The study, conducted by Andrzej Z. Kotarba of the Space Research Centre at the Polish Academy of Sciences, evaluated a cloud-top height retrieval method first proposed by Šoljan and colleagues in 2024. The technique estimates how tall a storm cloud is by matching the brightness temperature measured in the 11-micrometre infrared window channel against the temperature of a rising parcel of moist air. Deep convective clouds develop along what meteorologists call the moist adiabat, the characteristic cooling curve followed by saturated air as it ascends. If the satellite can measure how cold the cloud top is, and the atmosphere’s moist adiabat is known, the cloud-top height follows directly. The innovation lies in how the adiabat is handled: instead of computing it iteratively, which is computationally expensive, the method approximates it with compact polynomials requiring only 30 coefficients, making it fast enough to process enormous archives of satellite data.

The mathematical machinery builds on earlier work by Bakhshaii and Stull, who devised a non-iterative way to compute saturated pseudoadiabats, and by Moisseeva and Stull, who refined the approximation using high-order polynomials with 231 coefficients. The newer method deliberately trades a little precision for speed, reducing the polynomial orders to fifth and fourth degree. The trade-off limits its use to cloud tops warmer than minus 75 degrees Celsius, but that range comfortably covers deep convective clouds. According to its developers, the approximation estimates cloud-top pressure with a maximum error of just 28 metres compared with a full iterative calculation. Crucially, the method needs only three standard meteorological inputs, air temperature, dew point temperature and pressure, which are freely available from reanalysis datasets such as ERA5, taken from the most unstable atmospheric level between 1000 and 700 hectopascals.

To find out whether this elegant shortcut could stand in for operational cloud-top height products, Kotarba ran two independent validation experiments. The first exploited a rare alignment of fortune in Earth observation: between 2006 and 2011, the Aqua satellite flew in close formation with CloudSat and CALIPSO, a pair of satellites carrying a cloud-profiling radar and a sensitive lidar. Together these active instruments provide some of the most accurate measurements of cloud vertical structure ever made from space. By matching roughly 1.7 million deep convective cloud observations from MODIS, Aqua’s imaging radiometer, with lidar-radar profiles collected about a minute later, the study assembled a rigorous reference dataset spanning both the tropics and the mid-latitudes.

The verdict was nuanced. Compared with the lidar-radar truth, the polynomial method underestimated cloud-top height by 2.7 kilometres on average, about 18 percent, placing typical storm tops at 12.5 kilometres rather than 15.2. Against MODIS’s own operational retrieval the bias shrank threefold to 0.9 kilometres, but that comparison flatters the method, because MODIS itself is known to underestimate the height of high clouds by a kilometre or more. Part of the discrepancy has a clear physical explanation: the 11-micrometre brightness temperature averaged 2.4 kelvin warmer than the true cloud-top temperature, because the signal mixes radiation from the cloud top with warmer emission from atmospheric layers below. Near the tropopause, where temperature barely falls with height, that small temperature error can translate into an altitude error exceeding a kilometre.

The good news is that the bias is systematic and therefore correctable. By fitting a simple linear regression between the estimated and reference heights, averaged over one-degree latitude bands, the study showed that a straightforward bias adjustment cuts the normalized error to below 7 percent and the mean absolute error to under 1.2 kilometres. Because this calibration is a monotonic, one-to-one transformation, it rescales the retrieved heights without distorting or smoothing genuine physical signals, preserving any trends or spatial gradients in the data. The second validation, comparing the method against EUMETSAT’s operational CLAAS-3 cloud product derived from the SEVIRI imager on Meteosat Second Generation over two decades, found a bias of 1.1 kilometres, or about 10 percent, sitting neatly between the MODIS and lidar-radar results.

The decisive test, however, was not how well the method estimated heights in the abstract, but whether those estimates were good enough to fix parallax. Applying the technique to SEVIRI observations over a pan-European domain during the summer of 2005, the study computed parallax correction vectors and compared them with those derived from the operational CLAAS product. The reference corrections displaced deep convective cloud tops by an average of 32.5 kilometres, a striking illustration of how badly uncorrected imagery misplaces storms. The unadjusted polynomial method produced vectors just 1.3 kilometres shorter on average, a relative error of only 6 percent. At the level that actually matters for mapping, 84 percent of cloud pixels ended up in exactly the same image location as with the operational correction, rising to 97 percent once the CLAAS product’s own retrieval uncertainty was taken into account.

The climatic stakes of this apparently technical detail are considerable. Because deep convective clouds are relatively rare, even small geolocation errors translate into large relative uncertainties in how often they occur. The study found that skipping parallax correction altogether introduces errors of 13 to 20 percent into monthly and seasonal estimates of deep convective cloud frequency over Europe. Using the polynomial method reduced those errors to below 7 percent monthly and below 5 percent seasonally, comparable to what the operational product achieves. Since the method depends only on an infrared window channel and freely available reanalysis data, it works identically for the earliest Meteosat first-generation sensors, the current second-generation SEVIRI, and the third-generation imagers now entering service, none of which require multispectral absorption bands, stereo viewing or machine-learning training data.

That generational neutrality is the study’s most consequential finding. Climate research on thunderstorms has long been hampered by the difficulty of building homogeneous long-term records from satellites whose instruments and algorithms have changed repeatedly over the decades. A single, uniform parallax correction method, applicable to more than 40 years of geostationary observations, opens the door to consistent storm-cloud climatologies spanning the entire satellite era, not only for Meteosat but potentially for other geostationary families such as GOES, Himawari and the Chinese Fengyun satellites, all of which carry the same basic infrared window channel. The method is not perfect: its tropical performance against the operational product was weaker, and the study notes that calibration coefficients derived from short reference periods may not generalise across seasons. But as a pragmatic tool for turning four decades of raw thermal imagery into a physically consistent record of where, and how often, the planet’s most violent clouds truly are, a handful of polynomial coefficients may prove to be worth their weight in gold.

Subject of Research: Validation of a moist-adiabat cloud-top height retrieval method for parallax correction of deep convective clouds in multi-generation Meteosat satellite records

Article Title: Evaluation of a moist-adiabat cloud-top height retrieval for parallax correction of deep convective clouds across Meteosat generations

Article References: Kotarba, A. Z. (2026). Evaluation of a moist-adiabat cloud-top height retrieval for parallax correction of deep convective clouds across Meteosat generations. Atmospheric Measurement Techniques, 19(19), 6193-6208. https://doi.org/10.5194/amt-19-6193-2026

Image Credits: AI Generated

DOI: 10.5194/amt-19-6193-2026

Keywords: deep convective clouds, cloud-top height, parallax correction, Meteosat, SEVIRI, MODIS, CloudSat, CALIPSO, moist adiabat, infrared brightness temperature, satellite climatology, ERA5 reanalysis