Determining the height of a volcanic SO2 cloud after a volcanic eruption is a challenging task in UV satellite retrievals. The height is nevertheless the most important yet uncertain parameter required to forecast the movement of the volcanic cloud and to determine the total SO2 column and ejected SO2 mass, especially for local authorities and aviation safety applications. Retrieval algorithms developed so far use direct fitting and optimal estimation techniques to determine the height information, which is hidden in the spectral signature. They are computationally very expensive and time consuming and therefore are not practical in near-real time operational retrievals, especially for current and future satellite UV instruments with high resolution and related high data amount. We have therefore developed the ‘Full-Physics Inverse Learning Machine’ (FP_ILM) retrieval algorithm that combines principle component analysis and neural network , which performs an extremely fast (3 ms per TROPOMI pixel) yet accurate (<2km average accuracy) SO2 LH retrieval based on UV satellite measurements. The algorithm was first applied to GOME-2 and introduced by Efremenko et al. (2017) . Hedelt et al. (2019) further improved the algorithm and applied it to Sentinel-5p/TROPOMI data. The algorithm was optimized and validated in the framework of ESA’s Sentinel-5p Innovations project (S5P+I) and is already performing SO2 LH retrievals in a semi-operational near-real time environment. Recently, Fedkin et al. (2021) applied the FP_ILM algorithm to OMI and OMPS data. Application to future UV LEO (Sentinel-5) and GEO (Sentinel-4, GEMS, TEMPO) satellite missions will follow. We present here SO2 LH results based on GOME-2, OMI/OMPS, and TROPOMI measurements of the Raikoke (Kuril Islands) volcanic eruption in June-July 2019 and La Soufriere St. Vincent volcanic eruption in April 2021 and make intercomparisons between the UV sensors as well as other (IR) independent measurements.