STORED VALUE
Integer delivered for efficient storage
Standalone advanced lesson · Thermal infrared
From emitted radiance to a defensible surface-temperature claim
Separate what the detector receives from what a product retrieves—then carry emissivity, atmosphere, time, spatial support and validation into every ecological interpretation.
This lesson uses an official Landsat decoding example and clearly labelled synthetic teaching records. It does not reproduce a real heat event, diagnose vegetation stress, or provide a universal calibration for satellite or UAV sensors.
Measurement physics
A thermal pixel is an energy balance observed through an atmosphere
In the thermal infrared, the surface emits energy according to temperature and emissivity. It can also reflect downwelling atmospheric radiance. The atmosphere transmits part of the surface-leaving signal and adds path radiance before the detector records it.
Lsensor = τ[εB(Ts) + (1−ε)L↓] + L↑
This conceptual radiative-transfer relation names the dependencies; operational retrievals use sensor response functions, calibrated radiance and product-specific algorithms.
SENSOR · LsensorL↑ atmosphereτ transmissionATMOSPHEREεB(Ts) emission ↑L↓ reflected by surface ↕SURFACE · Ts · ε
Do not collapse the processing ladder
Five quantities that answer different questions
AT-SENSOR RADIANCE
Thermal energy reaching the detector
BRIGHTNESS TEMPERATURE
Blackbody-equivalent temperature inferred from radiance
LAND-SURFACE TEMPERATURE
Retrieved radiometric skin temperature of the observed surface
ECOLOGICAL MODEL OUTPUT
Stress, ET or another response estimated using thermal and supporting evidence
Interactive radiance reasoning
Temperature is not the only control on the signal leaving a surface
Adjust temperature and emissivity. The bars are a dimensionless teaching proxy, not a sensor calibration or land-surface-temperature retrieval.
Choose by question and product—not by colour
Satellite and UAV thermal evidence have different support
SOURCEWHAT THE PRODUCT REPRESENTSSUPPORTVALUELIMIT
Surface temperature in Kelvin after decoding
Scene product; inspect native thermal/product support and grid
Long, consistently processed archive
Single-channel retrieval uses emissivity and atmospheric auxiliary data; cloud-adjacent errors and emissivity gaps/issues remain.
Land-surface temperature and emissivity retrieved across five active TIR bands
Swath or 70 m gridded/tiled products
Different overpass times from the ISS support diurnal investigations
Use product cloud/quality information and current collection documentation; a 70 m pixel is a mixed surface.
Sensor-specific radiance or apparent-temperature-related imagery
Fine nominal pixels within a flight footprint
Local detail and flexible campaign timing
Calibration, emissivity, atmosphere, drift, viewing geometry and ground validation do not disappear at fine resolution.
Product-specific decoding
A stored integer is not yet a temperature
For the USGS Landsat Collection 2 Level-2 surface-temperature band, use the documented scale and offset below. Do not copy this conversion to another product or collection.
ST (K) = DN × 0.00341802 + 149.0- Kelvin
- 302.6 K
- Celsius
- 29.5 °C
The default DN 44,947 reproduces the USGS example: 302.6 K, or 29.5 °C after Kelvin-to-Celsius conversion.
End-to-end procedure
Question → radiance/product → valid temperature → qualified inference
Each gate protects the next. A beautiful heat map cannot repair a product-level, emissivity, cloud, timing or validation failure.
Define the temperature claim before selecting a product
ACTION
- Name the surface, time window, spatial support and comparison of interest.
- Write what would count as independent validation.
- State non-claims such as air temperature, soil moisture, evapotranspiration or plant stress.
CHECK
- The target is surface temperature or thermal radiance—not a colour on a map.
- The expected temperature contrast can exist at the sensor's time and scale.
OUTPUT
A bounded thermal question and validation plan.
Choose a product level that matches the question
ACTION
- Distinguish raw counts, calibrated radiance, brightness temperature and land-surface-temperature products.
- Select Landsat, ECOSTRESS, UAV or another source from coverage, acquisition time, resolution and product maturity.
CHECK
- The variable, units, scale/offset, valid range and QA layers are documented.
- Product version and processing level are recorded.
OUTPUT
A product decision with one intended use and explicit limitations.
Inventory acquisition and processing metadata
ACTION
- Record product ID, acquisition and processing time, sensor, view geometry, CRS, grid and native resolution.
- Keep source QA, emissivity and uncertainty layers with the temperature data.
CHECK
- UTC/local time conversion is explicit.
- No layer has been separated from the product version that defines it.
OUTPUT
A traceable source inventory.
Apply scale, offset, fill and valid-range rules
ACTION
- Mask fill and invalid values before arithmetic.
- Decode stored integers using the exact metadata for that product and collection.
- Convert Kelvin to Celsius only after the physical temperature value exists.
CHECK
- Sample pixels reproduce a documented example or metadata calculation.
- Units survive export and plotting.
OUTPUT
A physically labelled temperature layer plus a decode audit.
Build a joint validity mask
ACTION
- Decode cloud, cloud-shadow, fill, saturation and product-specific quality bits.
- Exclude or stratify low-quality emissivity and uncertainty evidence where the product provides it.
CHECK
- The mask uses the same grid and shape as temperature.
- Valid fraction and exclusion reasons are reported—not hidden.
OUTPUT
A reproducible valid-pixel mask and exclusion summary.
Audit emissivity and atmospheric assumptions
ACTION
- Identify how emissivity and atmospheric transmittance/radiance enter the supplied retrieval.
- Inspect known gaps, blockiness or uncertainty relevant to the product.
- For custom retrievals, document the radiative-transfer method and every auxiliary input.
CHECK
- Brightness temperature is not relabelled as surface temperature.
- Emissivity is treated as wavelength-, material- and condition-dependent.
OUTPUT
A retrieval-assumption record tied to the selected product.
Align space, time and measurement support
ACTION
- Keep native grids until a justified target grid is defined.
- Pair field observations within a defensible time window and footprint.
- Aggregate raster values to the field support rather than treating neighbouring pixels as independent plots.
CHECK
- CRS, transform, resolution, extent and NoData match after alignment.
- Reprojection and resampling methods suit continuous temperature data.
OUTPUT
A support-compatible temperature and reference table.
Control acquisition context before comparison
ACTION
- Record local solar time, season, weather, cloud history, wind and surface wetness where available.
- Compare like observation contexts or model those differences explicitly.
CHECK
- A warmer value is not explained from temperature alone.
- Day/night and pre-event conditions are visible in the analysis table.
OUTPUT
A comparable observation set or a documented reason it is not comparable.
Validate against independent, support-aware observations
ACTION
- Use calibrated reference measurements with timestamp, footprint and uncertainty.
- Predeclare matching and exclusion rules.
- Report bias, MAE/RMSE, sample count and error by surface/context when justified.
CHECK
- Validation data did not tune the final correction or threshold.
- Reference and raster support are compatible enough for the claim.
OUTPUT
A validation table, error summary and unresolved limitations.
Test a hypothesis without turning association into cause
ACTION
- Calculate contrasts, anomalies or models only inside valid support.
- Carry acquisition context and uncertainty into grouping or modelling.
- Use spatial or grouped validation when estimating predictive performance.
CHECK
- The result distinguishes observed temperature, derived contrast and modelled quantity.
- Alternative causes remain visible.
OUTPUT
A qualified thermal result with uncertainty and applicability domain.
Export an analysis-ready thermal evidence package
ACTION
- Write temperature, QA mask, context table, validation results and methods together.
- Preserve native and analysis grids separately.
- Publish a human-readable limitations and non-claims statement.
CHECK
- Files reopen with correct units, CRS, scale and NoData.
- Every map legend states the variable, units and time.
OUTPUT
A reviewable thermal package—not only a coloured image.
Time is part of the measurement
The same place can tell a different thermal story through one day
Incoming radiation, shading, wind, evaporation and thermal inertia change the surface energy balance. Comparisons must carry acquisition time and recent conditions.
Prediction before reveal
Which statement stays inside the evidence?
Ecological interpretation ladder
Add evidence before climbing from temperature to process
Valid radiance, brightness temperature or land-surface temperature at a stated time and support.
Anomaly, contrast, zonal summary or time-series feature calculated with explicit masks and context.
Evapotranspiration, water stress, habitat response or hazard class estimated with supporting predictors and validation.
A process explanation that remains conditional on design, alternatives, uncertainty and domain knowledge.
LAND-SURFACE TEMPERATURE ≠ AIR TEMPERATURE
WARMER ≠ AUTOMATICALLY DRIER OR STRESSED
A THERMAL PALETTE ≠ A THERMAL MEASUREMENT
Analysis-ready handoff
Release the evidence, not only the heat map
A defensible package keeps the source product, scale/offset, QA, masks, emissivity/atmospheric context, acquisition metadata, validation and limitations attached to the temperature values.
Authoritative references