China Natural Forest 500-m 8-day Gap-filled Gross Primary Productivity Dataset, 2005–2020

CN-NF-MOD17GPP-Fill v1.0 is a quality-controlled and gap-filled gross primary productivity (GPP) dataset developed for China’s natural forests. The dataset is derived from the MODIS MOD17A2HGF Version 6.1 product, using both the Gpp_500m science data layer and the associated Psn_QC_500m quality-control layer. It provides a temporally complete 500-m, 8-day GPP time series for the constant natural forest extent of China from 2005 to 2020. The original MOD17A2HGF product provides 46 8-day composites per year, yielding 736 temporal layers over the 16-year study period. Raw integer GPP values were converted to physical units using the product scale factor, screened using bitwise interpretation of the Psn_QC layer, and spatially constrained to the constant natural forest mask. Continuous GPP values and QC layers were processed separately to preserve their respective data characteristics. Missing values caused by quality-control masking, cloud contamination, or data gaps were reconstructed using a hierarchical pixel-level gap-filling algorithm. Valid observations were retained unchanged. One-step gaps were filled by linear interpolation, two- to three-step gaps were reconstructed using Savitzky–Golay smoothing, and longer gaps were filled using a pixel-level climatological estimate. When pixel-level observations were insufficient, the algorithm progressively fell back to neighbouring DOY windows, the pixel-wise temporal median, and finally the forest-wide DOY climatology. Each output value is accompanied by a provenance-oriented quality code that identifies whether the value is observed, smoothed, climatology-filled, or interpolated. This design allows users to filter or weight observations according to their reconstruction reliability. The dataset is designed to support studies of forest productivity dynamics, phenological transitions, drought responses, ecosystem resilience, and cross-product robustness analyses in China’s natural forests. CN-NF-MOD17GPP-Fill v1.0 should be described as a quality-controlled and gap-filled derivative dataset, not as an independent ground-truth validation product.CN-NF-MOD17GPP-Fill v1.0是一款针对中国天然森林研发的经过质量控制与间隙填充的总初级生产力（Gross Primary Productivity，GPP）数据集。该数据集源自MODIS MOD17A2HGF 6.1版产品，同时使用了其中的Gpp_500m科学数据层与配套的Psn_QC_500m质量控制层。它为2005至2020年中国恒定天然森林范围提供了时间上完整的500米分辨率、8天合成的GPP时间序列。原始MOD17A2HGF产品每年提供46幅8天合成数据，在本16年研究周期内共计生成736个时间层。原始整数型GPP数值通过产品缩放因子转换为物理单位，借助Psn_QC_500m层的按位解析结果进行质量筛选，并通过空间约束限定至恒定天然森林掩膜范围。连续型GPP数值与质量控制层分别进行处理，以保留各自的数据特性。 因质量控制掩膜、云污染或数据间隙产生的缺失值，通过分层像素级间隙填充算法进行重构，有效观测值则保持不变。单步间隙采用线性插值填充，2至3步间隙通过萨维茨基-戈莱（Savitzky–Golay）平滑进行重构，更长时间的间隙则采用像素级气候学估计值填充。当像素级观测数据不足时，该算法会逐级回退，依次采用邻近年积日（Day of Year，DOY）窗口、像素级时间中位数，最终采用全森林范围的年积日气候学数据。每个输出值均附带面向数据来源的质量编码，用于标识该数值属于观测值、平滑值、气候学填充值还是插值值。该设计允许用户根据数据重构的可靠性对观测值进行筛选或加权。 本数据集旨在支撑中国天然森林的森林生产力动态、物候转变、干旱响应、生态系统恢复力以及跨产品稳健性分析等相关研究。 CN-NF-MOD17GPP-Fill v1.0应被归类为经过质量控制与间隙填充的衍生数据集，而非独立的地面真值验证产品。