Calculation of Sea Surface Temperature Forward Radiative Transfer Model Approach Alec Bogdanoff, Florida State University Carol Anne Clayson and Brent Roberts GHRSST Users Symposium May 28, 2009
Background Empirically based retrievals are biased towards those areas where in situ measurements are abundant. GOAL: Increases confidence in Sea Surface Temperatures (SSTs) in areas with few in situ data points with a minimum computing effort. Use technique of Merchant and LeBorgne (2004), which provides a more physical approach to the derivation of Brightness Temperatures (BT) coefficients. 2
Background con’t. Using a full Radiative Transfer Model (RTM) for Sea- Surface Temperature (SST) retrievals is computationally exhaustive. We use RTTOV, a fast RTM. Neural Networks have the ability to increase computational efficiency by times depending on atmospheric profile. Chavallier et. al. (1998) 3
Distribution of MDB 4 Match-up Database from Pierre Le Borgne and EUMETSAT's Ocean and Sea Ice Satellite Application Facility
Quick Introduction to Neural Networks Type of nonlinear regression Inputs are weighted and transformed nonlinearly. These nonlinearly transformed values are weighted and linearly combined to produce the output. For neural networks, one uses standard nonlinear optimization techniques such as steepest-descent No guarantee that a global minimum will always be found and so it sometimes requires a bit of trial and error 5
Neural Network Example 6
Neural Networks Used 100,000 randomly selected data (of 236, 704) to train the networks Using night-time measurements only Buoy to skin SST correction of -0.2 K Inputs of NWP and observed BTs 7
Skill of Neural Network Approach 8 NN – No RTMNN – RTM O&SI SAF AlgorithmMLR
Skill of Neural Network Approach 9 NN – No RTMNN – RTM O&SI SAF AlgorithmMLR
Neural Network emulation of O&SI SAF Nocturnal SST Algorithm 10 MLP – NO RTM MLP – RTM MLRO&SI SAF MLP – O&SI RMS Mean Bias Neural Net emulated O&SI SAF Algorithm
Can the Neural Network emulate the Radiative Transfer Model? 11
Neural Network Emulation of RTTOV μ m 10.8 μ m 12.0 μ m RMS Mean Bias μm 10.8 μm 12.0 μm
Neural Network Emulation of RTTOV Jacobian: dBT/dTCWV 3.7 μ m 10.8 μ m 12.0 μ m RMS Mean Bias μm10.8 μm 12.0 μm
Comparison to GHRSST Products This technique will be compared to GHRSST’s L2P AVHRR products, once expanded beyond MetOp-A. After optimization, the dataset will be compared to the closest spatial and temporal L4 data point, specifically those products that use AVHRR. 14
Conclusions As a first guess model, the neural networks do an exceptional job, with the inclusion of the RTM increasing the accuracy of the modeled approach. Will see much greater impact from the Radiative Transfer approach in cloudy sky and/or daytime condition. 15
Future Work Many more runs are required to determine the best regression. Further optimization of the networks will increase the accuracy of the approach. Expand beyond clear sky at night. A RTM approach will be greater during daytime and cloudy sky conditions Expand beyond MetOp-A satellite to allow for better global coverage of satellite derived SSTs. The error characteristics of the resulting SST datasets will be compared to GHRSST products. 16
Acknowledgements Brent Roberts and Carol Anne Clayson Pierre Le Borgne of Météo-France (MDB) Chris Merchant of the University of Edinburgh (RTM) The dataset is from EUMETSAT's Ocean and Sea Ice Satellite Application Facility NASA PO Program for their support of this research 17
References Chevallier, F., F. Chéruy, N. A. Scott, and A. Chédin, 1998: A neural network approach for a fast and accurate computation of a longwave radiative budget. J. Appl. Meteor., 37, 1385–1397. Merchant, C. J. and P. Le Borgne, 2004: Retrieval of Sea Surface Temperature from Space, Based on Modeling of Infrared Radiative Transfer: Capabilities and Limitations. J. Atmos. Oceanic Technol., 21, _____, P. Le Borgne, A. Marsouin, and H. Roquet, 2008: Optimal estimation of sea surface temperature from split-window observations. Remote Sens. Environ., 112, , doi: /j.rse Ocean & Sea Ice SAF Project Team, 2009: MetOp/AVHRR Sea Surface Temperature Product User Manual Version 1.5. O&SI SAF Product User Manual. SAF/OSI/CDOP/M-F/TEC/MA/127, 48 pp. Rumelhart, D. E., G. E. Hinton, and R. J. Williams, 1986: Learning internal representations by error propagation. Parallel Distributed Processing: Exploration in the Macrostructure of Cognition 1, D. E. Rumelhart and McClelland, Eds., MIT Press, Walton, C. C., W. G. Pichel, J. F. Sapper, and D. A. May, 1998: The development and operational application of nonlinear algorithm for the measurement of sea surface temperatures with the NOAA polar-orbiting environmental satellites. J. Geophys. Res., 103(C12),