en
0.25
0.5
0.75
1.25
1.5
1.75
2
Supermodeling: Consensus by Synchronization of Alternative Models
Published on Nov 08, 20114119 Views
Computational models of an ongoing objective process, as in weather forecasting, must continually assimilate new observational data as they run. Both "truth" and "model" are chaotic systems that thu
Related categories
Chapter list
Consensus By Synchronization of Alternative Models00:00
Data Assimilation00:25
Suppose the World is a Lorenz System and Only x is Observed01:38
Let a Collection of Models Assimilate Data From (Synchronize With) One Another; Adapt the Coupling Coefficients05:37
Summary08:15
Coupled Model Intercomparison Project11:34
Error in annual mean surface air temperatures13:01
Example: Divergent Model Projections of Regional Precipitation Change14:00
Test Case: Fusing 3 Lorenz Systems With Different Parameters14:21
Supermodeling Relies on 3-Way Synchronization of Truth and Alternative Models19:47
...Or Can Use Standard Machine Learning Methods to Adapt Inter-Model Connections (!)21:00
...Or Can Use Standard Machine Learning Methods to Adapt Inter-Model Connections (2)21:42
Learning Algorithm22:46
Supermodeling Works With Multi-time-scale Models24:04
What if all models err in the same way?27:23
What if all models are biased in same direction?28:19
What if parameters shift between training and testing?31:50
What if the connection scheme obtained by cost-minimization is only locally optimal?34:09
Extension to PDE's: What is the required spatial density of inter-model coupling?35:53
What variables should be coupled?39:44
Proposed Adaptive Fusion of Two QG Channel Models41:51
Models Synchronize With Each Other and With “Truth”44:12
... As the Adaptation Procedure Estimates the Intermodel Connection Coefficient c →1/246:27
Limits of Supermodeling46:43