Issues with Optimal Estimation retrieval of Sea Surface Temperature retrievals from MODIS

Abstract:

We use Optimal Estimation (OE) to retrieve a simplified state of the ocean-atmosphere system as described by vector z = [SST, TCWV] where SST is the sea surface temperature, and TCWV is the total column water vapor. Input data are measurements of radiance in MODIS 11 and 12µm channels. RTTOV is used as the forward model.

The OE retrieval calculates a correction to the a priori state from the difference between the actual satellite measurements and simulated measurements obtained by radiative transfer calculations for the prior state. It is understood that the difference between the model and the measurement represents the miss-specification of the a priori state which is then rectified but the calculated correction. However, any bias present in either the model or the measurement that is not specifically accounted for will also contribute to the correction.

We test the performance of the OE approach on a set of MODIS data. In order to assess the significance of biases we use the buoy SST (SSTb) as the prior that is well known and with the expectation that the retrieved SSTOE = SSTb. Prior TCWV and atmospheric profiles are from the ECMWF ERA5 reanalysis. In a setup with a well known prior SST, practically all corrections are applied to TCWV. Also, all biases present will contribute to the adjustment of TCWV and the retrieved TCWV would not be ‘correct’ but rather it would represent some effective TCWV which allows for the unbiased retrieval of SST. We use this modified TCWV in RTTOV to simulate MODIS radiances and Jacobians and repeat the OE retrieval to yield unbiased estimates of SST.

We found that such tuning the OE had only small effect on the mean offset between the SSTOE and SSTb decreasing it from -0.02 K to -0.01K. However, the standard deviation and the root mean square of the distribution of the differences are reduced almost twofold and threefold respectively. Similarly, the mean difference between measured MODIS brightness temperatures and those simulated by RTTOV decreases by about 25% while RMS decreases threefold. These results indicate that both the water vapor bias and the unknown biases in the MODIS OE retrieval using RTTOV as a forward model are small and should not hinder obtaining valid estimates of SST.