Abstract's details
Multi-mission sea state bias modeling: development and assessment
CoAuthors
Event: 2014 Ocean Surface Topography Science Team Meeting
Session: Instrument Processing: Corrections
Presentation type: Type Oral
Contribution: PDF file
Abstract:
Related to ocean surface waves, the sea state bias (SSB) remains as one of the largest uncertainties sources in the ocean altimeter range error budget. More accurate SSB models are still highly desired for both scientific applications and satellite altimeter climate record generation. Measurable advancement in modeling for a new SSB correction was recently reported when including information from ocean wave models. In particular, Tran et al. (2010) demonstrated that improvement could be gained with three variable (3D) SSB models that include mean wave period estimate from a global wave model (Wavewatch III) in addition to the traditional 2D input variables being the altimeter onboard measured significant wave height (Hs) and wind speed (U10). However, within the context of recent climate data record evaluations of Jason-1 altimeter correction methods, very limited gains were observed using such a new SSB correction. This study seeks to improve upon available multi-mission SSB models as well as to answer the questions of when and how can one certify and quantify the improvement in new candidate SSB corrections using the available suite of metrics derived from satellite altimeter datasets.
Classically, the altimeter SSB correction models have been empirically developed using input data from the alongtrack difference of residual sea level anomaly at either crossover or collinear positions (Gaspar et al., 2002; Labroue et al., 2004), and more recently using the sea level residual directly without differencing (Vandemark et al., 2002; Tran et al., 2010). We have shown that either choice of input data for SSB modeling has its own advantages and disadvantages (Vandemark et al., 2013). Specifically, we find that non-uniform space/time sampling usually exists in difference-based input data approach leading to significant model uncertainty, while correlation between the mean sea surface signal and SSB model inputs data can systematically corrupt that latter approach. Both methods require care in SSB model derivation and then in ascribing uncertainty when assessing model performance.
Regardless of the input data, statistical nonparametric estimation methodologies are now employed to develop the SSB estimators. These include 1) the local linear kernel LLK (Gaspar et al., 2002) and 2) spline smoothing SP (Feng et al., 2010). The two estimators are nearly equivalent when applied to an identical input dataset. An advantage of the SP approach is its ease and speed in computation and its particular suitability for high dimensional estimation when compared to the LLK approach.
Study goals are: 1) to develop 2D/3D SSB models for recent altimeter missions (Jason 1-2 and SARAL/AltiKa) in terms of direct data and using an SP estimator, 2) to include in this design a strategy for preprocessing direct sea level residual data to remove temporal trend in direct residual sea level data, and 3) to provide a thorough assessment for 2D and 3D SSB model performances. In this, we explore differences in variance reduction results amongst direct, collinear, and crossover analysis tests.
Classically, the altimeter SSB correction models have been empirically developed using input data from the alongtrack difference of residual sea level anomaly at either crossover or collinear positions (Gaspar et al., 2002; Labroue et al., 2004), and more recently using the sea level residual directly without differencing (Vandemark et al., 2002; Tran et al., 2010). We have shown that either choice of input data for SSB modeling has its own advantages and disadvantages (Vandemark et al., 2013). Specifically, we find that non-uniform space/time sampling usually exists in difference-based input data approach leading to significant model uncertainty, while correlation between the mean sea surface signal and SSB model inputs data can systematically corrupt that latter approach. Both methods require care in SSB model derivation and then in ascribing uncertainty when assessing model performance.
Regardless of the input data, statistical nonparametric estimation methodologies are now employed to develop the SSB estimators. These include 1) the local linear kernel LLK (Gaspar et al., 2002) and 2) spline smoothing SP (Feng et al., 2010). The two estimators are nearly equivalent when applied to an identical input dataset. An advantage of the SP approach is its ease and speed in computation and its particular suitability for high dimensional estimation when compared to the LLK approach.
Study goals are: 1) to develop 2D/3D SSB models for recent altimeter missions (Jason 1-2 and SARAL/AltiKa) in terms of direct data and using an SP estimator, 2) to include in this design a strategy for preprocessing direct sea level residual data to remove temporal trend in direct residual sea level data, and 3) to provide a thorough assessment for 2D and 3D SSB model performances. In this, we explore differences in variance reduction results amongst direct, collinear, and crossover analysis tests.