基于多源遥感数据的海南岛总初级生产力时空变化

Spatiotemporal Variation of Gross Primary Productivity in Hainan Island Based on Multi-Source Remote Sensing Data

摘要: 海南岛是我国最大的热带岛屿，也是全球生物多样性保护的热点地区之一。准确估算热带岛屿生态系统的总初级生产力(GPP)对理解全球碳循环至关重要，但目前仍存在较大不确定性。本研究集成GLASS、MOD17、FLUXCOM、VODCA2、VPM和改进的EC-LUE六种遥感GPP产品，系统评估了2001~2015年海南岛GPP的时空变化特征及其驱动机制。通过统一的时空预处理和多维度分析方法，发现海南岛GPP呈现显著的“核心–边缘”空间格局，中部热带雨林区域年均GPP达2829.78 gC m⁻² yr⁻¹，较全岛均值(1659~2138 gC m⁻² yr⁻¹)高32.6%，且年增长率最高达45 gC m⁻² yr⁻² (p < 0.05)。GPP空间分布具有显著的南北分异，以19˚N为界(R² = 0.73)，北部区域GPP年际波动剧烈(标准差 > 300 gC m⁻² yr⁻¹)，南部区域较为稳定(标准差 < 120 gC m⁻² yr⁻¹)。东北部城市化地区GPP呈现显著负增长(−30至−45 gC m⁻² yr⁻²)。不同遥感产品在估算值上存在系统性差异(范围479 gC m⁻² yr⁻¹)，光能利用率模型在雨季高估约15%，机器学习模型在旱季低估9.8%，主要受热带多云环境影响(云覆盖率75.44%)。融合微波与光学数据可降低估算误差12%~15%，为改善热带岛屿碳循环监测提供了新思路。

Abstract: Hainan Island, the largest tropical island in China, is a critical region for biodiversity conservation and a global hotspot for understanding ecological dynamics. Accurately estimating the Gross Primary Productivity (GPP) of tropical island ecosystems is fundamental to comprehending the global carbon cycle and its implications for climate change. However, substantial uncertainties persist in the current estimation methods, especially in regions characterized by complex tropical environments. This study integrates six remote sensing-based GPP products, namely GLASS, MOD17, FLUXCOM, VODCA2, VPM, and an improved EC-LUE model, to systematically evaluate the spatiotemporal variations of GPP across Hainan Island from 2001 to 2015. The study also investigates the underlying driving mechanisms of these variations. A unified spatiotemporal preprocessing approach was applied to harmonize the data from different remote sensing products, followed by multidimensional analytical methods to analyze the variations in GPP across the island. The results reveal a distinct “core-periphery” spatial pattern of GPP across Hainan Island, with the central tropical rainforest region exhibiting the highest GPP values. Specifically, the annual average GPP of the central tropical rainforest reached 2829.78 gC m⁻² yr⁻¹, which was 32.6% higher than the island-wide average GPP range (1659~2138 gC m⁻² yr⁻¹). The central area also exhibited the highest annual GPP growth rate of 45 gC m⁻² yr⁻², which was statistically significant (p < 0.05). These findings highlight the central region’s potential as a carbon sink within the tropical island ecosystem. Additionally, the study found a pronounced spatial differentiation in GPP between the northern and southern regions of the island, with 19˚N latitude serving as a boundary. This division was statistically significant, with an R² value of 0.73. The northern region showed greater interannual variability in GPP, with standard deviations greater than 300 gC m⁻² yr⁻¹, indicating more fluctuating carbon dynamics. In contrast, the southern region demonstrated a more stable GPP pattern with a lower standard deviation of less than 120 gC m⁻² yr⁻¹. This finding suggests that the southern part of Hainan Island may have more consistent carbon flux, possibly due to its tropical rainforest characteristics. In terms of urbanization impact, the northeastern region, which has undergone substantial urban development, experienced a significant negative GPP growth rate ranging from −30 to −45 gC m⁻² yr⁻². This indicates that urbanization and land-use changes in this area have contributed to a decrease in the island’s overall carbon productivity. The loss of vegetation cover and the associated decline in GPP underscore the need for sustainable urban planning and land management strategies to mitigate further carbon losses in this region. The study also highlighted systematic differences in the GPP estimates derived from different remote sensing products, with the variation spanning a range of 479 gC m⁻² yr⁻¹. The Light Use Efficiency (LUE) model was found to overestimate GPP by approximately 15% during the rainy season, while machine learning models tended to underestimate GPP by 9.8% during the dry season. These discrepancies were mainly attributed to the tropical cloud cover, which has an average coverage rate of 75.44%. The study suggests that integrating microwave and optical data can reduce estimation errors by 12~15%, offering a more accurate method for monitoring GPP in tropical island ecosystems.