Multi-Resolution Variational Analysis of Sea Surface Temperature and Uncertainty Estimation

The sea surface temperature (SST) is measured at very different spatial and temporal scales, from the kilometer-resolution satellite data sets with repeat-cycle of several days to the near-continuous buoy-based data spaced hundreds of kilometers apart. The application of SST demands representation at different scales as well. For example, a global SST mean over a long time period is often examined in climate studies, while an SST snapshot of sub-kilometer resolution may be desired in biological studies. In weather forecasting, the SST is often smoothed to a spatial scale that matches the dynamical resolution of the forecast model in order to avoid spurious dynamical behaviors. To blend the variety of SST measurements and produce maps at various spatial resolutions, we thus use a multi-resolution analysis method to interpolate the data using a wavelet basis. A variational (optimization) formulation is used to realize the analysis method to improve robustness against data noise and to facilitate computation of formal uncertainty, resulting in the "Multi-Resolution Variational Analysis" (MRVA) method. Production of daily, 1-km resolution SST maps using MRVA is under way, sponsored by NASA's MEaSUREs (Making Earth System Data Records for Use in Research Environments) Program. Our project focuses on merging measurements from multiple satellite sensors to facilitate horizontal and temporal coverage. In particular, the microwave (MW) sensors have typically coarser 25-km resolutions than the infra-red (IR) sensors which can resolve down to a 1-km scale. On the other hand, the IR-based measurements are prone to data-voids due to cloud contamination, which does not affect MW sensors nearly as much. Satellite-based SST data often contain systematic biases due to atmospheric conditions such as aerosol concentration as well as calibration issues unique to each sensor. Moored and drifting buoy SST (in addition to in-situ IR radiometer data where available) are often used for bias reduction. The RMS differences between the buoy values and the map values analyzed at various wavelet scales are nearly constant over a wide range of horizontal resolutions. This indicates that the traditional buoy match-up approach may have limitations when used for validation of the high-resolution SST maps, due to relatively sparse buoy locations. Several additional validation methods are thus suggested, including independent satellite measurements, a set of other blended products (ensemble uncertainty), and use of simulated high-resolution SST fields to quantify the errors introduced by the blending procedure (an observation system simulation experiment). Efforts are also under way to improve the formal estimates of the analyzed SST field by: improving evaluation of single-sensor errors (bias and variance); determining error budget for diurnal warming; accounting for pre-analysis procedures including corrections for inter-sensor bias and diurnal heating; quantifying correlations among the sensor data realistically at high resolutions.