Method and system for generating deterministic analysis set based on reverse localization

Abstract

The invention discloses a method and system for generating deterministic analysis set based on reverse localization. The method comprises the following steps: calculating the analysis state mean valuebased on the control variable analysis value, the prediction set state mean value and the localized prediction error covariance square root matrix, and constructing the state variable analysis errorcovariance square root matrix and calculating the analysis increment for each set member; according to the mean value of analysis states for each set member and summing of the analysis increment, obtaining the final set of analysis states for each set member as the initial set of states. The localization scheme and the generation mode of the analysis set are improved, so that the sampling error and long-range pseudo-correlation can be eliminated better, , a set of analytical dispersion can be generated, which is more inclined to the analytical error, and finally the performance of data assimilation can be improved significantly, and the initial state of the numerical prediction problem can be obtained more accurately and reliably.

Description

technical field [0001] The invention relates to the field of numerical forecasting or forecasting, in particular to a method and system for generating a deterministic analysis set based on reverse localization. Background technique [0002] For fields such as numerical weather prediction and numerical ocean prediction, the state dimension of the simulation system in these application fields is very high (O(10 8 )~O(10 9 )), the amount of calculation itself is huge, when the set method is used, the increase in the number of members will lead to a multiplied increase in the amount of calculation, and the number of set members used in practical applications will be much smaller than the system state dimension. The numerical forecasting or forecasting problem is an initial value problem, that is, given an initial state, the state at a certain time in the future can be predicted or predicted under the condition of known system evolution law. Therefore, whether the initial state...

AI Technical Summary

Problems solved by technology

[0030] However, this scheme has the following disadvantages: (1) In Step 6, this scheme uses the observation random disturbance method, because the observation error statistics itself is biased, the random disturbance method may lead to greater systematic deviation and sampling error; (2) In step 8, the localized information of background field covariance cannot be well preserved in the process of observing random disturbances, resulting in unsatisfactory elimination of sampling error, which in turn makes the dispersion of the set smaller Finally, the aggregate system does not converge and eventually diverges

However, the use of localization technology makes it difficult to analyze the generation of set members. At present, the method of random perturbation observation data is often used for processing

However, due to the deviation of the observation error statistic itself, the random perturbation method may lead to greater systematic deviation and sampling error.

Images