Home page  
Home   Your Room   Login   Contact   Feedback   Site Map   Search:  
Discover this product  
About Us
Overview
Getting here
Committees
Products
Forecasts
Order Data
Order Software
Services
Computing
Archive
PrepIFS
Research
Modelling
Reanalysis
Seasonal
Publications
Newsletters
Manuals
Library
News&Events
Calendar
Employment
Open Tenders
   
Home > Research > Ifsdocs > PHYSICS >  
 

DYNAMICS

IFS documentation Front Page


Table of contents

Chapter 1. Introduction

Chapter 2. Basic equations and discretization

Chapter 3. Semi-Lagrangian formulation

Chapter 4. Computational details

REFERENCES
 

  Next Section
Previous Section


3.7 Optimization of vertically non-interpolating scheme




In the `vertically non-interpolating' scheme, the departure point of each modified trajectory lies on a model level. For the set of arrival points on each model level, it is of interest to determine the frequency distribution of the corresponding departure points. The results of an experiment run to collect these statistics led to a significant optimization of the code for the vertically non-interpolating scheme.


The statistics were obtained from a 10-day forecast using the model in its operational configuration: T213, 31 levels, with a 15-minute timestep. The results are summarized in Table 3.1, which shows that the vast majority (99.67% overall) of modified trajectories are horizontal; no departure point was ever more than three model levels away from its corresponding arrival point.


The implication of these results is that a great deal of redundant calculation was being performed in the vertically non-interpolating scheme. For each horizontal modified trajectory, the interpolation of the horizontal winds in the trajectory calculation itself becomes two-dimensional rather than three-dimensional, as do the interpolations of `right-hand side' terms at the mid point of the trajectory, while the additional interpolations to calculate terms of the form in (3.12) are not required at all. Consequently, special routines were written to perform interpolations which are two-dimensional everywhere except at a set of `flagged' points where they become three-dimensional, and similarly to perform two- or three-dimensional interpolations at the flagged points while skipping all other points. The use of these special routines reduced the `semi-Lagrangian overhead' for the vertically non-interpolating scheme by about 30%.


Table 3.1 Frequency distribution (%) of departure points in the `vertically non-interpolating' scheme
Arrival
level
Departure levels




1-6
100.00



7-9
100.00
*


10
99.99
0.01
*

11
99.96
0.04
*

12
99.89
0.11
*

13
99.76
0.24
*
*
14
99.60
0.40
*
*
15
99.43
0.57
*
*
16
99.28
0.72
*

17
99.16
0.83
0.01

18
99.08
0.92
*

19
99.05
0.94
0.01

20
99.05
0.94
0.01
*
21
99.09
0.91
*
*
22
99.14
0.85
0.01
*
23
99.22
0.78
*

24
99.31
0.69
*

25
99.44
0.56
*

26
99.60
0.40
*

27
99.78
0.22
*

28
99.92
0.08
*

29
99.99
0.01


30
100.00
*


31
100.00






Asterisks indicate less than 0.005% frequency





Next Section
Previous Section



 

Top of page 08.04.2002
 
   Page Details         © ECMWF
shim shim shim