The dataset includes spatially aggregated canopy temperature measurements (°C) of one to two tree crowns obtained using infrared (IR) point sensors.

The dataset was quality screened using a multi-step outlier detection procedure based on Median Absolute Deviation (MAD) statistics within a defined analysis period (= 1 year).

Three complementary quality-control checks were applied:

1. Global MAD screening – detects observations that strongly deviate from the overall distribution of each individual sensor time series.

2. Local spike detection – identifies short-term anomalous spikes within each sensor time series using rolling median and rolling MAD statistics.

3. Cross-sensor consistency screening – compares measurements across all sensors recorded at the same timestamp to identify extreme deviations relative to the network-wide median behaviour.

Each observation received individual QC flags for all three checks, followed by a combined final quality flag. Observations passing all criteria were labelled as good, while flagged observations were labelled as outlier if considered suspect, or as bad if associated with known sensor drifts. Observations labelled as outlier may still represent physically plausible values, while observations labelled as bad should not be considered for further analysis. All observations were retained in the exported dataset for traceability.