SUPER v2: A 3-Hourly Global Precipitation Dataset Optimized for Sparse Data Challenges

Abstract. The Statistical Uncertainty analysis-based Precipitation mERging (SUPER) methodology can optimally merge different precipitation datasets with minimal use of ground-based information and is therefore better suited for data-sparse regions. Although a proof-of-concept SUPER framework has already been introduced previously, it contains substantial uncertainties and is only available at a daily timescale, which is inadequate for land surface modeling. In response, we present here a new 3-hourly, 0.1-degree, global SUPER version 2 (v2) dataset, spanning 2000–2023. SUPER v2 is unique in three key aspects: i) it optimizes the number of input precipitation datasets, which reduces data redundancy and mitigates negative biases in extreme precipitation events; ii) it optimally evaluates its internal merging weights and filters out false-alarmed events without reliance on extensive gauge networks; and iii) it employs a multi-scale (i.e., monthly–daily–3-hourly) temporal correction/merging procedure that enhances the robustness of precipitation estimates. The SUPER v2 product is comprehensively evaluated using 5,972 independent gauges. Results show that it has a root-mean-squared-error of 3.64 mm d⁻¹ and correlation coefficient of 0.68 [-] for daily precipitation estimates. These error metrics outperform traditional approaches over 81 % to 86 % of the validation gauges. The superiority of SUPER v2 with regards to rain/no-rain classification skill is even more evident, with Heidke’s Skill Score 22 % higher than commonly used datasets. Similar findings are also demonstrated in the 3-hourly SUPER v2 precipitation dataset. As such, SUPER v2 provides a unique opportunity for enhancing global-scale hydrology and land surface modeling — particularly for data sparse regions.