Assessment of gap-filling techniques applied to satellite phytoplankton composition products for the Atlantic Ocean

• DOI: 10.5194/egusphere-2025-112
• Publication Date: 2025-03-12T09:59:29Z
• Authors: Ehsan Mehdipour; Hongyan Xi; Alexander Barth; Aida Alvera-Azcárate; Adalbert Wilhelm; Astrid Bracher

Abstract. Phytoplankton are vital to marine biogeochemical cycles and form the base of food web. Comprehensive datasets offering a spatiotemporal perspective on phytoplankton composition essential for assessing impacts climate change ecosystems. functional types (PFTs) classify based their functions, enabling assessments nutrient cycling, primary productivity, ecosystem structure. However, satellite-derived ocean colour products like PFTs chlorophyll-a (Chla) concentrations challenged by limited temporal spatial coverage due exclusion data collected under non-optimal observing conditions such as strong sun glint, clouds, thick aerosols, straylight, large viewing angles or specific sensor configuration malfunction. This highlights importance gap-filling techniques producing consistent datasets, which currently missing operational sets. study evaluates two robust methods satellite observations: Data Interpolating Empirical Orthogonal Functions (DINEOF) Convolutional Auto Encoder (DINCAE). These were applied Sentinel 3A/B OLCI-derived Chla concentration in several regions Atlantic Ocean over three years data, including total (TChla) five major PFTs, namely diatoms, dinoflagellates, haptophytes, green algae, prokaryotic phytoplankton. The reconstructed assessed using test dataset evaluation validated with situ measurements during transatlantic RV Polarstern expedition PS113 2018. indicates that DINCAE outperforms DINEOF, particularly capturing transient-scale features. achieves an average root-mean-square-logarithmic-error (RMSLE) cross-validation is 66 % lower TChla 16 compared DINEOF. external validation better performance DINEOF than DINCAE, improved regression metrics 12.5 slope, 13.6 intercept, 68 higher coefficient determination (R²). gap-filled exhibit slightly reduced but still accuracy original while preserving statistical trends, improving structure restoration, increasing matchup validation. It concluded each have unique strengths products. performs well complex water bodies, effectively reproducing patterns from product. In contrast, shows overall reliability, supported independent validation, suited larger areas its computational demands.