OCms WFd algorithm. The black colour in the water area of the map is for significance level of p ≤ 0.05 and white for p ≥ 0.05. ........... 70 Figure 4.11 The mean Chl-a of the climatology from 2009-2013 was plotted for the a) OC3M-retrieved Chl-a, b) OCms and c) OCms WFd, respectively. The mean Chl-a was extracted in 3 sections of Malacca Strait, which are north (blue line), middle (green line) and south (red line), with the standard deviation value as the error bar showed in the plots. ................................................................................. 71 Figure 4.12 Plot of the mean Chl-a value at 3 sections of Malacca Strait and the graph plot of the differential value plot of Chl-a between the 3 algorithms (OC3M-OCms; OC3M-OCmswfd; OCms-OCmswfd), the red, green and blue line represents Chl-a value from OC3M, OCms and OCms WFd respectively, for the selected months. ............... 72 Figure 4.13 Climatology plot of river discharge near the Malacca Strait (19751997). Red, green, pink and blue lines are referred as Kerian river (Perak), Selangor river, Klang river and Langat river (Selangor), respectively. ............................................................................................ 73 Figure 4.14 The selected data in October and November is an example of the map of normalized water-leaving radiance (nLw) at wavelength 554nm as the turbidity index. .................................................................. 74 Figure 4.15 Plot of precipitation climatology of area averaged near in-situ location, ST1 and ST2 from 2009 until 2013, (monthly GPCC rainfall, TRMM). ..................................................................................... 75 Figure 4.16 Plots of climatology of wind speed and wind stress curl from 2009 until 2013. ............................................................................................... 76
xvii
LIST OF ABBREVIATION
APD
Mean absolute percentage difference
BNO
Bagan Nakhoda Omar
Cal/Val
Calibration and Validation
Chl-a
Chlorophyll-a
CDOM
Colored dissolved organic matter
CI
Confidence Interval
CZCS
Coastal Zone Colour Scanner
EMR
Electromagnetic radiation
Gof
Goodness of fit
ICRM
Iterative Conditional Regression Model
MERIS
Medium Resolution Imaging Spectrometer
MODIS
Moderate Resolution Imaging Radiometer
MBR
Maximum Band Ratio
NASA
National Aeronautics and Space Administration
NIR
Near infrared
Rrs
Remote sensing reflectance
RPD
Relative percentage difference
RMSE
Root mean square error
xviii ST1
Station 1
ST2
Station 2
SeaDAS
SeaWiFS Data Analysis System
SeaWiFS
Sea-Viewing Wide Field-of-View Sensor
SSE
Sum of Squares Error
TSS
Total suspended sediment
VIIRS
Visible and Infrared Imager/Radiometer Suite
xix
LIST OF SYMBOLS
a
Absorption
aT
Total absorption coefficient
bb
Backscatter
bbT
Total backscatter coefficient
Ca
Chlorophyll concentration
Lw
water-leaving radiance
Chlret
Chl-a retrieved by the Chl-a algorithms
Chlis
In situ Chl-a
nLw
Normalized water-leaving radiance
Rrs
Remote Sensing reflectance
R3M
Maximum 3 band ratio
λ
Wavelength
xx
LIST OF APPENDICES
APPENDIX A
Fundamental of Ocean Remote Sensing
78
APPENDIX B
Additional Result
80
APPENDIX C
Calibration and Validation Data
103
1
CHAPTER 1
INTRODUCTION
1.1
Introduction
Phytoplankton is a marine photosynthetic microorganisms formed by the green biomass called chlorophyll-a (Chl-a) which is the primary molecule of chlorophyll pigment and responsible for the photosynthesis process. Phytoplankton plays major role in the oceanic food chain and has become the oxygen production agent to ocean bio-creatures and the environment regulator in the ocean carbon cycle.
Phytoplankton intrinsically helps to regulate the world climate and by
knowing the spatial and temporal attributes would improve understanding of its influences to the world climate pattern. Measuring phytoplankton in the ocean is literally a tedious and complicated practice.
However, by the advancement of
satellite remote sensing the phytoplankton estimation is plausible thanks to spectroscopic measurement, through that the Chl-a optical properties can be determined as a function of the absorption and scattering representing the magnitude of concentration and spatio-temporal distribution of phytoplankton abundant. In fact the optical properties variant provide synoptic and continuous mapping of Chl-a at promising resolution in time and space.
Optically sensing Chl-a applies the electromagnetic radiance (EMR) to define the colour or spectral related feature of Chl-a in the bio-optical model and this application literally known as ocean colour remote sensing is very prevalent in
2 marine biological research.
Satellite based ocean colour bio-optical model has
evolved to cope with different mapping scales and various ocean climate and as a result, different algorithm and application have been demonstrated. Remote sensing image is composed of pixels representing the water optical properties that geometrically registered to earth coordinates. To estimate the remotely sensed Chla, two ocean colour model have been devised. First, the empirical model in which statistical regression is applied between sea truth Chl-a and satellite derived apparent optical properties (AOP) (e.g, the remote sensing reflectance, Rrs) by assuring both measurements are highly correlated in time and space.
The most favourable
empirical model depends on the spectral bands (typically by blue and green bands) and the water types (Case-1 and Case-2 water). Secondly is the analytical model which based on the inversion of a forward radiance model.
Other than that,
integration of both modelling schema was also devised (known as semi-analytical model) but requires theoretical AOP estimation optimized by in-situ inherent optical properties (IOP) (all definition of AOP and IOP are described in the glossary). The present thesis discusses on the application of empirical model to estimate the Chl-a due to the fact this model is straightforward and no dependent to ocean and geophysical parameters but completely dependent to satellite remote sensing products..
To study potential of the empirical model in Chl-a estimation, Malacca Strait is chosen in this thesis. Malacca Strait is one of the marginal seas in the Peninsular Malaysia and has significant value to Peninsular Malaysia as one of the productive fishing grounds (692,985 metric tons of fishes which valued at RM2.263 billion per year) as reported by Kasmin (2010) and the prominent ocean trade network in the Silk Road. This area is surrounded by different water types, receiving continuous water disposal from the major river outlets and experiencing distinctive seasonal climate every year which make it the best ocean water to examine the quality of satellite derived Chl-a by empirical model and assses the impact of spatial variation.
3 1.2
Background of Study
Optical satellite remote sensing basically equipped by passive sensor to observe all reflected and emitted EMR coming from the ocean surface at visible to near infra-red (NIR) wavelength. The NASA Earth Observation System (EOS) program has commissioned series of passive ocean colour remote sensing in space such as Coastal Zone Colour Scanner (CZCS) (Gordon et al., 1980); Sea-Viewing Wide Field-of-View Sensor (SeaWIFS) (Hooker et al., 2000); Moderate Resolution Imaging Spectroradiometer (MODIS) (Esaias et al., 1998); Visible and Infrared Imager/Radiometer Suite (VIIRS) (Feldman, 2015), and Medium Resolution Imaging Spectrometer (MERIS) (Le et al., 2013).
Amongst them, MODIS is
currently the most distinctive ocean colour mapping sensor that provides continuous, long-term and the most reliable Chl-a related products for ocean and atmospheric studies in the last decade. Prior to MODIS mision, the SeaWiFS brought 8 spectral bands ranging from 412 to 865 nm to collect global optical data at 4 km spatial resolution but the mission was completely shut down in 2010. MODIS offers 36 spectral bands at higher spatial resolution of 1km. The spectral bandwidth is narrower and more sensitive to the variation of bio-optical signatures because of the signal-to-noise ratio (SNR) is 2-4 times higher than the SeaWiFS (Hu et al., 2012). The recent MODIS data taken by Aqua platform (hereafter denoted as MODISA) has been released since 2013 (R2013 version) and the quality is greater owing to the higher SNR derived from the in-depth radiometric correction at band 8 and 9 (412 and 413nm respectively) (Feldman, 2014). However, it has yet a study that demonstrates the impact of using MODISA R2013 data for the Chl-a estimation in Malacca Strait.
Ocean colour retrieval algorithm is specifically designed either for Case-1 or Case-2 water in bio-optical model, (Morel & Prieur, 1977). The Case-1 water has the water optical properties that are mainly induced by the phytoplankton and the covarying in-water constituents. For the Case-2 water, the water optical properties are relatively more dominated by other non-co-varying in-water constituents either in the form of organic or inorganic particles than the phytoplankton. The empirical Chl-a estimation is complicated to be applied simultaneously for Case-1 and Case-2
4 that leads to inherent bias. In case of Malacca Strait, different water types would exist in a field-of-view (FOV) of EOS ocean color satellite representing Chl-a in a single pixel because in 1km x 1km areal pixel there are active nutrient rich sediment discharge from the nearby river outlets and continuous upwelling and downwelling currents from various depth variation at near and off the coast that diversified the ocean salinity and temperature rate.
Technically, correlated satellite derived Chl-a is based on the concept of ratio of the remote sensing reflectance at blue to green band (Tassan, 1981). This rationale lies on the fact that the photosynthetic pigment of Chl-a absorbs much blue and red radiance than of the green and reflects much radiance in blue to green. The hypothesis is that band ratio increases as the amount of the Chl-a abundant being sensed is higher. Though, the band ratio sometimes impaired by the lower band ratio value (i.e., in the case of 443/555 nm) when the higher Chl-a abundant escalates the Rrs at 555nm (Lee & Carder, 2000; Martin, 2014). Therefore, the maximum band ratio (MBR) is introduced and taking advantage of significant SNR remains as high as possible even over a broad range of Chl-a concentration. The above mentioned band ratio methods completely rely on the Rrs at different ocean color bands (Dierssen, 2010) and this has proved that two bands (OC2), three bands (OC3) and four bands (OC4) have been applied in EOS missions. In the present thesis, three ocean color bands was used in MODIS Chl-a estimation and commonly known as OC3M. The significant usage of OC2, OC3 or OC4 was discussed thoroughly in O’Reilly et al. (1998). To date, there are other latest empirical algorithm have been devised such as color index (CI) (Hu et al., 2012), normalized difference chlorophyll index (NDCI) (Mishra & Mishra, 2012), and semi-analytical algorithm (SAM_LT) (Pieri et al., 2015), however, those variants are mainly introduced to optimize the typical band ratio algorithm for estimating Chl-a concentration in oligotrophic water and turbid water area in low Chl-a concentration (below 1 mm3/mg).
Empirical Chl-a algorithms such as OC4v4 and OC3M devised for SeaWiFS and MODIS respectively have been proved as the global Chl-a algorithm. Though, the satellite derived Chl-a may differ if these algorithms are applied locally (within 1 pixel or 9 pixels) or regionally (more than 9 pixels) because the Chl-a diversity in the
5 ocean is exceptionally dynamic. It is a need to calibrate and thus validate the global Chl-a algorithm by downscaling the Chl-a at local scale which have been done by several studies (Cannizzaro & Carder, 2006; Lee & Hu, 2006; Le et al., 2013). The OC3M algorithm was designed to estimate the Chl-a for Case-1 water and this would yield misleading Chl-a if it was applied in Case-2 water (Gordon & Clark, 1981; Moses et al., 2009; Yang et al., 2010). Calibration and validation exercise (Cal/Val) is therefore compulsary to apply on the satellite derived Chl-a in all cases of water as long as the absolute percentage difference (APD) is less than 35% (accuracy set by the NASA). However, this accuracy is nearly hard to achieve on OC3M and SeaWiFS OC4v4 algorithm particularly for Case-2 water (Esaias et al., 1998; Darecki & Stramski, 2004; Volpe et al., 2007).
The Asian monsoon strongly influences the spatial distribution of Chl-a in Malacca Strait and satellite observation has proved as the most practical tool to measure the impact (Tan et al., 2006). Interannual Chl-a variation in the northern, middle and southern part of Malacca Strait was majorly associated with the ElNino/Southern Oscillation (ENSO) and river runoff as reported in (Siswanto & Tanaka, 2014). The study shows that the Chl-a variation was influenced by the north-east (December to January) and south-west (May to August) monsoon however the impact of local Chl-a algorithm towards spatial variability was not presented.
1.3
Problem Statement
Based on the background study, issues of this study can be drawn as follows: 1. Studies by Ab.Lah et al. (2014) and Darecki (2004) on proved that the MODISA empirical Chl-a algorithm (OC3M) exhibits fairly acceptable Chl-a estimates with the APD 90%) near the coast water (probably by Case-2 water). By the recent MODISA R2013 data released, the empirical Chl-a estimation can be improved and performed at local scale. To date no study has been locally
6 conducted to test applicability and accuracy of the R2013 on the Strait Malacca water.
2. MODIS pixel that matchs up with the corresponding in-situ point is needed for the empirical OC3M algorithm. To perform the pixel matching, different kernel window (starting from 3x3 kernel) is possible to use. Yet, size of kernel is limited because the Chl-a representing in the pixel varies with the corresponding in-situ Chl-a. Pixel averaging is commonly practiced but this would lead to spatial bias as the Chl-a concentration is fairly homogeneous within the 0.1 m2 water column and it is arguable to compare the averaged MODIS Chl-a concentration of one pixel in approximate 1 km2 water column. In this case, the spatial variability impact may reduce the correlation of OC3M with the in-situ (Chen et al., 2013).
3. Calibration and validation exercise (Cal/Val) requires at least 30 match-up samples (to achieve normal distribution) that are sparsely located in the study area. Yet, match-up samples are located at two independent in-situ stations where continuous daily Chl-a was measured in this study. No study was conducted to assess the spatial impact on Cal/Val by means of static sample.
1.4
Research Objectives
The aim of this study is to calibrate and validate the Chl-a derived from empirical Chl-a model using the latest reprocessed MODISA R2013 data for Malacca Straits. The objectives are the followings;
1. To develop local empirical model for estimating satellite derived Chl-a over Malacca Strait by means of MODISA R2013 data and global OC3M;
2. To compare the performance of local Chl-a estimation using MODISA R2010 and R2013 data;
7
3. To assess the impact of spatial variability on local Chl-a algorithm by means of the estimation accuracy in different kernel window size and distance of pixel to in-situ point; and
4. To test the applicability of the new calibrated Chl-a algorithm in the Malacca Strait water in regard to the impact of seasonal monsoon, coastal outputs and precipitation.
1.5
Significant of Study
Retrieving accurate Chl-a estimate by using remote sensing over Malacca Strait water is worthwhile as this technique conveys reliable information of phytoplankton and nutrient at larger scale and faster acquisition. Synoptic Chl-a mapping implies the intensity distribution and origins of nutrient along the coast of Peninsular Malaysia and Sumatra Indonesia. Massive suspended sediment loading from rivers outlets in Peninsular Malaysia and Sumatra that caused variation of phytoplankton can be determined by map of Chl-a concentration derived from this study. Information of nutrient is essential to determine the degree of marine biological production in Malacca Strait. All these marine substances significantly influence the spatio-temporal variability of phytoplankton and hence the Chl-a density in Malacca Strait.
Knowing the accurate Chl-a estimation in marginal seas is essential to understand the global ocean production. Study on spatio-temporal Chl-a variability is foreseen in future research to improve nowcasting of ocean climate change and algal bloom prediction model. This thesis shows the quality assessment procedure on local MODIS Chl-a algorithm for Malacca Straits that accounted impact of spatial variability of low spatial resolution.
8 The state-of-the-art and accurate local Chl-a algorithm with low impact of spatial variability induced by low spatial resolution in MODIS pixel. The new calibrated Chl-a MODIS could help any future research of marine biology in Malacca Straits (Tan et al., 2006) and to promote the application of MODISA R2013 in marine research. Besides, the MODIS R2013 Chl-a product could support the development of marine database of Malaysia National Oceanographic Data Centre (MyNODC).
Cal/Val exercise needs well-distributed in-situ to increase more match-up sample with satellite observation and this could be carried out by using vessel or by number of scattered bouys to sample Chl-a over the study area. Yet, it becomes more troublesome when the satellite data was hampered by cloud cover or limited number and distribution in-situ points were exist. This could no longer be the issue in this study as the new Chl-a algorithm demonstrates straightforward Cal/Val at promising results with static distribution of in-situ point. The procedure exhibits an alternative way to reduce spatial variability when the option of using the enormous pixel kernel size is needed.
1.6
Study Area
The Malacca Strait has relatively shallower in depth (absolute depth of 300 to 400 meter) than in the South China Sea (60 to 5500 meter depth from the margin to the northeast basin) but both marginal seas contain diverse salinity, temperature and optical water properties due to its geographical features and seasonal climatology. The Malacca Strait is one of the most productive waters in the Malaysia with high nutrient inputs discharged from the rivers (Ali Yousif, 2009) and this in turn intensifies the level of Chl-a abundant. Besides, its coastal water region typically exhibits higher temporal and spatial variations of Chl-a concentration induced from the climatic, biological, physical and chemical condition ( Thia-Eng et al., 2000; Abdul Hadi et al., 2013; Haldar et al., 2013). Two in-situ data stations in Malacca Strait measures continuous daily Chl-a located at Payar Island, Kedah (ST1) and Bagan Nakhoda Omar reservoir, BNO, Selangor (ST2).
All in-situ Chl-a
9 measurements are provided by Japan International Research Centre for Agriculture Science (JIRCAS) and Penang Fish Research Institute (FRI) for this study. The insitu Chl-a was measured based on the method proposed by Suzuki & Ishimaru (1990). Figure 1.1 showed the study region with two static in-situ stations and main rivers along the coast of Malacca Strait which are Kerian river (Sg. Kerian), Selangor river (Sg. Selangor), Klang river (Sg. Klang), and Langat river (Sg. Langat) where the river discharge inputs to the Malacca Straits. Malacca Straits has been divided into three sections which are, 1) north, 2) middle and 3) south, because the north and south region are very different in terms of water optical properties, physical oceanography (Andaman Sea water influence northern region and South China Sea water influence southern region), and its bathymetry where north is deep and wide compare to south which is shallow and narrow.
10
Figure 1.1 Map of Malacca Strait showing two stations, Station 1, ST.1 (northern part) and Station 2, ST.2 (southern part) where the in-situ Chl-a data were taken, three section of Malacca Strait (1-North, 2-Middle, 3-South) and 4 main rivers along the coast of the strait.
1.7
Scope of Study
This study will mainly focus on assessing the spatial variability of the local Chl-a estimation in terms of algorithm statistical analysis, Chl-a estimation bias, and local Chl-a algorithm applied map. The empirical algorithm (OC3M) was used to retrieve the Chl-a and it was calibrated using the in-situ data at two stations in Malacca Strait to achieve the objective of developing the locally-tuned OC3M algorithm using the latest MODISA R2013. The primary data used in this study is the MODISA Level 2 with the resolution of 1km. In this study, the MODISA data
11 was preferred instead of MODIS-Terra because of the 1km and 250m of the MODIS-Terra data has lower SNR and its application mission is not suitable for ocean studies (Xiaoxiong et al, 2005).
To establish the empirical model, the nonlinear regression with reduced major axis (RMA) method is applied. The fitting method presented in this study was implemented for the data sample that met the limitation of the study, which is data sample less than 40 match-up points and the in-situ data was from few static stations that were sparsely distributed. Throughout the Cal/Val process, the standard Chl-a algorithm was locally-tuned in the Case-1 and Case-2 water separately and also in combined water cases. This is to determine the best way that the local-tuned Chl-a algorithm give the best result for the estimated Chl-a value in the study area.
In the accuracy analysis, error of fitting is assessed based on the result of the confidence interval (CI), sum of squares error (SS) and goodness of fit test (gof) and error of data (i.e., induced by spatial variability of MODISA pixel) is evaluated by absolute percentage difference (APD), relative percentage difference (RPD), root mean square error (RMSE), mean normalized bias (MNB) and the correlation of determination R2. This study sets the APD lower than 35% the positive R2 as the major argument as the correlation of satellite pixels established with the static in-situ station. Practically, the satellite oceanography processing requires well-distributed in-situ points to increase match-up with synoptic satellite coverage, and this is not the case applied in this study.
Correlation to static in-situ introduces spatial
variability of MODISA Chl-a, therefore the assessment of this spatial variability impact is carried out by different different sizes of pixel kernels (e.g., 3x3, 5x5 and 7x7) and in turn, optimizing the chances of number of match-ups. To compensate the variability impact, the spatial weight function is employed to the Rrs satelliteretrieved and consideration on the temporal window for satellite acquisition to in-situ measurement time is also taken into account.
12 Malacca Strait is categorize as eutrophic water and in order to test the applicability of the derived locally tuned Chl-a model, the Malacca Strait water was divided into 3 parts (i.e., Northern part which basically is the Case-1 water, Middle part which is basically dual-classification waters and Southern part which is the Case-2 water). Applicability test encompasses estimation of p-value and statistical analysis of the related geophysical parameters (i.e., river discharge, rainfall rate, insitu SST and suspended sediment) to help in understanding the Chl-a variation during the study period.
13
CHAPTER 2
LITERATURE REVIEW
2.1
Introduction
Remote sensing has been widely used to measure the optical properties of the biological and chemical constituents of water bodies in the last few decades. The foundation of optical oceanography is the interactions of light with the water bodies and its content which resulting the spectral irradiance and radiance (see full definition in Appendix A). Ocean colour is defined as the spectral distribution of leaving radiance from the ocean at the visible wavelength. The sunlight is not merely reflected from the sea surface, the sunlight that has entered the ocean has been selectively absorbed, scattered and reflected by its in-water constituents in visible wavebands. The presence of pelagic organism dissolved organic material and particulates in the upper column of ocean water are known as in-water constituents. Literally water has no colour, but it is largely induced by its water optical properties that are divided into two classes, the inherent optical properties (IOP) and apparent optical properties (AOP) (see full definition in Appendix A). The water optical properties are different in different geographical location and environmental effect, where the in-water constituents can influence the water optical properties and thus exist two classification of water in ocean remote sensing study which are Case-1 and Case-2 water that refer to water optical properties mostly dominated by phytoplankton and water optical properties dominated by other substances (eg. Suspended sediment, colour dissolved organic matter (CDOM), geblstoff, etc.) than phytoplankton, respectively. This chapter highlights the following topic:
14 a) Remote sensing ocean colour for Chl-a estimation in Case-1 and Case-2 water b) Calibration and validation methods for Chl-a algorithm in marginal seas c) Spatial variability bias in Chl-a estimation d) Chl-a variability study in the Malacca Strait
2.2
Remote Sensing Ocean Colour for Chlorophyll-a Estimation
Satellite ocean colour data has successfully provided important information in bio-optical constituents (eg. Chlorophyll, CDOM and gelbstoff) in a long-term and synoptic observation. Nonetheless, the data accuracy requires the production of quality-controlled dataset accompanied by robust statistical analyses of the errors associated with the retrieval procedures, e.g. atmospheric correction, bio-optical algorithms. Thus, space agency (NASA) have collected vast databases of in-situ data for calibration and validation of the satellite product such as Chl-a for the development of empirical algorithm used in the satellite data processing chains (e.g., OC4v4 for SeaWiFS, OC3 for MODIS and Algal1 for MERIS). Most researchers working in the bio-optical algorithm development tend to adopt the method of dualcase classification of sea water.
The organic particulates, the dissolve organic
matter, the suspended particulate matter and the inorganic particulates (sand, clay) that enter the ocean through river runoff can remarkably determine the ocean colour. Given the diversity of this dissolved and suspended material, Morel & Prieur (1977) in Martin (2014) study divided the ocean into Case-1 and Case-2 waters. Different region possessed different inherent optical properties of their water and previous study has found that ratio of red and green Rrs has stronger correlation with Chl-a concentration (Tzortziou et al., 2007). This is because the water leaving radiances in the near-infrared (NIR) is resulting from the strong scattering of suspended particles which is often happened in turbid or optically complex water and thus gives substantial uncertainties.
15 2.2.1
Chlorophyll-a Estimation in Case-1 and Case-2 Water
In ocean colour algorithm development, the conventional method commonly applies dual-case classification (integration of Case-1 and Case-2 water) where the huge differences in Chl-a derived value between both cases has been detected and thus makes it difficult for algorithm tuning. Some study chose to separately classify the water into Case-1 or Case-2 depending on how that classification could help in algorithm development, and even there is some study that did not classify the water type to avoid ambiguity of retrieved Chl-a (Chen et al., 2013; Mobley et al., 2004). Besides, in Lee and Hu (2006) study had stated that the global distribution and percentage of Case-1 water can spatially change with season. This suggests that the calibration and validation study period is long enough to observe the variability of Chl-a estimation during the seasonal month. The standard algorithms for Chl-a proposed by space agency have nominal accuracy of ~ 35% in the Case-1 water (O’Reilly et al., 1998). However, some studies has showed that the standard Chl-a algorithm does not always works well in coastal and optically complex water (associated with Case-2 water) study region, which give the purpose to validate and calibrate the Chl-a retrieved by standard Chl-a algorithm in the local region. Previous study had demonstrated that the standard algorithms were inappropriate to estimate the chlorophyll in Mediterranean Sea which is the case of oligotrophic water (Volpe et al., 2007). In their research, the standard OC4v4 algorithm was performed worse (uncertainty of ~100%) than the two regional algorithms that were tested.
Moreover, study by Darecki and Stramski (2004) has evaluated the
performance of standard MODIS Terra and SeaWiFS ocean colour algorithms in the Baltic Sea which is optically complex water, and the result showed large overestimation in chlorophyll retrievals with the mean normalized bias (MNB) of ~ 150 – 200%. Then, the regionally tuned MODIS Chl-a algorithm has reduced the overestimation of Chl-a to 26%, but the root mean square (RMS) error was still large.
Both studies have not given the absolute percentage difference of the
estimated Chl-a and the method to reduce the bias in the Chl-a estimation. Later, other study by Dierssen (2010) has given the perspective of Chl-a estimation from satellite could resulted an error by a factor of 5 and more. A report of global evaluation of Chl-a estimation by standard algorithm MODIS in Case-2 water
16 conclude that the high inaccuracies were expected in the regional Chl-a estimation due to the layer systems such as river discharged and ice-melt plumes with monospecific blooms and vertical structure. But as more field data become available, regional accuracies with the global algorithms can be assessed, and local improvements can be implemented.
As the climate changes, the chlorophyll
distribution and other in-water constituents may shift in response to altered environmental forcing and further induce to the uncertainty of the Chl-a estimation. Thus, accurate global assessments of phytoplankton will require improved technology and modelling, enhanced field observations, and ongoing validation of the bio-optical product.
2.3
Calibration And Validation Methods For Chlorophyll-A Algorithm
According to Morel and Gordon (1980), there are three approaches to interpret ocean colour data in terms of in-situ optical constituents, which are empirical, semi-empirical or semi-analytical. O’Reilly et al. (1998) reported various algorithms for Chl-a retrieval were tested using SeaWiFS Bio-optical Miniworkshop (SeaBAM) data.
From this evaluation and assessment workshop,
empirical variant as found the most straightforward and less parameter dependent algorithm that only relies on statistical regression of radiance to chlorophyll concentration. Literally, the empirical algorithm is a function of the wavelength dependent Rrs which is a linear function of subsurface reflectance R(λ) at the range of ocean color wavelength. On the other hand, the semi-analytic or semi-empirical algorithm combined theoretical models of the Rrs dependence of the backscatter and absorption ( bb / a ) with empirical formulas that describe the dependence of absorption and backscatter on the oceanic constituents such as CDOM and phytoplankton pigments, the algorithm analytically estimate bb / a from R. The semi-analytic approach can reduce the dependence on illumination conditions that exist in the R versus C relationship of the empirical approach. But, the limitation of this approach is that it requires a spectral dependency such as backscattering and absorption by detritus which could influence the absorption spectra on pigment concentration, which is yet well characterized (Sathyendranath, 2000). However,
17 this study is not a case of semi-analytic algorithm and more details of semi-analytic approach can be found in He et al. (2000), Carder et al. (2003), and Carder et al. (2004).
The inaccuracy of Chl-a estimation in near coastal water is commonly due to other in-water constituents. The empirical algorithm based on the spectral band-ratio estimates the Chl-a accurately and efficiently in Case-1 water, where the optical properties and bottom reflectance is negligible (Cannizzaro & Carder, 2006). However, different water region may have different sensitivity towards the bluegreen ratios combination due to the different amount of in-water constituents in that region.
The space borne ocean colour sensor has many different bands to derive different optical ocean properties. In the case of Chl-a detection, wavelengths range from green to blue spectra is commonly used.
The fundamental assumption
underlying the empirical band ratio retrieval method is that the optical properties of the water are dominated by phytoplankton absorptions of incoming solar radiation as comprehensively mentioned in the Appendix A. Due to the fact that Chl-a absorbs most radiation at shorter wavelengths (blue band) and very less in the green band, high reflectance in blue means less concentration in Chlorophyll, and vice versa. This is because as the Rrs can be assumed proportional to the bb / a and thus, when the Chl-a concentration increases, the Rrs at blue band decreases dramatically and the maximum Rrs shifts towards the green.
The influence of backscattering and
absorption of water constituents at blue and green band helps to determine the Chl-a concentration. For example, if the water is pure or phytoplankton dominated with low Chlorophyll concentration at 0.1 mg m-3, the backscattering at the blue band is high and low absorption by phytoplankton at green band resulted in high ratio of blue-green band. When the concentration of Chlorophyll is high, the absorption at green band decreased and the backscattering at blue band decreased and resulted in a low reflectance ratio of the blue-green band. As mentioned in the Appendix A where the absorption of yellow substance also strong in blue band, so if the seawater is dominated by gelbstoff along with phytoplankton, it is hard to distinguish because of the spectral response of the two constituents at the wavelength of 440-550nm not
18 differ very much. Hence, if the concentration of yellow substance in the water is higher than the chlorophyll concentration, it resulted in increasing the absorption at blue band, then the blue-green ratio increased and causing the overestimation of the Chl-a concentration.
Therefore, the bio-optical algorithm developer starts to use the blue-green band ratio to retrieve the Chl-a value. An overview of this band ratio relation with Chl-a concentration is illustrated in Figure 2.1 where the reflectance of water dominated by phytoplankton, gelbstoff and sediment at blue to green wavelength with different Chl-a concentration is presented.
Figure 2.1 Spectra profile of remote sensing reflectance (based on Bricaud et al., 2004) for phytoplankton in the range of blue to green. For Case-1 profile the lower blue-green ratio presenting the higher Chl-a retrival, while for Case-2 profile the higher blue-green ratio reflecting the overestimation of Chl concentration in the algorithm.
19 In bio-optical algorithm development, the most common algorithm model used is the empirical model.
The empirical algorithms are derived from the
regression of in-situ Chl-a measurement and radiometric observations of Rrs at several wavelengths. There are algorithm that uses two-band empirical (Tzortziou et al., 2007; Moses et al., 2009), three or four-band empirical such as OC3M in MODIS (Ocean Chlorophyll 3-bands for MODIS) or OC4v4 in SeaWiFS (Ocean Chlorophyll 4-bands version 4) (Le et al., 2013), three-band semi-analytical (Gons et al., 2002; Cannizzaro & Carder, 2006) models and other different spaceborne sensors with their algorithms for Chl-a retrieval.
An attempt to develop an ocean colour algorithm for Case-2 water was conducted by Siswanto et al., (2011) and the scope of their study is using SeaWiFS algorithm (OC4v4). The study was carried out with the goal on the development of empirical ocean colour algorithms to retrieve ocean colour products in East China Seas and the results of using the modified OC4v4 has retrieved Chl-a with better accuracy compared to the standard algorithm OC4v4. Tilstone et al. (2011) has validated three Chl-a algorithm for SeaWiFS in coastal area of Bay of Bengal and open ocean area of the Arabian Sea. The empirical algorithms OC4v6, OC5 and Carder semi-analytical algorithm were tested in their study. The use of standard empirical and semi-analytical (OC4v6 and Carder) algorithms tended to overestimate Chl-a in the coastal area by factor of 2-3 and the empirical OC5 has performed better and preferred to use in coastal and open ocean water of their study area. This indicates that even using the semi-empirical algorithm, the overestimation of Chl-a can still exists. Moreover, study by Pan et al. (2010) has made evaluation of SeaWiFS and MODIS-derived Chl-a concentrations in Northern South China Sea (NSCS) and also resulted in overestimation of Chl-a (~77%) by global algorithms (OC2v4, OC4v4 and OC3M) due to oligotrophic water (Chl-a concentration < 0.1 m mg-3). Thus, a local algorithm was developed for the study area which resulted in better performance estimating Chl-a during summer season with low RMSE (0.25~ 0.28).
20 In regionally tuned standard algorithm, there are other technique than using non-linear regression that were practiced by other researcher (basic description about Chl-a algorithm calibration can be refer in Appendix A). In Werdell et al. (2009) study, they had derived the local Chl-a algorithm by using spectral-matching technique and resulted in less bias for both sensor MODIS and SeaWiFS. Other than that, Mishra and Mishra (2012) has proposed a normalized difference chlorophyll index (NDCI) to estimate the Chl-a concentration from remote sensing data in estuarine and coastal water. The research findings imply that NDCI successfully estimate the Chl-a on the MERIS data at the very turbid water and the application of the algorithm is quite similar to the NDVI used for terrestrial vegetation. Moreover, the usage of green-red wavelength such as synthetic chlorophyll index (SCI) (Le et al., 2013) in deriving the Chl-a concentration that works well for MERIS but not for MODIS in the study area such as turbid estuaries. In the case of clear water or oligotrophic and the ground truth is limited, the method introduced by Hu et al., (2013) has the advantage of estimating the uncertainty of the Rrs from MODIS and SeaWiFS, but this approach need the ground truth Rrs. The development of the multilayer perceptron (MLP) neural network algorithm for deriving the Chl-a and the definition of their range of applicability through the novelty detection technique was describe in the study by Alimonte et al., (2003). Table 2.1 listed the previous works that contributes to the Chl-a algorithm development in Case-1 and Case-2 water for global and regional area.
21 Table 2.1 Summary of the previous studies on the standard NASA Chl-a algorithm development and evaluation in local region. Reference
Type of water
Siswanto et al. 2011
Case-1 and Case-2
Tilstone et al. 2011
Case-1 and Case-2
Lapucci et al. 2012
Moore et al. 2009
Mishra & Mishra, 2012
Le et al. 2013
Hu et al. 2013 Alimonte et al. 2003
Input variable In-situ Chl-a Yellow Sea & SeaWiFS Rrs East China Sea In-situ Rrs Bay of Bengal In-situ Chl-a & SeaWiFS Rrs Arabian Sea In-situ Rrs Site
Lingurian & In-situ Chl-a Oligotrophic North MODISA Rrs Tyrrhenian Sea Case-1 and Case-2
Global ocean
Global In-situ Chl-a (NOMAD) MODISA Rrs
Chesapeake Bay, Delaware In-situ Chl-a Case-2 Bay, Mobile (for validation (Turbid) Bay & only) Mississippi MERIS Rrs Delta In-situ Chl-a MODISA Rrs Case-2 Tampa Bay MERIS Rrs (Estuary) Rrs (λ) lineheight North Atlantic SeaWiFS Rrs Oligotrophic & MODISA Rrs Case-1 South Pacific Rrs true Mediteranean Case-1 & Sea, North In-situ Chl-a Case-2 Sea & Baltic SeaWiFS Rrs Sea
Algorithm
Type of algorithm
OCChl MOCChl
Empirical
OC4v6 Empirical & Carder SemiOC5 empirical OC3M MedOC3 Empirical OC5 SAM_LT OC3M
Empirical
NDCI
Bandsubtraction
OC2 OC3 OC4 SCI
Empirical & Bandsubtraction
OC4v4 OC3M
Empirical
MLP
Neural Network
Past studies by O’Reilly et al. (1998); Blondeau-Patissier et al. (2004) and Siswanto et al. (2011) had used the Type II regression or Model-II regression to perform any iterative fitting routine to modify the standard empirical algorithms. This type of regression model was used as it relatively minimizes the areas between the data points and the line of best fit. The empirical model produced by regression method is subject for error fitting. In previous study, it was pointed out that the nonlinear regression programs generally estimated the standard error of the parameters,
22 but this errors are neither additive nor symmetrical, and exact confidence limits cannot be calculated (Motulsky & Ransnas, 1987). Thus, the reported standard error has underestimated the true uncertainty of any nonlinear equation, but the extent of how much the underestimation was depending on the particular equation and data being analysed. In order to demonstrate the certainty of the values determined by a nonlinear regression, Monte Carlo method was used and from the comparable result conducted between nonlinear and linear method (Xiao et al., 2011), they suggested that nonlinear regression method is preferable. Hence, it is necessary to carefully select equation or iteration technique option when using the nonlinear regression program. More comprehensive explanation of the regression analysis and non-linear regression method is described in Appendix A.
2.4
Spatial Variability Bias In Chlorophyll-A Estimation
Accurate assessment of Chl-a concentration by remote sensing is challenging in the optically complex or turbid Case-2 water. The locally-tuned Chl-a algorithm has been developed in many study area by previous researcher with the main goal to have better accuracy in Chl-a estimation than the Chl-a retrieved by the standard global-developed Chl-a algorithms.
The accuracy in Chl-a estimation could be
induced by many factors that relate with the spatial variability of the Chl-a such as the in-water constituents, seasonal or monsoon effect at the study area and the satellite data spatial resolution. A study by Zimba et al. (2006) found that the assessment of the model accuracy in the hyper-eutrophic water is hampered by spatial and temporal variation in Chl-a concentration by 20% and 8%, respectively. This indicate that data used as “ground-truth” can vary significantly in time and space and might not represent constituent concentrations during the measured reflectance. Besides, the sampling procedure can significantly change constituent concentrations resulting in poor correlations with the remotely measured reflectance.
The high spatial variability at coastal region was also caused by the in-water constituents that influence the water optical properties. A study by Paudel et al. (2016) has stated that the environmental variables such as river discharge that drives
23 the salinity and organic matter into the estuaries can cause variation in phytoplankton growth and then the variability in Chl-a. Moreover, phytoplankton patchiness caused by monsoon wind was partially responsible for unexplained Chl-a variation (e.g. Tan et al. 2006).
Bio-optical profiles (e.g. Chl-a, diffused attenuation at 490nm (K490), CDOM and TSM) measured at a single station are representative of a spatial scale that is only a small fraction of a kilometre.
The use of window kernel when
extracting data at a location is required to estimate the spatial average of the optical properties represented by the satellite pixels. The bio-optical measurements by ship over the grid are usually not instantaneous and temporal variability in bio-optical properties can add additional uncertainty. In the case of static station, the bio-optical measurement is continuous because the measurement does not have to move from station to another station. Previous study has demonstrated that finer resolution is required for validation of coastal products in order to improve match ups of in situ data with the high spatial variability of satellite properties in coastal regions (Ladner et al., 2007). The Chl-a concentration is homogeneous within the 0.1m2 water column as compared to the averaged MODIS Chl-a concentration in ~1km2 water columns. This may lead to low degree of pixel-to-point correlation between satellite pixel and in-situ point of measurement location. Ocean color data with high spatial resolution and radiometrically correct could give better Rrs of the water optical properties (Patissier et al. 2004; Gregg et al. 2009). The degree of bias in spatial variability during the Rrs extraction is unknown because there has not been discussed yet in the previous studies.
2.4.1
MODIS-Retrieved Chl-a Empirical Algorithm
The MODIS retrieved Chl-a empirical algorithm known as OC3M employs the maximum of blue to green band ratio and the empirical algorithm OC3M by using the coefficient of OC3M version 4. The OC4v4 for SeaWiFS uses four bands, while the OC3M algorithm is based on three bands because MODIS does not have 510nm band as used in the OC4v4 SeaWiFS algorithm, instead MODIS used 551nm which is more sensitive to the absorption peak.
The MODIS OC3M is the
24 replacement for the CZCS and SeaWiFS empirical algorithms that used to measure the Chl-a. The previous OC4 used by SeaWiFS has given better fitting using more bands in the algorithm, but not all the bands were used when estimating the Chl-a. So, they pick the maximum band to estimate the Chl-a and by using the maximum band ratio (MBR), low Chl-a concentration was retrieved.
More details of the algorithm version from previous to the latest calibrated by the OBPG can be found in the post launch calibration and validation analysis report from the ocean colour website. Previous study reported that the Chl-a retrieved utilizing the OC3M algorithm has underestimated the in-situ Chl-a at concentrations below than 1mg m-3 and overestimation at higher concentration (Dierssen, 2010). Although the global Chl-a product with high resolution derived by OC3M is available at no cost, it is widely accepted that OC3M only working well in retrieving Chl-a particularly in Case-1 water. While in the Case-2 water such as coastal water and river estuary, OC3M algorithm fails to reliably retrieve Chl-a, due to non-trivial amount of suspended sediment and CDOM (Moses et al., 2009). This was supported by Wang et al. (2010) in their study, where they had retrieved and validated the ocean colour products around the China coast using MODIS, and the results proved high overestimation for Chl-a at the coastal region.
2.4.2
MODISA R2010 and MODISA R2013 Data
Ocean Biology Processing Group (OBPG) has pre-processed all the MODIS (Terra and Aqua) data to produce the Level 2 and 3 products for users and distributed by the NASA Goddard Space Flight Center’s Ocean Data Processing System (ODPS), which can be accessed from the NASA Ocean Colour Website. In March 2013, OBPG has done the new calibration for MODISA bands 8 and 9 (412nm and 443nm) and reprocessed the previous MODISA (R2010). The R2013 variant employs the calibration of MODIS Aqua bands 8 and 9 (412nm and 443nm) to adjust the temporal trends in the response versus scan angle (RVS). This temporal adjustment which could not be fully characterized by the on-board calibration (i.e., based on lunar and solar attributes) was derived by cross-calibration with SeaWiFS (Ocean Colour Webmaster, 2012). Unfortunately, the SeaWiFS mission was ended
25 in late 2010 due to a terminal spacecraft anomaly. The objective of MODISA R2013 is to implement a new instrument calibration that is fully independent of the SeaWiFS mission.
2.5
Chlorophyll-A Variability Study In The Malacca Strait
Since there are many previous studies (concluded in Table 2.2) found that the Chl-a retrieval in the coastal and turbid water region by using the standard empirical algorithm is problematic due to high uncertainties, the need to develop local Chl-a algorithm for coastal water or Case-2 waters is essential. Some cases of regional water are optically complex Case-2 or oligotrophic water. To this present, the global algorithms for ocean colour remote sensing do not always provide a reliable retrievals in all areas of the ocean, due to the fact that the empirical algorithm is only as good as the data located on (Pan et al., 2010). As our ocean is highly dynamic, the outcome from the river runoff influences the optical properties of the water. Geographically different water bodies led to different variations of their optical properties (Mobley et al., 2004), and hence, calibration and validation is required to access the accuracy of ocean colour algorithm and later rectify the Chl-a estimates at the local scale. Besides, some discrepancy on the validation is subject to the impact of environmental aspects (e.g., monsoon effect and other geophysical parameter) and lack of availability of satellite and in-situ data.
26 Table 2.2 List of previous study of Chl-a retrieval in the marginal sea. Reference Dall’Olmo et al. 2005 Bierman et al. 2009
Accuracy of estimated Chl-a Relative random uncertainty ~28% 54% of uncertainty
Uncertainty of 10-50 % 40-100% Overestimate Pan et al. 2010 77% Chl-a < 0.1 mg m3 Overestimate Chl-a < 5mg m-3 Vazyulya et Underestimate al. 2014 Chl-a > 5 mg mWerdell et al. 2009
Satellite Mission
Type of water
Factor influence Chl-a inaccuracy
SeaWiFS MODIS
Turbid productive
Other in-water constituents
MODIS
Case-1 and Case-2
Water depth and atmospheric correction
SeaWiFS MODIS
Case-2
Radiometric error
SeaWiFS MODIS
oligotrophic
Summer season and other in-water constituents
MODIS
Case-2
Atmospheric correction and river runoff
3
Study of biological production estimation in Malacca Straits is significant as one of the main fishing ground area for Malaysia and several studies has been done in Malaysian water to understand the spatial and temporal variability of the phytoplankton and the seasonal Chl-a variation affected by the geophysical factors which is significant to assist in fisheries management and to detect harmful algal bloom. Study by Tan et al. (2006) had verified the Chl-a retrieved by SeaWiFS and the result showed that chl-a in Malacca Strait experienced seasonal variability and overestimation of Chl-a in the south part of the Malacca Strait is due to the high turbidity. Moreover, the study that investigates the spatial and temporal distribution of SeaWiFS derived-Chl-a in the Malacca Strait also relate the Chl-a variation with the seasonal factor and sea surface temperature (SST) distribution (Ali Yousif, 2009). Other study has validated the MODIS data along the east coast of Malaysia and comparing algorithm for mapping Chl-a that eventually suggested Aiken’s algorithm (Marghany & Hashim, 2010) as the best tool to use. However no effort has been made to calibrate and validate the standard MODIS Chl-a algorithm OC3M for local scale application especially in Malacca Strait.
27 Previous studies has outline the standard Chl-a algorithm in global and local region, where most study given the overestimation of Chl-a analysed based on the systematic and random errors such as RMS error, but the percentage of the uncertainty of the Chl-a estimation has not being highlighted.
Studies in
oligotrophic and optically complex water showed that one of the major factors of the overestimation of Chl-a by standard algorithm is the atmospheric correction. The impact of spatial bias when extracting the Chl-a product or the Rrs has not been discussed in the previous study. A simple method to reduce the spatial bias has yet been introduced; instead they had suggested a more complex algorithm to derive the Chl-a concentration. Furthermore, in calibration of the algorithm, most study used cruise observation data where the in-situ data were collected well distributed in the study area. A case of static in-situ data with continuous observation has yet been discussed. In the previous study, none had validate and calibrate the standard Chl-a algorithm in the unique geographical location such as Malacca Strait where the water is eutrophic and consist of Case-1 and Case-2 water. Most of the research had highlighted seasonal impact on Chl-a variation in the Malacca Strait, thus a local Chl-a algorithm need to be developed.
28
CHAPTER 3
RESEARCH METHODOLOGY
3.1
Introduction
The methodology of this research is divided into three parts; 1) data acquisition includes types and source of satellite data, and the method of in-situ Chla acquisition, 2) satellite data pre-processing, and 3) main data processing consists of four parts; (i) spatial and temporal consideration, (ii) calibration and validation (verification of OC3M, local-tuned method and weight function method), (iii) accuracy assessment consist of Cal/Val routine with statistical analysis and lastly (iv) the application of the new locally-tuned Chl-a algorithm. Figure 3.1 shows the the main flowchart of the research methodology which highlighting the main process of the research and the sub-flowchart of the main process expanded comprehensively later in Section 3.4. The sub-flowchart of (a), (b) and (c, d, e) are going to be discussed in the following sections; 3.4 Calibration and validation exercise; and 3.5 Accuracy Assessment, respectively.
There are three exercises involved in this Cal/Val process; 1) verification of OC3M algorithm applicability in the Malacca Straits using the latest MODISA R2013 data; 2) modification of OC3M based on MODISA R2013 Rrs to improve the previous exercise of regionally-tuned Chl-a algorithm for the Malacca Straits and 3) the applicability test and map for Chl-a estimated by locally-tuned algorithms.
29
Figure 3.1 The main flowchart of the study consist of several sub-flowcharts that shows detail process of the method involved, (a) Figure 3.6; (b) Figure 3.7; and (c), (d), (e) in Figure 3.8.
3.2
Data Acquisition
In the preliminary study of this research, the satellite data MODISA R2010 has been used and due to the recent update from NASA Ocean Biology Processing Group (OBPG) where the reprocessed data of R2013 have replaced the previous reprocessed data with significant improvement in the band calibration. following sub-sections describe the data were used in this study.
The
30 3.2.1
Remote Sensing Data
Section 2.4.2 has previously mentioned the significant improvement of MODISA data version R2013 compared to the version R2010. The latest reprocessed data were used to understand the performance and how the improvement would affect MODIS Chl-a estimation. This study requires the Level 2 data product which can be accessed
freely
from
the
NASA
Ocean
Color
Website
(http://oceancolor.gsfc.nasa.gov/). Level 2 data product was generated from Level1B where the data have been applied with the instrument or radiometric calibration (description of data level can be found in Appendix A) and all contents are being stored in the Hierarchical Data Format (HDF).
The Level 2 product contains
numerous metadata to characterize the quality of each retrieved pixel values, such as the date, time, and coordinates of data collection. Additionally, numerous flags are provided to indicate if any algorithm failures or warning occurred for each pixel of the scene. The ordered data product for this study contains the Chl-a estimated by OC3M, the Rrs at ten ocean colour wave bands (refer Table 3.1) and the l2_flags band. The l2_flags band contains quality flag information such as listed in the Table 3.2 and in this study, the LAND flag was used to mask out the land in the data.
Then, by using SeaWiFS Data Analysis System (SeaDAS) version 6.3 software, the value of Rrs was derived from each data of daily MODISA Level 2 data where the period of study spans from 14th October 2011 to 10th August 2012. The study period is based on the time where the in-situ data collection were performed. The level 2 data products have been calibrated and corrected atmospherically and geometrically by the OBPG. Atmospheric correction method designed by Gordon & Wang (1994) was used. Next, the Rrs values were used as an input to the Maximum Band Ratio (MBR) in the OC3M algorithm which is a fourth degree polynomial regression in log-transform (O’Reilly et al., 2000) as follows
2 3 4 Ca 10C0 C1 R C2 R C3 R C4 R
where
(3.1)
31
Rrs 443 Rrs 488 R log10 max , Rrs 547 Rrs 547
(3.2)
the log-transform Chl-a concentration is denoted by Ca , superscript C0,C1,C2,C3, and C4 are the polynomial coefficients, and R is the MBR with the Rrs from several wavelength bands as input. The MBR method is widely used and the standard OC3M algorithm also using the Rrs derived from MBR technique. The limitation and challenges of using the band-ratio algorithm such as OC3M was explained in the previous chapter, hence the performance of this algorithm in the coastal region was determined in this study.
Table 3.1 Data specification of Level 2 product used in this study. Properties Study period Area of interest
Data parameter 2 October 2011 to 10 August 2012 9.0N and 0.0N ; 97.0E and 105.0E Chlorophyll-a
Units
mg m-3
Parameter extracted from Level 2 data Rrs412, Rrs443, Rrs 469, Rrs488, Rrs531, sr-1 product Rrs547, Rrs555, Rrs645, Rrs667, Rrs678
Table 3.2 Description of l2_flags of MODISA level 2 data, the list can be accessed from the data product through SeaDAS. Name ATMFAIL LAND PRODWARN HIGLINT HILT HISATZEN COASTZ SPARE8 STRAYLIGHT CLDICE
Description Atmospheric correction failure Land One (or more) product algorithms generated a warning High glint determined High (or saturating) TOA radiance Large satellite zenith angle Shallow water (0.5) as the Chl-a variation in ST1 and ST2 is higher and the number of available match-up sample is insufficient (the best is more than 30). Yet, the positive correlation is essential in order to retain positive correlation in evaluating the local OC3M algorithm. Furthermore, a higher R2 is insignificantly indicates that the model is a good fit. This has been tested earlier in each Cal/Val exercise; when the objective function is set to have maximum R2 in the fitting routine, the results of R2 were all high but it showed systematically over and under-predicts the data and the algorithm coefficients turned out extreme values. Besides, in non-linear regression (considering the form of Eq. 3.1), the R2 value is best determined by calculating the proportion of the total variation in the observations that cannot be explained by the fitted model and subtracting this proportion from one; and the standard formula of R 2 used in linear regression case is inappropriate in the nonlinear case (Cornell and Berger, 1987).
62 Now that we know the statistical evaluation of the local-tuned OC3M with and without WFd, a suitable and applicable algorithm was chosen to map the Chl-a in Malacca Strait water. From section 4.2 until this section, the first and third objective of this study has been achieved and the next section will discuss how the best algorithm was chosen among the sets of exercises and the test of applicability was performed to see the local-tuned algorithm acceptable area in the Malacca Strait and the nearby areas. .
63
Figure 4.5 Plot of regression between MODIS derived Chl-a and in-situ Chl-a using standard, local and local with weight function distance in dual-case, Case-1 and Case-2 water respectively. The red, blue and dashed line indicates unity, Type-II regression fit and general regression fit, respectively. Error bars are designed by the standard deviation and the average Chl-a of the sample points. The result is basically produced by 63
the kernel size of 3×3.
64 4.6
Local Algorithm Applicability
From the previous section, the suitable kernel size for optimum used in Cal/Val has been determined. This section discusses on the decision of choosing the best locally-tuned algorithm in terms of the algorithm reliability and its application for Malacca Strait region. There are four characteristics used to determine good algorithm, which are 1) the sum square of error (SSE) is reduced after the localtuned being employed; 2) the slope=1 or closer and intercept=0; 3) the confidence interval (CI) of the regression should be smaller than before the algorithm was being locally-tuned and 4) the goodness of fit test (Gof) should have positive value. Referring Table 4.5, the performance of the local-tuned OC3M algorithm in terms of SSE, slope, intercept, CI and Gof can be compared for data with k3 and k7. Table 4.5 shows that the results from k3 produced four out of six positive Gof compared to the results from k7. Hence, to determine which algorithm is suitable for application in the Malacca Strait and the water cases alike, the algorithms of data using k3 were employed in mapping the climatology Chl-a in this study area.
Based on the statistical performance of the iterative fitting for combined and separated cases, the local OC3M algorithm worked better in separate water cases (Case-1 and Case-2). This is because when the samples of Case-1 and Case-2 waters were combined for local-tuning, the Gof was unsatisfactory due to massive difference between the average value of Chl-a in Case-1 and Case-2.
So, the
estimation of Chl-a in dual-case is less accurate and tends to underestimate the Chla. Thus, to map the Chl-a for the application of the local-tuned algorithms, the OCms1 and OCms2 were used for mapping the OCms Chl-a, while the OCms1 WFd and OCms2 WFd were used to map the OCmsWFd Chl-a (refer Figure 4.7 and 4.9, respectively).
65 Figure 4.6, 4.7 and 4.8 are the climatology of Chl-a from 2009 until 2013. The OC3M-retrieved Chl-a (Figure 4.6) shows higher Chl-a value along the coastal area, especially at the south part of the Malacca Strait. The Chl-a retrieved shows significant differences from the in-situ value, where the Chl-a value is higher than the average in-situ Chl-a at ST2 (2.76 mg m-3) and each month presents the highest Chl-a at ST2 (9.53 mg m-3). During the Northeast monsoon months (Nov- Mar), the Chl-a retrieved by standard OC3M algorithm indicates very high Chl-a concentration along the Malacca Strait, and during the Southwest monsoon months (May-Sept), there are only 3 months (July, August and September) showing very high Chl-a concentration.
Figure 4.6 Map of climatology OC3M-retrieved Chl-a from 2009 until 2013.
66 By applying the local-tuned OC3M algorithm (OCms and OCmsWFd), the Chl-a value presented in the climatology map presents the overestimation of standard OC3M derived Chl-a in the south part of Malacca Strait has been reduced but very high Chl-a is existed along the coastal water that is mapped in the red spot along the Malacca Strait in Figure 4.7.
In other region than the coastal area, the Chl-a
concentration is higher and most of the open ocean water given the Chl-a value at ~ 1 mg m-3. When applying the local-tuned algorithm with WFd, higher Chl-a at coastal water is reduced. The Chl-a map produced by OCms WFd (Figure 4.8) has shown promising improvement where the overestimation along the coastal water was reduced and the Chl-a at the open ocean water is within the average and minimum value of in-situ Chl-a at ST1. The very high Chl-a which is above 10 mg m-3 is coexisting in the map of OCms WFd but only on certain months with monsoon.
Figure 4.7 Map of climatology Chl-a retrieved from locally-tuned algorithm, OCms from 2009 until 2013.
67
Figure 4.8 Map of climatology Chl-a retrieved from applying locally-tuned algorithm, OCms WFd from 2009 until 2013.
To study the reliability of the local-tuned algorithm, a p-value map was produced from the climatology of Chl-a. Figure 4.9 and 4.10 show the p-value map of the Chl-a estimated by OCms and OCms WFd algorithms. Back to the Chapter 3 where the null and alternative hypothesis was set, the significance level of p-value in the map was set as black for p ≤ 0.05 and white for p ≥ 0.05. The p-value map for OCms algorithm shows the null hypothesis was accepted for area that the Chl-a value is greater than 10 mg m-3 and rejects the null hypothesis at the area where the Chl-a value is smaller than 10 mg m-3. This can be seen from the Chl-a climatology map of OC3M and OCms, where the areas that have very high Chl-a (≥10 mg m-3) are remaining at almost the same area. Only a small part of the area which has very high Chl-a in the OC3M map was reduced by the OCms algorithm, but the area with Chl-a less than 1.0 mg m-3 also been changed to value that is closer to 1.0 mg m-3.
68 This shows that OCms algorithm insignificantly reduce the overestimated Chl-a pixels in the OC3M map. This could be due to the APD of OCms1 that is higher than the target APD (35%) and also combining with the OCms2 that resulted to irrelevant the Chl-a estimate as the larger dynamic range of Chl-a from the Case-1 and Case-2 water. In the p-value map for OCmsWFd, the null hypothesis was accepted for area with the Chl-a lower than 10 mg m-3 and above than 0.1 mg m-3, and reject the null hypothesis for the area with the Chl-a greater than 10 mg m-3 and below than 0.1 mg m-3. Based on the minimum in-situ Chl-a which is 0.588 mg m-3 in the Case-1 water and 0.704 mg m-3 in the Case-2 water, the used of OCmsWFd algorithm can reduce the Chl-a overestimation and underestimation of in-situ Chl-a by OC3M.
69
Figure 4.9 The p-value map from the climatology Chl-a retrieved by using OCms algorithm. The black colour in the water area of the map is for significance level of p ≤ 0.05 and white for p ≥ 0.05.
70
Figure 4.10 The p-value map from the climatology Chl-a retrieved by using OCms WFd algorithm. The black colour in the water area of the map is for significance level of p ≤ 0.05 and white for p ≥ 0.05.
Besides the p-value map, the mean Chl-a of the climatology were extracted from 3 sections of Malacca Strait (north, middle and south) and plotted in the graph with its standard deviation value as the error bar (Figure 4.11) to see in details whether the Chl-a from the OC3M was improved when using the local-tuned algorithms. The Chl-a trend of the 3 sections for every month were plotted, the mean Chl-a trend at the north part of Malacca Strait is consistent around value
