Researches On Oceanic Field Recovery And Front Detection

It is meaningful for the sustainable development of oceanic environment and oceanic resources to detect oceanic features by using satellite remote sensing techniques. In this paper, spatial-temporal oceanic data reconstruction and front detection have been researched based on marine field analysis. Spatial-temporal sea surface temperature (SST) and chlorophyll-a (Chl-a) data in the northern South China Sea (NSCS) were reconstructed by using DINEOF algorithm. In addition, a modified DINEOF algorithm was proposed in this paper and it was verified to be better than the ordinary DINEOF algorithm based on the tests on real SST data and synthetic SST data. A front detection algorithm based on gravitational model was proposed. Because of the subjectivity of threshold setting that is a necessary step in the front detection procedure, a front detection algorithm based on threshold interval instead" of single threshold was proposed. For many front detection researches are on basis of satellite-derived data rather than satellite data, we proposed a front detection algorithm that can be directly used for satellite data to detect oceanic features. Due to the disadvantages of front frequency and average gradient that are obtained from the time-series images, a front composite algorithm was proposed to enhance the visuality of fronts. Finally, according to the peoposed methods above, monthly disttibutions and regional anomalies of SST and Chl-a data in NSCS were analyzed. Additionally, monthly distributions of fronts in Bohai, Yellow and East China seas were also researched. Specific research contents and conclusions are:(1) SST and Chl-a data in the NSCS were reconstructed by using DINEOF algorithm and a modified DINEOF algorithm was proposed. Monthly 4-km daytime Advanced Very High Resolution Radiometer (AVHRR) SST data in 1985-2009 (without 2002-2004) and monthly mapped 9-km Sea-viewing Wide Field-of-view Sensor (SeaWiFS) Chl-a data in 1998-2010 were reconstructed by using DINEOF algorithm. The results indicate that 40 modes are the most suitable to reconstruct SST data, giving the highest Pearson correlation coefficient 0.9765 (P<0.001), the number of cross-validation points is 182,357 and the lowest RMSE difference is 0.5438; the optimal number of EOFs to reconstruct the Chl-a data is 39 and the corresponding Pearson correlation coefficient is 0.9075 (P<0.001) and the number of cross-validation points is 23,719. The error with the optimal number of EOF modes is 0.1447 in log RMSE for Chl-a data, which is smaller than the algorithm "noise" (0.22 in log RMSE) accepted in O'Reilly et al. (2000). A modified DINEOF algorithm was proposed which is not necessary to set cross-validation points. And in the new algorithm, the optimal EOF modes were changeable and determined based on the local data instead of the entire data set. Through the reconstructions on real SST data and synthetic SST data, the accuracies from the modified algorithm are all higher than the ordinary algorithm and the new algorithm can be less affected by the miss data.(2) A front detection algorithm based on gravitational model was proposed. In the algorithm, median filter and zero-elimination were first applied; then in order to remove the influences from original data, local normalization was operated; and a local enhancement algorithm was used to protrude the fronts as well as to depress noises; finally, the gravition edge detection algorithm was exploited to detect fronts. The results verified that this algorithm could effectively detect fronts and decrease the influences of noises.(3) The threshold setting is necessary for some front detection methods, such as gradient or entropy methods, yet the subjectivity of threshold setting is difficult to solve. Hence, in this paper, a front detection algorithm based on threshold interval was proposed, which could partly eliminate the subjectivity of threshold setting. Tests on Kuroshio front verified that this algorithm could take advantage of merits from upper and lower thresholds, which means it could keep the continuities of fronts and meanwhile reduce the influences of noises.(4) Oceanic features are mainly detected based on satellite-derived data, so the researches on front detection algorithms that can be directly used for satellite data seem to be not enough. Hence, a front detection algorithm for satellite data was proposed and "Beijing-1" small satellite data were used as experimental data to verify the validity of the proposed algorithm.(5) To test the stabilities and strengths of fronts and to reduce the influences of missing data and noises, front frequency and average gradient are usually calculated based on time-series images. However, these two parameters both have some limits, though they can partly reflect some characteristics of fronts. In this paper, a front composite algorithm based on these two parameters was proposed to better reflect the stabilities and strengths of fronts.(6) Based on cloud-free SST and Chl-a data obtained from DINEOF reconstruction, the monthly distributions of SST and Chl-a data were first analyzed; then, according to anomaly analysis algorithm, the regional anomalies and corresponding relationships of SST's and Chl-a's anomalies were discussed.[7] Based on the proposed front composite algorithm, daily 4-km MODIS Terra SST data in 2000-2013 were used as experimental data to analyze the monthly distributions of fronts in Bohai, Yellow and East China seas.