Global aboveground and belowground Net Primary Productivity grids with 0.05° resolution from 1981 to 2018 using the Deep Learning model

Net primary productivity (NPP) in global ecosystems is a critical component of terrestrial carbon cycling and an essential indicator of ecosystem carbon sequestration capacity. However, the size and spatial patterns of NPP across global ecosystems remain uncertain, particularly for the divided estimation of above- and below-ground net primary productivity (ANPP and BNPP), which limits our understanding of global NPP carbon cycling and the sustainability of carbon sequestration. Here, an updated database was compiled with 5,184 filed observations to estimate the spatial patterns of ANPP and BNPP using a modified multilayer perceptron neural network (MLP) method. Based on the 10-fold cross-validation,<i> </i>the model performance<i> </i>of ANPP and BNPP were 0.74 (<i>R</i>^<em>2</em><i>, RMSE</i>: 138.37 g C m^-2 yr^-1) and 0.73 (<i>R</i>^<em>2</em><i>, RMSE</i>: 76.33 g C m^-2 yr^-1), respectively. The total ANPP and BNPP were 33.4 ± 0.5 and 19.2 ± 0.73 Pg C yr^-1, respectively. ANPP and BNPP exhibited decreasing trends from low latitudes toward the poles. Over the last 38 years, global 65.38% and 60.83% areas of the ANPP and BNPP showed a growth trend, and annual total ANPP and BNPP were significant positive growth trends (<i>p</i> &lt; 0.01), about 0.02 and 0.05 Pg C yr^-1, respectively. Our findings suggest that field data-driven prediction of NPP using the MLP model could be a useful approach for obtaining spatially explicit NPP of ecosystems, particularly BNPP. This study contributes to the understanding of the dynamics of terrestrial carbon sequestration in the context of global change and may provide a new method and basis for making climate change response proposals.Filename: ANPP_MLP_1981_2018_0.05.zip; BNPP_MLP_1981_2018_0.05.zipFile information：‘ANPP_MLP_1981_2018_0.05.zip’ is the predicted global ANPP grid with 0.05° spatial resolution from 1981 to 2018; ‘BNPP_MLP_1981_2018_0.05.zip’ is the predicted global BNPP grid with 0.05° spatial resolution from 1981 to 2018. The unit for these data is 'g C m^-2a^-1'. Besides, both zip files contain 38 grid images, and each grid image is named according to the following convention: 'data type' _ 'model type' _ 'year' _ 'spatial resolution'.tif. For example, the ANPP data for the year 1981 would be named:' ANPP_MLP_1981_0.05.tif'.<b>Corresponding author</b>: Xiaolu Tang ([e-mail])

全球生态系统的净初级生产力（Net primary productivity, NPP）是陆地碳循环的核心组成部分，亦是衡量生态系统固碳能力的关键指标。然而，全球生态系统中NPP的总量与空间格局仍存在不确定性，尤其是地上净初级生产力（above-ground net primary productivity, ANPP）与地下净初级生产力（below-ground net primary productivity, BNPP）的拆分估算，这极大限制了我们对全球NPP碳循环过程以及固碳可持续性的认知。

本研究整合了5184项野外观测数据，构建了更新后的数据库，并采用改进的多层感知器神经网络（modified multilayer perceptron neural network, MLP）方法，估算了ANPP与BNPP的空间分布格局。基于10折交叉验证，ANPP与BNPP的模型性能分别为0.74（决定系数R²，均方根误差RMSE：138.37 g C m⁻² yr⁻¹）和0.73（R²，RMSE：76.33 g C m⁻² yr⁻¹）。全球总ANPP与BNPP分别为33.4±0.5和19.2±0.73 Pg C yr⁻¹。

ANPP与BNPP均呈现出从低纬度向极地递减的空间分布趋势。在过去38年间，全球ANPP和BNPP分别有65.38%和60.83%的区域呈现增长趋势，且年总ANPP与BNPP均表现出显著的正向增长趋势（p < 0.01），年增长率分别约为0.02和0.05 Pg C yr⁻¹。

本研究结果表明，依托野外数据、采用MLP模型预测NPP，是获取生态系统空间显式NPP的有效途径，尤其适用于BNPP的估算。本研究有助于深化对全球变化背景下陆地固碳动态的理解，可为制定气候变化应对策略提供全新的方法与理论依据。

文件名：ANPP_MLP_1981_2018_0.05.zip；BNPP_MLP_1981_2018_0.05.zip
文件说明：‘ANPP_MLP_1981_2018_0.05.zip’为1981—2018年全球0.05°空间分辨率ANPP预测栅格数据集；‘BNPP_MLP_1981_2018_0.05.zip’为1981—2018年全球0.05°空间分辨率BNPP预测栅格数据集。上述数据的单位为“g C m⁻² a⁻¹”。此外，两个压缩包均包含38幅栅格图像，每幅图像的命名遵循以下规则：“数据类型_模型类型_年份_空间分辨率.tif”。例如，1981年的ANPP数据命名为：ANPP_MLP_1981_0.05.tif。

通讯作者：唐晓璐（[e-mail]）