Chapter 3 Validation of the estimated upper-ocean thermal structure in the Northwest Pacific Ocean

Chapter 3 Validation of the estimated upper-ocean thermal structure in the Northwest Pacific Ocean

3.1 The importance of validation

In Chapter 2, upper-ocean thermal structure is estimated using combination of satellite data and the 2-layer reduced gravity model. How accurate the model-derived upper-ocean thermal structure is in the NWPO is an important question to ask before applying these model-derived temperature profiles to study of the typhoon intensification. For this reason, this study uses in-situ measurements to validate the estimated; comparison to the NPANFS full ocean model is also made. The NRL’s NPACNFS model is a real-time North Pacific Ocean data-assimilating model with 26 sigma levels. The assimilated data including satellite-derived SSHA and Muti-Channel Sea Surface Temperature (MCSST) (Ko et al., 2003). The data assimilation is through the MODAS system (Modular Ocean Data assimilation System), currently the NPACNFS is considered as a one of the best ocean model in the NWPO.

In order to effectively validate the derived temperature profiles in the NWPO, the NWPO was separated into four geographical regions. These regions include two known eddy-rich zones, i.e. the North Eddy-Rich Zone (NERZ) and the South Eddy-Rich Zone (SERZ) (Yasuda et al. 1992; Qiu 1999; Yang et al. 1999; Roemmich and Gilson 2001; Hwang et al. 2004), Interval between the 2 Eddy-Rich Zone (IERZ), and the Southern part of the North Pacific Gyre (SNPG). Figure 3.1 shows the location of these four zones. In the two eddy-rich zones (NERZ and SERZ), both warm and cold eddies are frequently found throughout the year (Yasuda et al. 1992;

Qiu 1999; Yang et al. 1999; Roemmich and Gilson 2001; Hwang et al. 2004). The

NERZ (Yasuda et al. 1992) is located at the Kuroshio extension region (140º~170ºE,

30~40ºN). The SERZ (Qiu 1999; Yang et al. 1999; Roemmich and Gilson 2001;

Hwang et al. 2004) is located in the North Pacific Subtropical Countercurrent (STCC).

It is well known that eddy is one of the important factors affecting the upper-ocean thermal structure in the ocean (Yasuda et al. 1992; Qui 1999; Hong et al. 2000; Shay et al. 2000;), so validating the satellite-derived temperature profiles in these two eddy-rich zones is important. In this validation, the two in situ observational data sets used are the GTSPP (the Global Temperature-Salinity Profile Program) from the NOAA/NODC (www.gtspp.org) and the ARGO buoy data sets (Gould et al. 2004;

www-argo.ucsd.edu) data. Three parameters, D20, D26, and the TCHP are validated.

The validation period is taken from May to October of 2003, so as to cover the main typhoon season in the NWPO. A total of 2258 in-situ profiles were found in these four validation zones and their locations are shown in Figure 3.1. Since we only have 3-months (July to September) of NPACNFS data, the validation of NPACNFS profiles is performed only for these 3 months.

3.2 In situ data sets

3.2.1 The Global Temperature-Salinity Profile Program (GTSPP)

GTSPP is a cooperative international project; countries contributing to the project

are Australia, Canada, France, Germany, Japan, Russia and the United States. The

project is leaded by Canada’s Marine Environmental Data Service (MEDS). GTSPP

seeks to develop and maintain a global temperature and salinity resource with data

that are as up-to-date and of the highest quality as possible. The data in the GTSPP

database is generated by ships and buoy measurement from all regions of the world’s

oceans. Instruments are used to collect the data include thermistor chains on buoys,

expendable bathythermographs (XBTs), digital bathythermographs (DBTs), bottle

samplers, and CTDs (conductivity-temperature-depth sensors). MEDS operationally

gathers and processes the temperature profiles, and eventually transfers them to NOAA/NODC, who has the responsibility to maintain the database of GTSPP and provide online access the data. GTSPP mainly offers two modes of data; they are real-time and delayed-mode. The delayed-mode records are generally of higher resolution than the real-time records (Keeley et al. 2003). In this study, delayed-mode records are used for the validation. All data are downloaded from the NOAA/NODC database (ftp.nodc.noaa.gov).

3.2.2 The ARGO floats data set

ARGO is a global array of 3,000 free-drifting profiling floats that measure the

temperature and salinity profiles from the upper 2000m of ice-free global ocean and

currents from intermediate depth (Gould et al. 2004), which is a part of Global

Climate Observing System/Global Ocean Observing System (GCOS/GOOS). The

ARGO can continuously monitor temperature, salinity, and velocity of the upper

ocean with all data being relayed and made publicly available within hours after

collection. The first floats were launched in 1999; approximately, there have about

1890 ARGO floats were deployed in the global ocean in the moment. However, the

finial array of 3000 floats will provide 100,000 temperature and salinity profile

measurements each year over global oceans at every 3-degrees of latitude and

longitude (Gould et al. 2004). A typical distribution of ARGO array for a day is shown

in Figure 3.2. The principle of the ARGO floats is that the data come from

battery-powered autonomous floats that spend most of their lifetime to drift at a

constant pressure depth at about 2000db, at typically 10-day intervals. The floats rise

to the surface from the 2000m underwater with rising speed about 10 cm/s. In this

procedure, they can measure the temperature and salinity profiles. It takes about 6~12

hours at the surface to transmit data to satellite. After transmission, the floats will sink

back to the original pressure level and go on to its next cycle. Figure 3.3 illustrates the principle of the ARGO float. This set of data was downloaded from the Global Ocean Data Assimilation Experiment (GODAE) (www.usgodae.org/cgi-bin/argo_select.pl).

3.3 Results

3.3.1 Validation of the depth of the 20 ℃ sotherm i

D20 was the first parameter estimated from the Two-Layer reduced gravity ocean Model (herein after refer as TLM_NWPO) and SSHA. It is the first parameter needed to be validated before validating D26, because D26 was derived based upon D20.

With incorrect D20, D26 will be subsequently incorrectly estimated. For validation, each comparison pair between the in situ and estimated data is chosen at the same location within 0.5 degree radius and simultaneous day. Figures 3.4-3.7 shows the comparison between estimated and observed D20 for each month between May to October 2003. It can be seen in Figure 3.4 that the estimated of D20 is generally bad in the NERZ for all 6 months. Evident overestimation appear in the positive SSHA regions (e.g. warm eddies) and underestimation in the negative SSHA regions (e.g.

cold eddies). In contrast, much better estimation of D20 in the other three regions (IERZ, SERZ and SNPG) are found (Figure 3.5-3.7). Figure 3.8 summarises the monthly average of D20 in the 4 zones. It can be seen that the difference between the estimation and observation D20 show reasonable agreement with Root Mean Square (RMS) around 30m for IERZ, SERZ and SNPG. In the NERZ, however, the estimated D20 is not reasonable because the RMS is about 60-100m (Figure 3.8e). The observational mean of D20 and RMS of TLM_NWPO are shown in the Table 3.1.

Given observation mean and RMS, error percentage can be calculated; its definition is

as follow:

%

× 100

= Mean

Error RMS , (3.1)

where Mean

is observational mean. Figure 3.8f shows that the error of D20 of TLM_NWPO in the NERZ is extremely large (80%~210%); in contrast to IERZ SERZ and SNPG, their errors are lower than 40%. The result suggests that the D20 of TLM_NWPO is applicable in these 3 regions (IERZ, SERZ and SNPG), but it is totally not usable in the NERZ.

As such, we demonstrate that the two-layer reduced gravity model scheme in estimating is not usable in NERZ; but it is usable in the IERZ, SERZ and SNPG to a first order. The complex hydrographic condition in the NERZ should be the main reason that leads to the failure of the two-layer model in producing unreliable D20.

NERZ is where the warm northward Kuroshio, a southward Oyashio ,and a cold sub-Arctic ocean current meet (Figure 3.9). Also there exist the Kuroshio Extension which is an eastward-flowing inertial jet accompanied by large-amplitude meanders and energetic pinched-off eddies (Qiu 2001). Thus NERZ water is mixed and complex.

It is reasonable to understand why the simply 2-layer scheme failed in this complex water.

The validation of the D20 in the NPACNFS full ocean model is shown in Figure

3.10. It can be seen that the D20 from NPACNFS is significantly better than the

estimation from the two-layer reduced gravity model in the NERZ with RMS about

30m. In the IERZ and SERZ, the performance between NPACNFS and the 2-layer

scheme is similar with RMS about 30m. In the SNPG, the NPACNFS is also better

than the 2-layer reduced scheme with RMS about 20m (Figure 3.10e). The above

suggest that the D20 from the full-ocean model NPACNFS is better in all 4 regions

because the full ocean model can handle more complicated ocean hydrographical

settings. But the error in the NERZ is still large (over 40%), so we suggest that the

D20 from NPACNFS is also unreliable in the NERZ.

3.3.2 Validation of the depth of the 26℃ isotherm

Given D20, D26 can be estimated as discussed in Chapter 2. Figures 3.11-3.13, are scatterplots of the estimated and in situ D26 in the IERZ, SERZ, and SNPG. Since D20 estimated from the two-layer reduced gravity model is not usable in the NERZ region (section 3.3.1), D26 can also not be estimated here and is thus not included in the validation. In addition, there is no climatological ratio between D26 and D20 in the IERZ in May because the SST is lower than 26℃ (Figure 2.5). Thus the D26 of May in IERZ is not retrieved. Figure 3.14 summarises the monthly mean D26 in the IERZ, SERZ and SNPG, respectively. It is found that the estimated D26 very well matches the in-situ D26 except in May and June of SERZ where undersestimation is found. Otherwise, all RMS between in-situ and estimation is between 10-20m, suggesting accurate prediction of D26 by the TLM_NWPO. The observational mean of D26 and RMS of TLM_NWPO are shown in the Table 3.2. Additionally, Figure 3.14f shows the error in SERZ during May is over 40%, suggesting it is not usable in SERZ during May. The inaccuracy in SERZ during May is attributed to the uncertainty in the D26/D20 ratio (Figure 2.5) because during the climatological SST was approaching the minimum of 26℃, i.e. D26 approaches zero; thus the D26/D20 ratio become unusually low (Figure 2.5). Therefore, resulting the unreliable D26 found in SERZ during May. Similar problem is found in the IERZ in the month of June and July.

From Figure 2.5, it is noted that in the NERZ region, the climatological D26/D20

ratio in the NERZ can not be retrieved in most months because the SST is generally

lower than 26℃ or even lower than 20℃. Clearly, the D26 in the NERZ should be

very close to zero or even non-existent. However, in the NOAA/AOML’s web, error

messily high D26 are posted operationally without quality control (Figure 3.15a). This overestimation is especially evident in the warm eddy regions (positive SSHA) that the D26 estimated can be as high as > 150m (Figure 3.15a). Figure 3.16 depicts the co-located in-situ measured ARGO profile of the same day located in the center of the warm eddy at 145.5°E 34.6°N. It clearly illustrates that even in the warm eddy region the D26 was about 25m and D20 was about 144m. In comparison with NOAA/AOML’s D26 and D20 output shown in Figure 3.15, its D26 and D20 product in the NERZ is grossly overestimated. The past in situ report from Yasuda et al. (1992) and Uehara et al. (2003) also confirm similar hydrographical structure that D26 is <

20m in the NERZ warm eddy region and are shown in Figure 3.17 and 3.18. As discussed in the validation report shown Section 3.3.1 that the 2-layer reduced gravity scheme is not applicable in the NERZ. NOAA/AOML’s product apparently neglects this aspect and the erroneous product in the NERZ is still posted on the web to the world.

Figure 3.19 shows the D26 validation of the NPACNFS model. It can be seen that its performances is slightly not as good as the simple TLM_NWPO scheme (Figure 3.14) in the IERZ, SERZ, and SNPG. A prevailing underestimation of D26 is found in the NPACNFS full ocean model. The RMS of SERZ and SNPG are between 20m and 35m, higher than the RMS from the TLM_NWPO of 10-20m in the validation months of July-September. Although the RMS of the NERZ is relatively low of the order of 5-15m (Figure 3.19e), but since the mean D26 is only about 10m (Table 3.2), the relatively error is still over 70% (Figure 3.19f). As such the D26 of the NPACNFS in the NERZ also has large error and is not unreliable.

3.3.3 Validation of the TCHP

Figures 3.20-3.22 show examples of the temperature profiles in September 2003

that is estimated from the TLM_NWPO (green), the NPACNFS full ocean model (blue), and in situ profiles (red) in the IERZ, SERZ and SNPG. The comparison suggests that the simple TLM_NWPO can effectively retrieve the upper-ocean thermal structure with good approximation, even though it only comprise four elements (SST, mixed-layer depth, the depth of 26℃ and the depth of 20℃) in the upper-ocean temperature profile. With the vertical temperature profiles, TCHP can be calculated.

Table 3.3 is the validation of TCHP between TLM_NWPO, NOAA/AOML, and the NPACNFS full ocean model by in situ measurements in the IERZ, SERZ and SNPG. Bias, RMS and error (defined in equation 3.1) are calculated. It can be found that among the 3 products, the NPACNFS has the largest error with large underestimation (bias around -25 to -48 and error in percentage about 41-71%). The TCHP from TLM_NWPO and the NOAA/AOML both perform reasonably well in the SERZ and SNPG with bias between 8-20 kJ/cm

(error about 26-37%), though the TLM_NWPO show slightly better performance than the NOAA/AOML, but the TLM_NWPO has slight overestimation. Both the NOAA/AOML and the NPACNFS are generally underestimated. The RMS in both two-layer reduced gravity models were less than 35 kJ/cm

; in contrast to NPACNFS, the RMS was higher than 40 kJ/cm

. The error shows that the uncertainty of the TLM_NWPO was less than 30%

except in the region of the IERZ, because the value of TCHP in the IERZ is relatively lower than the other regions, which leads to the higher error (50%) in the IERZ. From Table 3.3, the TLM_NWPO model is generally better than the NOAA/AOML’s two-layer reduced gravity model. In addition, the error of the NPACNFS is extremely large and is grossly underestimating TCHP.

The unexpected extremely under-estimated TCHP from the NPACNFS is probably

caused by NPACNSF’s internal mixed-layer model. In the D26 validation (Section

3.3.2), it is noted that the D26 is under-estimated by the NPACNFS. In order to further understand the problem of NPACNFS, SST comparison was made. Figure 3.23 shows the SST comparison between the NPACNFS and observational measurements. It can be seen that the SST from NPACNFS is generally under-estimated by > 1℃ in all validation areas in spite of NPACFS’s assimilation of MCSST. This underestimation generally appears in the upper-ocean, and consequently leads to the underestimation of the TCHP. The reason of the underestimation is probably caused by data assimilation or internal mixed-layer model in the NPACNFS, though is not the scope of this study. The above is somewhat surprising since the full 26-sigma levels NPACNF model was anticipated to perform better than the simple 2-layer scheme (TLM_NWPO) established in this work.

3.4 Summary

According to the validation results, the applicability of the TLM_NWPO, NOAA/AOML, and the NPACNFS model in different month and regions are summarized in Table 3.4-3.7.

In summary, we found that upper-ocean thermal structure in the NWPO can be

reasonably estimated by SSHA and SST field as input to TLM_NWPO in the IERZ,

SERZ, and SNPG; but is totally not suitable in the NERZ due to its complicated

hydrographic condition. In addition, the climatological D26/D20 ratio is unstable in

May and June for the IERZ and SERZ in the TLM_NWPO scheme. Besides from the

above, the rest of the time the TLM_NWPO is workable. As such, the ‘confidence

map’ of the TLM_NWPO scheme is depicted in Figure 3.24. Though TLM_NWPO

can not work in NERZ and partially in IERZ and SERZ, it is fortunately found that

the intensification location of typhoons mostly fall in to the ‘safe’ zones. This suggests

the applicability of the TLM_NWPO to be used in the typhoon intensification study in

Chapter 4.

It also clear that the NOAA/AOML two-layer reduced gravity model’s operationally product is not globally applicable, though is released via web. It is very risk to use this data to study typhoon intensity in the NWPO without validation.

Through the inter-comparison, it is also found that the complex NPACNFS full ocean

model is not necessarily better than the simple two-layer reduced gravity model. This

complicated model has its own problem to be improved. This work suggests that the

used of satellite SSHA and SST data as input to the two-layer reduced gravity model

is the most realistic method currently to estimate upper-ocean thermal structure in the

NWPO.