@Override
public AbstractGridDataset createDataset(String id, String location)
throws IOException, EdalException {
NetcdfDataset nc = null;
try {
/*
* Open the dataset, using the cache for NcML aggregations
*/
nc = openAndAggregateDataset(location);
/*-
* We may in future be able to use forecast model run collection aggregations for
* dealing with the case of overlapping time axes. To do this the code will look
* something like this:
*
* StringBuilder sb = new StringBuilder();
* Formatter formatter = new Formatter(sb, Locale.UK);
* Fmrc f = Fmrc.open(location, formatter);
*
* in openAndAggregateDataset. It will need to build up an NcML document which
* does this. It should look something like:
*
* <netcdf xmlns="http://www.unidata.ucar.edu/namespaces/netcdf/ncml-2.2" enhance="true">
* <aggregation dimName="run" type="forecastModelRunCollection" timeUnitsChange="true">
* <!-- scanFmrc actually works, but what we want is something like the following bit -->
* <scanFmrc location="/home/guy/Data/POLCOMS_IRISH/" regExp=".*\.nc"/>
* <netcdf location="/home/guy/Data/POLCOMS_IRISH/polcoms_irish_hourly_20090320.nc" coordValue="2009-03-20T00:00:00Z" enhance="true" />
* <netcdf location="/home/guy/Data/POLCOMS_IRISH/polcoms_irish_hourly_20090321.nc" coordValue="2009-03-21T00:00:00Z" enhance="true" />
* <netcdf location="/home/guy/Data/POLCOMS_IRISH/polcoms_irish_hourly_20090322.nc" coordValue="2009-03-22T00:00:00Z" enhance="true" />
* </aggregation>
* </netcdf>
*
* For more documentation see:
* http://mailman.unidata.ucar.edu/software/thredds/current/netcdf-java/ncml/FmrcAggregation.html
*
* We then can do stuff like:
*
* ucar.nc2.dt.GridDataset gridDataset = f.getDatasetBest();
*
* To get the single best aggregation of the overlapping time axis
*
* Then we need to work with GridDatasets in place of NetcdfDatasets. Stuff like:
*
* for(Variable variable : gridDataset.getNetcdfFile().getVariables()) {
* // blah blah
* }
*
* will be necessary. We need to check that that works with remote datasets too
*/
/*
* We look for NetCDF-U variables to group mean/standard-deviation.
*
* We need to do this here because we want to subsequently ignore
* parent variables
*/
Map<String, String[]> varId2AncillaryVars = new HashMap<String, String[]>();
for (Variable variable : nc.getVariables()) {
/*
* Just look for parent variables, since these may not have a
* grid directly associated with them
*/
for (Attribute attr : variable.getAttributes()) {
if (attr.getFullName().equalsIgnoreCase("ancillary_variables")) {
varId2AncillaryVars.put(variable.getFullName(), attr.getStringValue().split(" "));
continue;
}
}
}
ucar.nc2.dt.GridDataset gridDataset = CdmUtils.getGridDataset(nc);
List<GridVariableMetadata> vars = new ArrayList<GridVariableMetadata>();
/*
* Store a map of component names. Key is the compound name, value
* is a 2-element String array with x, y component IDs
*
* Also store a map of whether these components are really
* eastward/northward, or whether they are locally u/v
*/
Map<String, String[]> xyComponentPairs = new HashMap<String, String[]>();
Map<String, Boolean> xyNameToTrueEN = new HashMap<String, Boolean>();
/*
* Store a map of variable IDs to UncertML URLs. This will be used
* to determine which components are mean/std/etc.
*
* TODO implement more than just Mean/SD
*/
Map<String, String> varId2UncertMLRefs = new HashMap<String, String>();
/*
* Here we store the parent variable IDs and their corresponding
* title.
*/
Map<String, String> parentVarId2Title = new HashMap<String, String>();
for (Gridset gridset : gridDataset.getGridsets()) {
GridCoordSystem coordSys = gridset.getGeoCoordSystem();
HorizontalGrid hDomain = CdmUtils.createHorizontalGrid(coordSys);
VerticalAxis zDomain = CdmUtils.createVerticalAxis(coordSys);
TimeAxis tDomain = CdmUtils.createTimeAxis(coordSys);
/*
* Create a VariableMetadata object for each GridDatatype
*/
for (GridDatatype grid : gridset.getGrids()) {
VariableDS variable = grid.getVariable();
String varId = variable.getFullName();
String name = getVariableName(variable);
/*
* If this is a parent variable for a stats collection, we
* don't want it to be a normal variable as well.
*/
if (varId2AncillaryVars.containsKey(varId)) {
parentVarId2Title.put(varId, name);
continue;
}
/*
* If it is a child variable is (potentially) referenced by
* UncertML, store its ID and the (possible) UncertML URI
*/
for (Attribute attr : variable.getAttributes()) {
if (attr.getFullName().equalsIgnoreCase("ref")) {
varId2UncertMLRefs.put(varId, attr.getStringValue());
}
}
Parameter parameter =
new Parameter(
varId,
variable.getShortName(),
variable.getDescription(),
variable.getUnitsString(),
name);
GridVariableMetadata metadata =
new GridVariableMetadata(
variable.getFullName(), parameter, hDomain, zDomain, tDomain, true);
vars.add(metadata);
if (name != null) {
/*
* Check for vector components
*/
if (name.contains("eastward_")) {
String compoundName = name.replaceFirst("eastward_", "");
String[] cData;
if (!xyComponentPairs.containsKey(compoundName)) {
cData = new String[2];
xyComponentPairs.put(compoundName, cData);
xyNameToTrueEN.put(compoundName, true);
}
cData = xyComponentPairs.get(compoundName);
/*
* By doing this, we will end up with the merged
* coverage
*/
cData[0] = varId;
} else if (name.contains("northward_")) {
String compoundName = name.replaceFirst("northward_", "");
String[] cData;
if (!xyComponentPairs.containsKey(compoundName)) {
cData = new String[2];
xyComponentPairs.put(compoundName, cData);
xyNameToTrueEN.put(compoundName, true);
}
cData = xyComponentPairs.get(compoundName);
/*
* By doing this, we will end up with the merged
* coverage
*/
cData[1] = varId;
} else if (name.matches("u-.*component")) {
String compoundName = name.replaceFirst("u-(.*)component", "$1");
String[] cData;
if (!xyComponentPairs.containsKey(compoundName)) {
cData = new String[2];
xyComponentPairs.put(compoundName, cData);
xyNameToTrueEN.put(compoundName, false);
}
cData = xyComponentPairs.get(compoundName);
/*
* By doing this, we will end up with the merged
* coverage
*/
cData[0] = varId;
} else if (name.matches("v-.*component")) {
String compoundName = name.replaceFirst("v-(.*)component", "$1");
String[] cData;
if (!xyComponentPairs.containsKey(compoundName)) {
cData = new String[2];
xyComponentPairs.put(compoundName, cData);
xyNameToTrueEN.put(compoundName, false);
}
cData = xyComponentPairs.get(compoundName);
/*
* By doing this, we will end up with the merged
* coverage
*/
cData[1] = varId;
}
/*
* We could potentially add a check for zonal/meridional
* here if required.
*/
}
}
}
CdmGridDataset cdmGridDataset =
new CdmGridDataset(id, location, vars, CdmUtils.getOptimumDataReadingStrategy(nc));
for (Entry<String, String[]> componentData : xyComponentPairs.entrySet()) {
String commonName = componentData.getKey();
String[] comps = componentData.getValue();
if (comps[0] != null && comps[1] != null) {
cdmGridDataset.addVariablePlugin(
new VectorPlugin(comps[0], comps[1], commonName, xyNameToTrueEN.get(commonName)));
}
}
for (String statsCollectionId : varId2AncillaryVars.keySet()) {
String[] ids = varId2AncillaryVars.get(statsCollectionId);
String meanId = null;
String stddevId = null;
for (String statsVarIds : ids) {
String uncertRef = varId2UncertMLRefs.get(statsVarIds);
if (uncertRef != null
&& uncertRef.equalsIgnoreCase("http://www.uncertml.org/statistics/mean")) {
meanId = statsVarIds;
}
if (uncertRef != null
&& uncertRef.equalsIgnoreCase(
"http://www.uncertml.org/statistics/standard-deviation")) {
stddevId = statsVarIds;
}
}
if (meanId != null && stddevId != null) {
MeanSDPlugin meanSDPlugin =
new MeanSDPlugin(meanId, stddevId, parentVarId2Title.get(statsCollectionId));
cdmGridDataset.addVariablePlugin(meanSDPlugin);
}
}
return cdmGridDataset;
} finally {
CdmUtils.closeDataset(nc);
}
}
/**
* Opens the NetCDF dataset at the given location, using the dataset cache if {@code location}
* represents an NcML aggregation. We cannot use the cache for OPeNDAP or single NetCDF files
* because the underlying data may have changed and the NetcdfDataset cache may cache a dataset
* forever. In the case of NcML we rely on the fact that server administrators ought to have set a
* "recheckEvery" parameter for NcML aggregations that may change with time. It is desirable to
* use the dataset cache for NcML aggregations because they can be time-consuming to assemble and
* we don't want to do this every time a map is drawn.
*
* @param location The location of the data: a local NetCDF file, an NcML aggregation file or an
* OPeNDAP location, {@literal i.e.} anything that can be passed to
* NetcdfDataset.openDataset(location).
* @return a {@link NetcdfDataset} object for accessing the data at the given location.
* @throws IOException if there was an error reading from the data source.
*/
private NetcdfDataset openAndAggregateDataset(String location) throws IOException, EdalException {
NetcdfDataset nc;
if (location.startsWith("dods://") || location.startsWith("http://")) {
/*
* We have a remote dataset
*/
nc = CdmUtils.openDataset(location);
} else {
/*
* We have a local dataset
*/
List<File> files = null;
try {
files = CdmUtils.expandGlobExpression(location);
} catch (NullPointerException e) {
System.out.println("NPE processing location: " + location);
throw e;
}
if (files.size() == 0) {
throw new EdalException(
"The location " + location + " doesn't refer to any existing files.");
}
if (files.size() == 1) {
location = files.get(0).getAbsolutePath();
nc = CdmUtils.openDataset(location);
} else {
/*
* We have multiple files in a glob expression. We write some
* NcML and use the NetCDF aggregation libs to parse this into
* an aggregated dataset.
*
* If we have already generated the ncML on a previous call,
* just use that.
*/
if (ncmlString == null) {
/*
* Find the name of the time dimension
*/
NetcdfDataset first = openAndAggregateDataset(files.get(0).getAbsolutePath());
String timeDimName = null;
for (Variable var : first.getVariables()) {
if (var.isCoordinateVariable()) {
for (Attribute attr : var.getAttributes()) {
if (attr.getFullName().equalsIgnoreCase("units")
&& attr.getStringValue().contains(" since ")) {
/*
* This is the time dimension. Since this is
* a co-ordinate variable, there is only 1
* dimension
*/
Dimension timeDimension = var.getDimension(0);
timeDimName = timeDimension.getFullName();
}
}
}
}
first.close();
if (timeDimName == null) {
throw new EdalException("Cannot join multiple files without time dimensions");
}
/*
* We can't assume that the glob expression will have
* returned the files in time order.
*
* We could assume that alphabetical == time ordered (and
* for properly named files it will - but let's not rely on
* our users having sensible naming conventions...
*
* Sort the list using a comparator which opens the file and
* gets the first value of the time dimension
*/
final String aggDimName = timeDimName;
Collections.sort(
files,
new Comparator<File>() {
@Override
public int compare(File ncFile1, File ncFile2) {
NetcdfFile nc1 = null;
NetcdfFile nc2 = null;
try {
nc1 = NetcdfFile.open(ncFile1.getAbsolutePath());
nc2 = NetcdfFile.open(ncFile2.getAbsolutePath());
Variable timeVar1 = nc1.findVariable(aggDimName);
Variable timeVar2 = nc2.findVariable(aggDimName);
long time1 = timeVar1.read().getLong(0);
long time2 = timeVar2.read().getLong(0);
return Long.compare(time1, time2);
} catch (Exception e) {
/*
* There was a problem reading the data. Sort
* alphanumerically by filename and hope for the
* best...
*
* This catches all exceptions because however
* it fails this is still our best option.
*
* If the error is a genuine problem, it'll show
* up as soon as we try and aggregate.
*/
return ncFile1.getAbsolutePath().compareTo(ncFile2.getAbsolutePath());
} finally {
if (nc1 != null) {
try {
nc1.close();
} catch (IOException e) {
log.error("Problem closing netcdf file", e);
}
}
if (nc2 != null) {
try {
nc2.close();
} catch (IOException e) {
log.error("Problem closing netcdf file", e);
}
}
}
}
});
/*
* Now create the NcML string and use it to create an
* aggregated dataset
*/
StringBuffer ncmlStringBuffer = new StringBuffer();
ncmlStringBuffer.append(
"<netcdf xmlns=\"http://www.unidata.ucar.edu/namespaces/netcdf/ncml-2.2\">");
ncmlStringBuffer.append(
"<aggregation dimName=\"" + timeDimName + "\" type=\"joinExisting\">");
for (File file : files) {
ncmlStringBuffer.append("<netcdf location=\"" + file.getAbsolutePath() + "\"/>");
}
ncmlStringBuffer.append("</aggregation>");
ncmlStringBuffer.append("</netcdf>");
ncmlString = ncmlStringBuffer.toString();
}
nc = NcMLReader.readNcML(new StringReader(ncmlString), null);
}
}
return nc;
}