Key terms used throughout this course, grouped by topic.
File Formats & Data Structure¶
HDF5 (Hierarchical Data Format 5) A self-describing binary file format that stores data as a hierarchy of groups and datasets inside a single file — like a file system within a file. Tanager-1 data is delivered in HDF5.
HDF-EOS5
An extension of HDF5 defined by NASA for Earth Observation data. It adds standardized structures for swath, grid, and point data, plus geolocation arrays. All Tanager-1 .h5 files follow this convention.
Group The HDF5 equivalent of a folder. Groups can contain datasets and other groups, forming a tree hierarchy.
Dataset The HDF5 equivalent of a file — a named array of data (e.g., the 3-D reflectance cube or the wavelength vector).
Attribute Metadata attached directly to a group or dataset in HDF5 (e.g., units, scale factor, description). Attributes are how HDF5 files are “self-describing.”
GeoTIFF A standard raster image format that embeds spatial reference information (coordinate system, geotransform) directly in a TIFF file. Used for Tanager-1 quality masks and visual assets.
ENVI-BIL A flat binary format used in professional remote sensing software (ENVI, GDAL). BIL = Band Interleaved by Line. TanagerSpec can export to this format.
STAC (SpatioTemporal Asset Catalog) A standard for describing geospatial data so it can be searched and accessed via web APIs. Planet’s Open Data STAC serves free Tanager-1 scenes.
Radiometry¶
Radiance (TOA Radiance) The physical energy measured by the satellite sensor, expressed in W·sr⁻¹·m⁻²·μm⁻¹. “Top-of-Atmosphere” because it includes atmospheric effects (haze, aerosols, water vapor).
Surface Reflectance (BOA Reflectance) The fraction of incoming sunlight that the ground surface reflects, after atmospheric correction removes haze and aerosols. A unitless ratio between 0 and 1. More stable across time and location than radiance.
Atmospheric Correction The processing step that converts TOA radiance to surface reflectance by modelling and removing atmospheric effects. Tanager-1 uses ISOFIT for this step.
Digital Number (DN) The raw integer value recorded by the sensor before any calibration. Must be converted to radiance using sensor-specific gain and offset coefficients.
Calibrated Radiance Radiance derived from DN using the sensor’s radiometric calibration coefficients. The first step from raw data to a physically meaningful measurement.
Geometry¶
Basic Scene (Sensor Geometry) Tanager-1 data delivered in the sensor’s native image space — not projected to a map. Each pixel has a lat/lon from the geolocation array, but the image is not on a cartographic grid.
Ortho Scene (Map Geometry) Tanager-1 data that has been orthorectified (terrain-corrected) and projected to a UTM map grid, ready for overlay with other projected layers.
Orthorectification Geometric correction that removes distortions caused by terrain relief and sensor perspective. Uses a Digital Elevation Model (DEM). Output pixels are on a regular map grid.
Geolocation Array A separate dataset in Basic HDF5 files containing per-pixel latitude and longitude values. Used to map sensor-space pixels to geographic coordinates.
UTM (Universal Transverse Mercator) A common cartographic projection that divides the Earth into 60 zones. Tanager-1 Ortho products are projected to the UTM zone of the scene.
GSD / Ground Sample Distance The physical size of one pixel on the ground. Tanager-1 has a native GSD of ~32 m at nadir; Ortho products use a 30 m grid.
Pushbroom Scanner A sensor design where the full swath is imaged at once with a linear detector array, building the image line by line as the satellite moves. Tanager-1 uses this design.
Spectral Concepts¶
VSWIR (Visible to Shortwave Infrared) The spectral range from approximately 380 nm to 2500 nm, covering visible light, near-infrared (NIR), and shortwave infrared (SWIR). Tanager-1 captures this full range.
Contiguous Bands Spectral bands that cover a wavelength range without gaps, sampling every ~5 nm. Enables detection of narrow absorption features invisible to broadband sensors.
Spectral Signature The unique pattern of reflectance across wavelengths for a material or surface type. Used to identify vegetation species, minerals, water quality, and more.
Absorption Feature A dip in a spectral signature caused by a chemical bond absorbing specific wavelengths (e.g., chlorophyll absorbs red light ~670 nm; water absorbs SWIR at ~1400 nm and ~1900 nm).
Red Edge The steep rise in vegetation reflectance between ~680 nm (red, absorbed by chlorophyll) and ~740 nm (NIR, reflected by leaf structure). A sensitive indicator of plant health.
NDVI (Normalized Difference Vegetation Index)
(NIR − Red) / (NIR + Red) — the most widely used spectral index for estimating vegetation density and health.
EVI (Enhanced Vegetation Index) An improved vegetation index that reduces atmospheric and soil background effects compared to NDVI, especially in high-biomass areas.
Narrow-Band Spectral Index A spectral index computed using specific narrow wavelength channels (possible with hyperspectral data), rather than the broad bands of multispectral sensors. Enables more precise characterization.
Preprocessing¶
Bad Bands Spectral bands with low signal quality, typically caused by strong atmospheric water-vapor absorption (around 1350–1450 nm and 1800–2000 nm) or detector noise. Removed before analysis.
Quality Mask / UDM (Usable Data Mask) A binary raster where each pixel is flagged as usable or invalid (cloud, shadow, saturated, etc.). Applied to exclude unreliable pixels before analysis.
Reflectance Clipping Constraining surface reflectance values to a physically valid range (typically 0–1 or 0–10000 in scaled integer form) to remove outliers caused by noise or calibration errors.
Denoising Reducing random noise in hyperspectral data while preserving spectral features. PCA-based denoising discards components that mostly capture noise (low-variance components).
Analysis & Machine Learning¶
PCA (Principal Component Analysis) A dimensionality-reduction method that transforms correlated bands into uncorrelated principal components (PCs) ordered by variance explained. The first few PCs capture most of the scene’s information.
Dimensionality Reduction Reducing the number of bands (or features) from hundreds to a smaller set while retaining the most informative variation. Essential for efficient machine learning on hyperspectral data.
K-Means Clustering An unsupervised algorithm that groups pixels into k clusters based on spectral similarity. Used for scene segmentation and land-cover mapping without labeled training data.
Random Forest A supervised ensemble learning method that combines many decision trees. Commonly used for land-cover classification from spectral features.
Spectral Library A collection of reference spectra for known materials (soil types, vegetation species, minerals, water). Used for supervised classification and spectral matching.
Tools & Packages¶
TanagerSpec Planet Labs’ purpose-built Python package for end-to-end Tanager-1 hyperspectral analysis. Covers loading, preprocessing, export, visualization, spectral indices, and machine learning.
h5py
The standard Python library for reading and writing HDF5 files. Used throughout Modules 1–4 to navigate and extract data from Tanager-1 .h5 files.
ISOFIT An atmospheric correction algorithm developed at NASA JPL, used to derive Tanager-1 surface reflectance products from radiance.
GDAL Geospatial Data Abstraction Library — the underlying engine for reading and writing most geospatial raster and vector formats, including GeoTIFF.