ObsGroup. More...
#include <array>#include <iostream>#include <string>#include <valarray>#include <vector>#include "Eigen/Dense"#include "ioda/Engines/Factory.h"#include "ioda/Exception.h"#include "ioda/Group.h"#include "ioda/ObsGroup.h"#include "unsupported/Eigen/CXX11/Tensor"
Go to the source code of this file.
Functions | |
| int | main (int argc, char **argv) |
Detailed Description
ObsGroup.
The ObsGroup class is derived from the Group class and provides some help in organizing your groups, variables, attributes and dimension scales into a cohesive structure intended to house observation data. In this case "structure" refers to the hierarchical layout of the groups and the proper management of dimension scales associated with the variables.
The ObsGroup and underlying layout policies (internal to ioda-engines) present a stable logical group hierarchical layout to the client while keeping the actual layout implemented in the backend open to change. The logical "frontend" layout appears to the client to be as shown below:
layout notes
/ top-level group nlocs dimension scales (variables, coordinate values) nchans ... ObsValue/ group: observational measurement values brightness_temperature variable: Tb, 2D, nlocs X nchans air_temperature variable: T, 1D, nlocs ... ObsError/ group: observational error estimates brightness_temperature air_temperature ... PreQC/ group: observational QC marks from data provider brightness_temperature air_temperature ... MetaData/ group: meta data associated with locations latitude longitude datetime ... ...
It is intended to keep this layout stable so that the client interface remains stable. The actual layout used in the various backends can optionally be organized differently according to their needs.
The ObsGroup class also assists with the management of dimension scales. For example, if a dimension is resized, the ObsGroup::resize function will resize the dimension scale along with all variables that use that dimension scale.
The basic ideas is to dimension observation data with nlocs as the first dimension, and allow nlocs to be resizable so that it's possible to incrementally append data along the nlocs (1st) dimension. For data that have rank > 1, the second through nth dimensions are of fixed size. For example, brightness_temperature can be store as 2D data with dimensions (nlocs, nchans).
Definition in file 05a-ObsGroup.cpp.
