The sensitivity analysis in accordance with the resolution of the satellite sea surface temperature data

Abstract

It is important to improve ocean initial fields through the assimilation of high resolution observation data in order to enhance predictability of ocean model. The reproducibility of the ocean models are improved through the observation data assimilation in several studies (e.g., Wang et al., 2006; Oey et al., 2005a; Larsen et al., 2007). By applying a variety of observations and data assimilation techniques, many operational institutions are performing ocean forecasting. However, the extremely high resolution as compared to the model grid may result in a waste of computational resources due to the increased operation time when the data assimilated in model. Also it can be a reason to decrease accuracy of model. In this study, the sensitivity of analysis fields of data assimilation system according to subsampling interval of GHRSST L2P data is analyzed using the global ocean circulation forecasting system. The assimilated satellite observed SST data were AVHRR NOAA-18G and AVHRR MEOP-A. To examine the proper spatial resolution of these data for the data assimilation system, we performed two experiments of Exp.MDT (Moderate Density Test) and Exp.LDT (Low Density Test). The resolutions of assimilated SST data for each experiment are shown in table 1. First, we collect the data from GHRSST L2P ftp site. Secondly, the contaminated data is removed in accordance with flags (e.g., proximity flags, confidence flag, ice, and land masked). And we reduce data quantity using simple thinning method. Analysis areas are global region and regional areas (e.g., Agulhas return, Cold tongue, Gulf stream, Kuroshio extension) showing strong horizontal gradient in sea surface temperature with more than 1 degree (Lei and Kawamura, 2003). In the result of Taylor diagram analysis, the values of both experiments are close to the reference. And the normal standard deviation is close to zero. So, we can confirm that the performance of data assimilation of both experiments is excellent. There was no big difference between two experiments in correlation coefficient with maximum of 0.01. However, the comparison with observation data (e.g., in-situ, AVHRR NOAA-18G and AVHRR METOP-A) show that the Exp.LDT has lower RMSE than that from the Exp.MDT. In the global region, the RMSE of Insitu in Exp.MDT, Exp.LDT is 0.45, 0.41, that of AVHRR NOAA-18G is 0.45, 0.43, and that of AVHRR METOP-A is 0.64, 0.63, respectively. The RMSE difference reaches about 0.04°C in the global region. These results also appear in the region where a strong horizontal gradient exists, such as Kuroshio-Extension, Gulf-stream, and Cold tongue. As a result, it is considered that the low-resolution data is proper to assimilate for data assimilation system assimilation. However, it is considered to require additional experiment results for high density test (Exp.HDT). In the future, we have a plan for deriving a proper subsampling resolution considering accuracy of model and computational cost.