RFC 4: Band Properties¶
Author: Andreas Janz
Last modified: 2022-01-10
It is proposed to implement functions or methods for reading and writing band properties from/to QGIS PAM.
So far we identified the following band properties to be required or useful:
- band name
- Spectral Properties (see RFC 2): wavelength, wavelength units, fwhm, bad band multiplier
- Temporal Properties (see RFC 3): start time, end time
- Data Properties: offset, scale, no data value
API support will be implemented in the enmapboxprocessing.rasterreader.RasterReader class.
GUI support will be implemented in the qps.layerconfigwidgets.rasterbands.RasterBandPropertiesConfigWidget class.
Band property management is a cruitial, but a not well support raster layer preparation step in QGIS and the EnMAP-Box.
We need for example:
- band name information to subset or match raster bands by name instead of band numbers, and for setting proper names for reports
- wavelength information for plotting spectral profiles
- time information for plotting temporal profiles
- bad band multipliers to exclude noisy bands from specific plotting/analysis tasks
- data offset and scale information for scaling data stored as integer into 0 to 1 reflectance range
Spectral and Temporal Properties is already detailed in RFC 2 and 3.
Note that band no data value handling is fully supported by QGIS API directly via QgsRasterDataProvider methods: setNoDataValue, setUserNoDataValue, setUseSourceNoDataValue, sourceHasNoDataValue, sourceNoDataValue, userNoDataValues, useSourceNoDataValue.
- Band name, offset and scale information can be accessed via QgsRasterLayer methods, but can’t be modified. This is especially limiting in GUI applications, where we usually have a read-only handle to a raster source that we aren’t supposed to modify.
The here proposed approach will integrate band property handling into QGIS PAM management. This allows to set/update band properties for QgsRasterLayer objects, which is critical for GUI applications. It also takes care of information stored as GDAL PAM.
We propose the following approach for fetching band-specific properties.
Note that fetching Spectral and Temporal Properties is already detailed in RFC 2 and 3.
Band-wise properties are fetched with the following priorisation.
1. Look at QGIS PAM band-level default-domain. This is mainly relevant for GUI applications, where we need to set/update band properties using QgsRasterLayer objects:
bandName: str = layer.customProperty('QGISPAM/band/42//name') offset: floast = layer.customProperty('QGISPAM/band/42//offset') scale: float = layer.customProperty('QGISPAM/band/42//scale')
Note that those information is only concidered by EnMAP-Box applications and algorithms and always ignored by QGIS and GDAL. To manifest those changes you are required to translate the layer to an intermediate VRT using the EnMAP-Box Translate raster layer algorithm, which will transfer all QGIS PAM information to GDAL PAM.
2. Use QgsRasterLayer and QgsRasterDataProvider methods for accessing the band name, offset and scale:: This is mainly relevant for processing algorithms:
bandName = layer.bandName(42) offset = layer.dataProvider().bandOffset(42) scale = layer.dataProvider().bandScale(42)
Note that when reading raster band data, the application or algorithm is responsible for properly applying the band offset and scale. Notice that GDAL won’t scale data automatically, when calling gdal.Band.ReadAsArray(). On the contrary, QGIS will automatically scale the data using the QgsRasterDataProvider.bandOffset and QgsRasterDataProvider.bandScale information, but will ignore potential modifications stored in QGIS PAM.
For that reason, we highly encourage the use of the RasterReader.array methode for reading raster data, which will take care of all the data scaling details.
- Guide line 1:
- If you need to set band-wise properties in a processing algorithm: set it to the GDAL PAM band-level default-domain or use approriate gdal.Band methods. This way, i) the information is accessible with the GDAL API, and ii) consecutive band subsetting via gdal.Translate and gdal.BuildVrt can easily copy the band domains to the destination dataset.
- Guide line 2:
- If you need to set/update band properties in a GUI application: set it to QGIS PAM. This is most flexible and secure. The band properties are i) available as custom layer properties, ii) stored in the QGIS project, and iii) can be saved to QML layer style files. Note that those modifications are only used by EnMAP-Box applications and algorithms, QGIS and GDAL will ignore it! To manifest those changes in GDAL PAM, translate the layer to an intermediate VRT using the EnMAP-Box Translate raster layer algorithm.
- Guide line 3:
- Do not update GDAL PAM *.aux.xml file, while the corresponding source is opened as a QgsRasterLayer in QGIS. QGIS will potentially overwrite any changes, when closing the layer.
Technically, we don’t need any new functions or methods, because we fully rely on existing QGIS/GDAL API functionality.
But, the handling of property keys, the assurance of fetching priorities, and proper data scaling, can be tedious and should be encapsulated in util functions or methods. An example implementation is given by the RasterReader class.
Note that Spectral and Temporal Properties is already detailed in RFC 2 and 3.
To query band properties for band 42, we can use:
from enmapboxprocessing.rasterreader import RasterReader reader = RasterReader(layer) bandName = reader.bandName(42) bandOffset = reader.bandOffset(42) bandScale = reader.bandScale(42)
To set band properties use:
reader.setBandName('band 42 (0.685000 Micrometers)', 42) reader.setBandOffset(0, 42) reader.setBandScale(10000, 42)
To read (scaled) band data use:
array = reader.array(bandList=) maskArray = reader.maskArray(array, bandList=)
Also check out the RasterReader unittests: https://bitbucket.org/hu-geomatics/enmap-box/src/develop/enmapboxprocessing/test/test_rasterreaderrfc4.py