A global near real-time filtered 500m 10-day FPAR dataset from MODIS and VIIRS instruments, suited for operational agricultural monitoring and crop yield forecasting

Climate change and extreme weather events pose challenges to food security, emphasizing the need for reliable and timely monitoring of crop and rangeland conditions. For this purpose, long-term consistent Earth Observation datasets on vegetation conditions are typically used in Early Warning and crop yield forecast systems. However, the near-real-time (NRT) use of such observations and the need to guarantee long-term records present various challenges. To address these, we present a NRT global dataset of Fraction of Photosynthetically Active Radiation (FPAR) at 500 m resolution, optimized for agricultural applications. Our dataset combines MODIS-FPAR (Collection 6.1) and VIIRS-FPAR (Collection 2) data, ensuring continuity from 2000 to well beyond 2030. We applied a robust filtering approach based on the Whittaker smoother to produce reliable FPAR estimates in NRT, accounting for sparse and irregular spaced observations due to cloud cover. The dataset is composed of two 10-day filtered timeseries: 1) MODIS-FPAR for 2000 to 2023, being the reference dataset, and 2) intercalibrated VIIRS-FPAR for 2018 onward. While several methods can effectively smooth and gap-fill FPAR data (i.e., using observations before and after the estimation date), our method is designed for optimal filtering in NRT (i.e., using only prior observations). Our approach yields six successive estimates of the same FPAR data point with increasing quality: a inital estimate immediately after the 10-day reference period, four subsequent estimates every 10 days using new observations, and a final consolidated estimate 90 days later. The implemented filtering ingests the available FPAR observations and their original quality assessment (QA) layers. To avoid unrealistic extrapolation when observations are sparse, we impose constraints, season and location specific, to FPAR estimates. We then intercalibrated the VIIRS-FPAR with the MODIS-FPAR filtered timeseries, using a mean difference correction approach, to ensure consistency between both series. This paper describes the filtering and intercalibration method used, the quality assessment of resulting timeseries, and a detailed description of the product and corresponding QA layers.