Abstract <p>Net primary productivity (NPP) of vegetation is a key indicator for assessing regional carbon sequestration capacity and ecological health. Addressing the limitations of traditional mechanistic models for estimating NPP in complex environments, this study proposes an innovative framework driven by both mechanisms and data, using Henan Province, China as a case study. This framework optimizes the water stress module of the CASA model by introducing the shortwave infrared moisture index (SIMI). A deep learning model integrating attention mechanisms with spatiotemporal sequence learning is constructed to achieve high-accuracy NPP estimation. Results show that the average annual NPP in Henan Province from 2001 to 2020 was <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(619.02\,\hbox {g}\cdot \hbox {m}^{-2}\)</EquationSource> </InlineEquation>, showing a significant upward trend and a spatial pattern of high in the center and low in the surrounding areas. Geographic detector analysis reveals that the synergistic effect of temperature and precipitation is the dominant driving force behind the spatial variation of NPP. This framework demonstrates superior performance in complex terrain, significantly improving estimation accuracy and providing a reliable methodological support and data foundation for regional carbon sink assessment, ecosystem management, and climate change response.</p> Graphical Abstract <p>This visual summary offers a concise overview of our research on vegetation net primary productivity(NPP)in Henan Province,China,a climatically transitional region with complex terrains and agricultural ecosystems.Data sources include meteorological data (temperature,precipitation),remote sensing products(MODIS NDVI,MOD17A3HGF NPP for validation),and socioeconomic data,enabling spatiotemporal analysis from 2001 to 2020.We employed an integrated approach:mechanism-driven analysis with the improved CASA-SIMI model(incorporating SIMI for water stress optimization)and data-driven learning via the CBAM-ConvLSTM framework(fusing attention mechanisms with spatiotemporal extraction).Key findings show an annual mean NPP of <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(619.02\,\hbox {g}\cdot \hbox {m}^{-2}\)</EquationSource> </InlineEquation> with a fluctuating upward trend, “high-center,low-periphery” spatial patterns, and temperature-precipitation synergy as dominant drivers.CBAM-ConvLSTM achieved superior performance(R²=0.832, reduced MAE).This abstract encapsulates our framework’s role in advancing NPP estimation,supporting regional carbon sink assessment and ecological policy optimization.</p>

A Data-Mechanism Integrated Approach for Spatiotemporal Patterns and Vegetation Net Primary Productivity (NPP) Dynamics :A Case Study in Henan Province, China

期刊论文
发表日期：

2026年3月17日

查看全文

Weidong Li,
Zhe Wang,
Jinlong Duan,
Zhenhua Jing,
Xuehai Zhang

摘要

Abstract

Net primary productivity (NPP) of vegetation is a key indicator for assessing regional carbon sequestration capacity and ecological health. Addressing the limitations of traditional mechanistic models for estimating NPP in complex environments, this study proposes an innovative framework driven by both mechanisms and data, using Henan Province, China as a case study. This framework optimizes the water stress module of the CASA model by introducing the shortwave infrared moisture index (SIMI). A deep learning model integrating attention mechanisms with spatiotemporal sequence learning is constructed to achieve high-accuracy NPP estimation. Results show that the average annual NPP in Henan Province from 2001 to 2020 was \(619.02\,\hbox {g}\cdot \hbox {m}^{-2}\) , showing a significant upward trend and a spatial pattern of high in the center and low in the surrounding areas. Geographic detector analysis reveals that the synergistic effect of temperature and precipitation is the dominant driving force behind the spatial variation of NPP. This framework demonstrates superior performance in complex terrain, significantly improving estimation accuracy and providing a reliable methodological support and data foundation for regional carbon sink assessment, ecosystem management, and climate change response.

Graphical Abstract

This visual summary offers a concise overview of our research on vegetation net primary productivity(NPP)in Henan Province,China,a climatically transitional region with complex terrains and agricultural ecosystems.Data sources include meteorological data (temperature,precipitation),remote sensing products(MODIS NDVI,MOD17A3HGF NPP for validation),and socioeconomic data,enabling spatiotemporal analysis from 2001 to 2020.We employed an integrated approach:mechanism-driven analysis with the improved CASA-SIMI model(incorporating SIMI for water stress optimization)and data-driven learning via the CBAM-ConvLSTM framework(fusing attention mechanisms with spatiotemporal extraction).Key findings show an annual mean NPP of \(619.02\,\hbox {g}\cdot \hbox {m}^{-2}\) with a fluctuating upward trend, “high-center,low-periphery” spatial patterns, and temperature-precipitation synergy as dominant drivers.CBAM-ConvLSTM achieved superior performance(R²=0.832, reduced MAE).This abstract encapsulates our framework’s role in advancing NPP estimation,supporting regional carbon sink assessment and ecological policy optimization.