Looking up, looking down: zenith-to-nadir lidar comparisons for cloud scenes (IPRAL and ATLID/EarthCARE)

Abstract. Intercomparisons between spaceborne lidar instruments and ground-based lidars are essential for calibration/validation of these instruments and for data quality assurance. However, zenith-to-nadir viewing geometry differences can create systematic discrepancies between space-borne and ground-based observations, even under perfect spatiotemporal collocation. This is particularly true for clouds: unlike aerosols, which are comparatively well-mixed and spatially homogeneous, clouds are structured and short-lived, so both the observation geometry and the collocation quality strongly affect whether a spaceborne and a ground-based profile can be meaningfully compared. Still, clouds are valuable targets because they push the signal into a different dynamic range, from strong backscatter to near-saturation, that clear-sky or aerosol scenes never exercise. The present paper addresses three questions: which parameters retrieved from lidar observations are better suited to zenith-to-nadir comparisons, which atmospheric scenes are favorable for such a comparison, and how the choice of collocation approach affects the comparison's reliability. We address these questions with a simulation framework that generates 15 synthetic cloud scenes, propagates them through the lidar simulator of the Cloud Feedback Model Intercomparison Project (CFMIP) Observation Simulator Package, version 2 (COSP2), for both nadir and zenith geometries, and adds instrument-specific daytime/nighttime noise. We compare attenuated backscatter (Level 1) and retrieved particulate backscatter βpart(z) (Level 2) using two scattering-ratio (SR) definitions, and introduce a Comparability Score (CS) quantifying profile similarity via correlation and shape. Thin clouds yield high CS and comparable zenith/nadir measurements; thicker and multi-layer clouds do not, due to directional attenuation and retrieval instabilities. We also found that βpart(z) comparisons remain valid over a broader range of conditions than SR(z). In real collocations, spatiotemporal separation is rarely negligible, so we select the best-matched profile pair by maximizing cross-correlation. We estimate this method's false discovery rate (the fraction of selected pairs that do not sample the same cloud) at 10–20%, using both a theoretical null model and an empirical one built from mismatched-date collocations. Applied to real collocations at 355 nm between the Atmospheric Lidar (ATLID) on the Earth Cloud, Aerosol and Radiation Explorer (EarthCARE) and the IPSL Atmospheric Research Lidar (IPRAL) across four representative cases, βpart(z) again agrees better than native-geometry SR(z), consistent with the simulation results. We conclude that a spaceborne-to-ground-based cloud lidar comparison is meaningful, that is, not dominated by viewing geometry or by mismatched scenes, when three conditions are met: (1) candidate collocations are screened using the correlation coefficient and cloud vertical structure, rather than accepted by spatiotemporal proximity alone; (2) the collocated scene has a Comparability Score above 50; (3) the comparison variable, SR(z) or βpart(z) is chosen according to the scene's optical depth: both variables are comparable below τ = 0.7 and only βpart(z) is comparable above this threshold. We propose this decision scheme, with its threshold values, as a starting point for other spaceborne-to-ground-based cloud lidar validations.