Home Random Page


CATEGORIES:

BiologyChemistryConstructionCultureEcologyEconomyElectronicsFinanceGeographyHistoryInformaticsLawMathematicsMechanicsMedicineOtherPedagogyPhilosophyPhysicsPolicyPsychologySociologySportTourism






Satellite oceanography: New results 2 page


 

  Option no. Input layer Hidden layer 1 Hidden layer 2 Output layer   chl (mg m-3) sm (g m-3) doc (gC m-3)
  0-70 0-20 0-20
  0- 0-15 0-15
  0-70 0-5 0-20
  0-70 0-20 0-5

 

Table 6.2. The architecture and respective retrieval errors for a set of built-up NNs. Net structure/number of neurons CPA concentration ranges

 

 

(Table 6.2, option no. 1) for a wide range of the CPA concentration vectors has been analyzed.

Comparing Figures 6.1 with 6.2, the retrieval accuracy of the built-up NN (option no. 1) proves to be inferior to the retrieval accuracy achieved with the L-M procedure. It is especially so in the range of small and high concentrations of CPAs. However, the retrieval error can be further reduced if the initially deter- mined CPA concentration values are then used for a judicious choice of a NN trained for less wide CPA concentration ranges of each of the CPAs (Table 6.2). Further, using such a specialized NN in the second step, it is possible to determine more accurately each of the CPA concentration.

Determined promptly and with reasonable accuracy, by means of the above set of NNs, the CPA values can eventually be considered as C0 values for initiating the L-M procedure. This combination of the L-M and NN techniques significantly improves the performance of the compound tool in terms of both the retrieval accuracy and computation time. The concept of this approach is illustrated in more detail in Figure 6.3.

Ignoring some other sources of input data noise (e.g., small portions of sun glint, stray light within the sensor, etc.), the inevitable imperfection of atmospheric correc- tion of satellite images results in the processed data being subject to distortions in terms of both their spectral distribution and actual magnitudes at each wavelength. Our analyses of the reported data (e.g., Land and Haigh, 1996; Ruddick et al., 2000) indicate that the atmospheric correction error can reach ± (10-15)% in the blue part of the spectrum, declining towards the red portion of the spectrum. The distribution of the error occurrence might be either normal or uniform. Following Schiller and Doerffer (1999), we assumed a normal distribution of noise occurrence. Accordingly, the CPA concentrations determined with the specialized NNs (Figure 6.3), were then turned into respective concentration ranges with widths corresponding to a ± 15% departure from the tentatively retrieved chl, sm, and doc. The values of chl, sm, and doc belonging to these established concentration ranges were then used, in a random order, as the components of C0for the initiation of the L-M procedure. The Lake Ladoga hydro-optical model (Kondratyev et al., 1990) was employed as being typical of moderately eutrified waters at temperate


6.1



 

 

Simulated (a)

 

Simulated (b)

 

(c)

Figure 6.2. Same as Figure 6.1 but utilizing a NN simulator.

 

latitudes. The relationship between R and a, bb suggested by Jerome et al. (1988), was used for the analyses of the developed algorithm.

Table 6.3 exemplifies the increase in the CPA retrieval accuracy due to the sequential use of a rough NN (i.e., the one trained for a wide concentration range for all CPAs, see option no. 1, Table 6.2) in combination with the special- ized NNs (i.e., option nos 2-4, Table 6.2).


 


6.1



Table 6.3. Improvement of the CPA retrieval accuracy (in %) due to a sequential use of a rough NN in combinations with specialized NNs for the CPA concentrations falling in the ranges 0-5mg m-3for chl, 0-5g m-3for sm, and 0-5gC m-3for doc.

 

  CPA rough NN (option 1, Table 6.2) specialized NNs (options 2-4, Table 6.2)
chl
sm
doc

 

 

Table 6.4. Improvement of the tool performance in comparison with the facility of the L-M procedure per se.

 

  L-M procedure Advanced compound tool Relative improvement
sm (mg l-1) (s/pixel) (the L-M and NNs) (s/pixel) (L-M/L-M + NNs)
::; 0.5 0.13 3.8
5 ::; 20 1.3 0.6 1.9

 

 

Table 6.5. Admissible errors in the retrieval of chl in case II waters.

 

range of chl (mg m-3) 0-1 1-2 2-5 5-10 10-20 20-50
admissible error (%) ± 50 ± 30 ± 30 ± 30 ± 20 ± 20

 

 

Table 6.4 illustrates the enhancement of the operational efficiency (the time required for processing one pixel) of the developed tool in comparison with the sole L-M procedure for a couple of options of sm concentrations. According to our numerical assessments the level of sm content in water proves to be a controlling factor: the operational efficiency of the tool increases with decreasing sm.

The data in Table 6.4 are explicitly indicative of a significant advantage that the developed tools offer in terms of increasing the operational efficiency of retrieval of water quality parameters in optically complex aquatic environments.

In the following, we now consider the feasibility of the advanced tool under the conditions of different CPA composition options of natural waters.

Realistically assessing the attainable (and at the same time admissible) retrieval accuracy when processing remote sensing data taken over case II waters, we assumed some error limits for the retrievals of chl in turbid and absorbing waters (i.e., case II waters) (Table 6.5). (These error limits were based on both our own experience and the requirements set up by potential end-users (e.g., Durand et al., 1998).)

Given the predefined admissible errors in the retrieval of chl (Table 6.5), it is possible to determine, for the developed CPA retrieval tool, the concentration ranges


 

Table 6.6. Concentration ranges for sm and doc given a 10% uniform ,\-independent noise.

 

sm (g m-3)   doc (gC m-3) 0-0.5 0.5-1 1-5 5-10 10-20
0-2 0-50 0-50 0-50 0-50 0-2
2-5 0-50 0-50 0-50 0-10 0-1
        20-50  
5-10 0-50 0-50 0-50 0-50 10-20
10-20   5-10 0-50 2-10 2-5
  20-50 0-50 20-50 20-50

 

 

Table 6.7. Concentration ranges for sm and doc, given a 15% uniform ,\-independent noise.

 

sm (g m-3)   doc (gC m-3) 0-0.5 0.5-1 1-5 5-10 10-20
0-2 0-50 0-50 0-20 0-10 0-5
2- 0-20 0-20 0-20 0-10 0-5
5-10 2-20 0-20 0-20 0-1  
        2-10  
10-20     0-10 5-10  

 

 

Table 6.8. Concentration ranges for sm and doc, given a 10% uniform ,\-dependent noise.

 

sm (g m-3)   doc (gC m-3) 0-0.5 0.5-1 1-5 5-10 10-20
0-2 0-10 0-10 2-5 2-5 1-2
2-     0-2 0-2  
5-10     0-1 0-1  

 

 

within which this is attainable under different scenarios of the magnitude, type of occurrence, and wavelength dependence (viz. independence from ,\ or decreasing linearly with ,\) of input data. Tables 6.6-6.10 illustrate the results of our numerical simulations.

The data given in Tables 6.6-6.10 indicate that the accuracy of retrieval results is highly dependent on the level and type of noise imposed upon the input data. These considerations should always be taken into account when appraising the adequacy of results of CPA retrievals from space images taken over case II waters.


6.1



Table 6.9. Concentration ranges for sm and doc, given a 15% uniform ,\-dependent noise.

 

sm (g m-3)   doc (gC m-3) 0-0.5 0.5-1 1-5 5-10 10-20
0-2 0-5 0-5 1-2 1-2 1-2
2-   0-2 0-2 1-2
5-10   0-1 0-1  

 

 

Table 6.10. Concentration ranges for sm and doc, given a 15% normal ,\-independent noise.

 

sm (g m-3)   doc (gC m-3) 0-0.5 0.5-1 1-5 5-10 10-20
0-2 0-50 0-50 0-50 0-50 0-1
          2-50
2-5 0-50 1-50 0-50 0-50 0-50
5-10 0-50 0-50 0-50 0-50 0-50
10-20 0-50 1-50 2-50 0-50 2-50

 

6.1.2 Advanced tool (AT) performance verification and seasonal variations of White Sea CPA concentrations as revealed by SeaWiFS data

In Onezhskiy Bay, which is a shallow-water body with an average depth well under 10 m, measurements of some basic water parameters and CPA concentrations were made at 15 stations from on board the Northern Water Problems Institute (NWPI) research vessel R/V Ecolog. Figure 6.4 displays the location of the stations (and the associated chl concentration determinations in situ) performed in Onezhskiy Bay, while Table 6.11 illustrates the results of shipborne in situ measurements of CPAs. These data are clearly indicative of case II waters.

It should be underlined that the in situ determinations of concentrations of chl, sm, and dom are prone to some inaccuracies arising collectively from instrumental and/or methodological errors and errors due to the routinely used sampling procedure performed with the help of a bucket (resulting in a scatter in the series of determinations from one and the same bucket given a high degree of inhomo- geneity of CPAs in case II waters). The resultant error is assessed at about 10% for chl and sm and 5% for doc (Petrova, 1990).

Figures 6.5 and 6.6 illustrate the spatial distributions of water temperature (Tw), salinity (S) and chl-a in the surface waters of Onezhskiy Bay from in situ measure- ments (both at stations and underway). These data indicate that the conditions of algae growth (with the predominance of freshwater species assimilated into the


 

 

Figure 6.4. Location of stations and the determined concentrations of chl (µg l-1) in Onezhskiy Bay.

 

brackish aquatic environment) are most favorable within the coastal zone, where the water temperature is considerably enhanced and the salinity levels are low. Approaching the Solovetskiy Archipelago and, hence, open tracts of the White Sea, the content of algae decreases. This pattern is also prompted by a quasi-permanent upwelling (within which the water temperature is considerably cooler than it is in the ambient parts of the sea) residing around the Solovetskiy Archipelago.

Interestingly, Figures 6.5 and 6.6 also reveal a specific feature in surface dis- tributions of water temperature Tw and salinity S. There are two centers of enhanced water temperature and low salinity - in the area of the Onega River mouth and near a small town called Belomorsk, located southward off the inlet of the Kem River. From these centers, both Tw and S decrease and increase, respectively, along imaginary axes intersecting in the region of the Solovetskiy Archipelago.

Owing to the specific features discussed above, the spatial distribution of chl is


6.1



Table 6.11. Results of in situ measurements in Onezhskiy Bay, July 2001, surface layer. The shipborne data was obtained on 14 July.

 

chl (µgl-1) chl (µgl-1) (fluorometric (spectrometric

Station no. Date of sampling sm (mg l-1) doc (mg C l-1) method) method)

 

1 10.07 0.25 8.3 - 1.01*

 

10.07 0.25 5.8 0.91 -
10.07 0.65 5.9 - 1.12**
10.07 0.70 5.8 - -
5 10.07 0.30 6.3 1.82 -
6 10.07 0.10 5.3 1.45 -
7 10.07 0.35 6.6 0.59 -
11.07 0.50 4.9 - -
11.07 0.30 5.9 0.29 -
11.07 0.85 5.9 1.21 -
11.07 0.70 5.5 1.02 -
11.07 0.10 5.3 1.33 -
13.07 0.30 5.8 - -
13.07 0.60 5.6 - -

15 13.07 1.35 5.4 - 1.01*

 

* Concentration of chl-a; ** data obtained at a distance of 10 km from the station location.

 

rather uniform in Onezhkiy Bay and varies only between about 1-2 µg l-1, whereby the lower concentrations are found mostly south of the Solovetskiy Archipelago.

SeaWiFS images over the White Sea were taken concurrent to the shipborne CPA measurements. Figures 6.7-6.10 illustrate the results of CPA retrievals from the image taken over Onezhskiy Bay in July, 2001. As Figures 6.7-6.10 illustrate, the retrieval results obtained with the use of SeaDAS (Figure 6.7) and the AT differ significantly, although both SeaDAS and AT retrievals were performed both after atmospherically correcting the SeaWiFS images with the use of the MUMM procedure (Ruddick et al., 2000): utilization of the SeaDAS atmospheric correction code resulted in hopelessly poor restoration of water-leaving radiances in the blue end of the spectrum (mostly negative values of upwelling radiance).

Table 6.12 compares in situ and remotely assessed concentrations of chl, sm, and doc for the AT code, and solely chl in the case of application of the SeaDAS code with the Management Unit of the North Sea Mathematical Models (MUMM) atmo- spheric correction. The comparison strongly favors the AT code: the retrieved chl concentrations are well within the interval of chl values obtained in situ, whereas the SeaDAS-produced chl values are more than three times higher. In addition, the AT procedure provides retrievals of doc with departures from the respective in situ measured values of only 20-25%. However, the retrievals of sm show much larger divergences from the respective in situ values at some stations. Several reasons might lead to these inconsistencies both for doc and sm (as well as chl, although they are much smaller). First of all, in situ measurements refer to a point at the surface of the


 

 

Figure 6.5. Shipborne measurements of near-surface water temperature (oC) in Onezhskiy Bay during 10-13 July, 2001.

 

bay where the sampling was taken, whereas the SeaWiFS pixel measures

1.1 km2x1.1 km2, and thus the retrieved value is in reality a spatially averaged value of the desired quantity. Besides, our own experience in shipborne in situ measurements indicates that the desired quantity can vary by about 20% in a series of samples taken from one and the same bucket, when the samples are taken near the water surface.

Finally, it should be noted that according to Table 6.12, the AT retrievals of sm are invariably higher than their in situ counterparts. One of the plausible explana- tions of this fact is the existence of fine particles in the water of the bay - particles smaller than the filter pores. Consequently, perhaps a considerable amount of sm remained unaccounted for.

In addition, the reader should be reminded that we have used the hydro-optical model, developed not specifically for the White Sea but the one for Lake Ladoga,


6.1



Figure 6.6. Shipborne measurements of surface salinity (0) in Onezhskiy Bay during 10-13 July, 2001.

 

when processing the SeaWiFS images taken over Onezhskiy Bay. This definitely will contribute to retrieval errors. However, since Lake Ladoga and the Onega River (as well as the Severnaya Dvina and Mezen Rivers) share very similar catchments, and are geographically close, we could anticipate (and the above results seem to corro- borate this premise) that their hydro-optical models are indeed very close.

In summary, even given these inaccuracies, the AT code definitely provides more accurate data than SeaDAS for Onezhskiy Bay.

From the perspective of the nature of inaccuracies arising from the application of the standard OC4 retrieval algorithm, our analyses of the chl field retrieved with SeaDAS indicate that it gives erroneously high values of chl in areas with enhanced concentrations of either sm or doc. Indeed, according to the inherent water circula- tion patterns, the waters from the Kem and Onega Rivers move counterclockwise (see Section 2.2) along the coastline of Onezhskiy Bay, being rich in doc and sm, respectively. It is preferably in coastal areas where OC4 overestimates concentrations of chl.


 

Figure 6.7. SeaDAS-based retrieval of concentration of chl (mg m-3) from a SeaWiFS image taken over Onezhskiy Bay in July, 2001.

 

Figure 6.8. New tool-based retrieval of concentrations of chl (mg m-3) from a SeaWiFS image taken over Onezhskiy Bay in July, 2001.


6.1



Figure 6.9. Same as Figure 6.8 but for the sm concentration (g m-3).

 

Figure 6.10. Same as Figure 6.8 but for the doc concentration (gC m-3).


 

Table 6.12. A comparison of water quality variables retrieved for Onezhskiy Bay with the SeaDAS code and our advanced bio-optical algorithm (July, 2001) with the respective in situ

data.                
Date of chl-a doc sm Date of chl-a chl-a doc sm
in situ (in situ) (in situ) (in situ) SeaWIFS (SeaDAS) (AT) (AT) (AT)
measurements (µgl-1 ) (mgC l-1) (mg l-1) overflight (µgl-1 ) (µgl-1 ) (mgC l-1) (mg l-1)
10/07/01 1.6 8.3 0.25 10/07/01 3.7 1.3 6.5 0.8
10/07/01 1.1 5.8 0.25 10/07/01 3.0 1.2 5.5 0.7
10/07/01 1.1 5.9 0.65 10/07/01 3.2 1.1 4.5 1.1
10/07/01 1.5 5.8 0.70 10/07/01 4.0 1.5 4.0 1.0
10/07/01 1.8 6.3 0.30 10/07/01 4.2 1.6 4.0 1.0
10/07/01 1.8 5.3 0.10 10/07/01 5.4 1.7 3.9 0.9

Thus, the above verification experiment has testified in favor of the high cred- ibility of CPA retrieval results as obtained with the use of the AT code.

The developed and validated AT code, together with the MUMM atmospheric correction procedure, were further employed to obtain seasonal variations in the spatial and temporal variations of CPA concentrations throughout the White Sea. Figures 6.11-6.15 illustrate the dynamics of the chl concentration throughout the sea.

According to the obtained chl distributions, in 2001 the micro-algae community in the White Sea started developing within a very narrow coastal zone as early as April. In the first 10-day period of May this burst in temporal variations of chl declined, but in late June there was a massive development of chl in the central part of the sea as well as in the north-eastern area. Later, the water blooming dilapidates in these parts of the sea. However, in late August, a new cycle of chl- intensive growth begins, this time in the southern part, viz. Onezhskiy Bay, which is shallow and well warmed. Finally in October, the biological activity in the White Sea declines leaving only slight traces of phytoplankton presence in the western and southern littoral zones.

It is interesting to compare the retrieved sequence of phytoplankton develop- ment in the White Sea with the historical data reported in a recently published monograph (Berger et al., 2001). The White Sea is at least partially covered with ice for nearly half of the year. Although the nutrients are relatively abundant in the wintertime, low water temperatures and low levels of light availability result in miniscule concentrations of phytoplankton. However, with the increasing incident solar radiation in March, starts a phenomenon characteristic of the White Sea: an extensive growth of sea ice algae (Horner, 1985). The micro-algae begin growing on the ice-water interface, and the sea ice algae reach their maximum abundance by the end of March. In April, while the sea remains frizzed , the combination of still high levels of nutrients and increasing incident solar radiation brings about a burst or bloom of phytoplankton growth. This is also conducive to the melting of ice in zones neighboring the coastline, predominantly in the western and southern parts of the


6.1



Figure 6.11. Spatial distribution of chl (µg l-1) in the White Sea as retrieved from a SeaWiFS image of 7 April, 2001.

 

 

Figure 6.12. Same as Figure 6.11 but for 25 May, 2001.


 

Figure 6.13. Same as Figure 6.11 but for 21 June, 2001.

 


Date: 2016-03-03; view: 549


<== previous page | next page ==>
Satellite oceanography: New results 1 page | Satellite oceanography: New results 3 page
doclecture.net - lectures - 2014-2024 year. Copyright infringement or personal data (0.017 sec.)