Improving temporal stability in geostationary land surface temperature for nocturnal cooling analysis in agricultural systems

This is the authors' abstract. We don't add key points for this paper.

Geostationary land surface temperature (LST) provides the temporal resolution required to characterise nocturnal cooling across agricultural landscapes. However, cooling metrics derived from LST can be affected by transient variability that disrupts the temporal structure of the signal. Existing approaches such as cloud-masking, temporal smoothing, and gap-filling address cloud classification, noise reduction, or data availability, but do not directly evaluate temporal structure and stability within the retained LST observations. We present ClearNights, a preprocessing framework that quantifies transient variability in geostationary LST and identifies intervals exhibiting sustained low or elevated short-term variability. Temporal-stability screening is implemented using rolling-window statistics and an operational threshold applied directly to the LST time-series, without requiring auxiliary environmental information. The framework was evaluated using three complementary lines of evidence. Firstly, variability regimes identified by ClearNights corresponded with a ceilometer-derived cloud occurrence, achieving a median area-under-curve value of 0.80 across 912 site-months. Secondly, temporal-stability screening reduced extreme cooling-rate values and improved agreement between satellite-derived and ground-observed cooling metrics relative to unfiltered, temporally smoothed, and cloud-masked LST. Thirdly, compared to smoothing and cloud-masking approaches, ClearNights produced substantial improvements in cooling metrics from non-filtering under disturbed nocturnal conditions and greater cross-scale consistency with ground measurements within the instrumented network. Strong agreement in minimum overnight temperature did not always imply equally strong agreement in cooling rates. These results indicate that temporal stability represents an observable property distinct from cloud contamination and local fluctuations and temporal-stability screening can improve the robustness of cooling metrics derived from geostationary LST. ClearNights’ contribution is in identifying and screening transient variability within the temperature signal, which is separate from and upstream of event detection and impact assessment. Downstream interpretation of retained observations remains conditional on scale consistency and process context.