270 likes | 404 Views
A new prototype AMSR-E SWE operational algorithm. M. Tedesco The City College of New York, CUNY, NYC With contributions from : Chris Derksen , Jouni Pulliainen , James Foster, Richard Kelly. Acknowledgment: NASA Terrestrial Hydrology and Energy and Water Cycle Programs. OUTLINE.
E N D
A new prototype AMSR-E SWE operational algorithm • M. Tedesco • The City College of New York, CUNY, NYC • With contributions from : Chris Derksen, JouniPulliainen, James Foster, Richard Kelly Acknowledgment: NASA Terrestrial Hydrology and Energy and Water Cycle Programs
OUTLINE • Current status of the AMSR-E SWE operational algorithm • Anew operational algorithm for SWE • Validation stage • Refinement and future directions for the product
Current status • No major issues to report • The code from SIPS is up and running at CCNY and will be modified once the new ATBD (see next) is going to be approved • Testing of the new code at CCNY and then migrating to UAH for production undergoing
Current algorithm Ingestion of Tbs and check on precipitation, wet snow and shallow snow
Identifying major issues: Current AMSR-E SWE vs. CMC CMC – AMSR-E CMC AMSR-E [cm] [cm]
Surface temperature from AMSR-E CMC – AMSR-E [cm] [C] January Surface Temperature from AMSR-E
The current algorithm has been attempting to account for the evolution of grain size to consider this aspect in the retrieval scheme • However, spatio-temporal evolution of grain size is difficult to model without ancillary data (given the sensitivity of PMW data to this parameter) • This is a large source of error
Proposed changes • Use of neural networks and electromagnetic model to derive quantities related used in retrieval coefficients (e.g., effective grain size) • Formula for retrieval coefficients (e.g., those relating the snow depth to Tbs) • Density used to convert snow depth to SWE • New formula for surface temperature estimate
What is available now that was not in the past, when the algorithm was originally developed ? • Electromagnetic theory advances • Computational power • Long time series of PMW data • Other snow depth products (from ground obs. and models) • Numerical techniques
Module 1 Module 2 Emissivity approach Module 1.1 Ground obs. Snow depth monthly climatology AMSR-E 10 GHz Kriging (or similar) Ground and surface temperature ANN Module 1.2 Other avail. prod Module 3 Snow depth Density model AMSR-E All channels Module 2.1 Electromagnetic model Day of the year Simulated Tbs Set of Snow parameters Module 2.2 ANN Training Grain size Simulated TBs Ts, Tg , et al. Monthly Climatology Snow density Maps
Example: map of effective grain size January 2006
Current vs. new coefficients [cm/K] [cm/K] Current dynamic coefficient map at 37 GHz New dynamic coefficient map at 37 GHz
Density Current algorithm New algorithm g/cm3 Using Sturm et al., 2011
Old vs. new algorithm OLD AMSR-E [cm] e.g., January 2004
Assessment and validation: Comparison with CMC product
Assessment: NEW vs. OLD algorithm CMC data set 2002 – 2010 Data (CMC data set is used as ‘truth’)
Assessment: NEW vs. OLD algorithm CMC data set (cont’d) 2002 – 2010 Data (CMC data set is used as ‘truth’)
Assessment: NEW vs. OLD algorithm WMO data set As in the case of CMC, the new algorithm provides better results than the original one for all months
Assessment: NEW vs. OLD algorithm GlobSNOW(SWE) As in the case of CMC and WMO data, the new algorithm provides better results than the original one for all months
MULTIPLE “LAYERS” WITHIN THE SAME PRODUCT • Effective grain size • Snow depth • Surface temperature • Snow bulk density Second layer = migrating operational algorithm Third layer = future operational algorithm
Steps for operational implementation ATBD ready Submitted to NASA next w.e. Evaluation and review Implementation and testing at UAH ATBD and software to UAH NSIDC distribution request RESEARCH product distribution Parallel distribution Of current and new products Migration to new product
Summary • The new proposed AMSR-E SWE algorithm makes use of ANN, electromagnetic models to compute grain size and use this information in the new retrieval scheme • Density is computed as a function of depth, day of the year and snow class in a dynamic fashion • New approach proposed for surface temperature • The new approach provides better results than the current algorithm when considering three different and independent validation data sets • The comparison of the two products allows also to move the AMSR-E SWE product to validated stage 2
Future plans • Introduce atmospheric correction (e.g., using other AMSR products) • Lake fraction (e.g., using a different ‘tuned’ algorithm for lake-rich areas) • Introduction of uncertainty maps • Extension to AMSR2 • Modifying the first part of the algorithm for flags and wet snow, precipitation events detection
The future generation of SWE global products • Multivariate outputs (snow density, depth, effective grain size) • Uncertainty will be included based upon quality flags depending on factors such as forest cover, lakes, atmosphere, etc. • The proposed approach is applicable ‘as is’ to other platforms (AMSR2, SSM/I) • The modules can be replaced • - e.g., Snow depth climatology with ground obs. assimilation • EM models can be replaced • Density model