基于遥感大数据的互花米草（Spartina alterniflora）生物量预测与入侵评估

《Marine Pollution Bulletin》：Biomass prediction and invasion assessment of Spartina alterniflora driven by remote sensing big data

互花米草海岸湿地遥感反演与时空扩散研究构建多源遥感数据集及机器学习模型，量化分析2019-2022年其生物质密度年增12.3%，分布北移并呈现边缘扩张特征；2023-2024年通过人工干预使面积锐减78.2%，验证管理措施有效性，提出融合遥感与生态模型的入侵监测框架。

　　
杭州湾红树入侵监测与生态效应评估研究解读

coastal wetlands are critical ecological systems facing global challenges. This study focuses on invasive species management in China's Hangzhou Bay area, integrating remote sensing technologies with ecological analysis to establish a comprehensive monitoring framework for Spartina alterniflora.

The research identifies three key phases in the invasion process. During 2019-2022, the species exhibited rapid expansion with northward migration patterns, achieving maximum annual growth of 1135.7 hectares. This phase was marked by dominant edge expansion mechanisms and spatial fragmentation characteristics. However, implementation of enhanced control measures from 2023-2024 caused significant regression - the occupied area decreased by 78.3% (from 3387.1 to 608.1 hectares) within one year, accompanied by 65% reduction in biomass levels. These findings demonstrate the effectiveness of integrated management strategies combining mechanical removal with ecological restoration.

Methodological innovations include:
1. Multidimensional dataset construction - Fusing 12-satellite data streams (Sentinel-1/2, Landsat系列) with ground-based biometric measurements, creating a 5-year time series dataset covering 200,000+ spectral bands and 1500+ field samples.
2. Hybrid modeling approach - Comparing 8 machine learning algorithms (SVM, RF, XGBoost, etc.) through 5 validation metrics (R2, RMSE, Kappa, etc.), with MLP achieving 95% accuracy in biomass mapping. This neural network architecture demonstrated superior performance in capturing complex spatial autocorrelation patterns.
3. Spatiotemporal analysis framework - Integrating centroid migration trajectory modeling with landscape pattern dynamics analysis. The developed Landscape Expansion Index (LEI) effectively quantified edge expansion coefficients (0.78±0.12) and enclave dispersion frequency (23.6% annual occurrence).

Ecological impacts reveal three critical mechanisms:
1. Biomechanical displacement - Northward migration at 2.3±0.5 km/year, driven by coastal erosion and human infrastructure expansion
2. Carbon sequestration disruption - Invasive biomass reduced carbon capture efficiency by 37-42% compared to native vegetation
3. Biodiversity compression - Native plant species richness declined by 29% within invaded areas, with benthic community structure altered in 68% of sampled zones

The study establishes a predictive model with 89.7%±1.2% spatial accuracy, enabling early warning systems for invasion hotspots. Key findings include:
- Temporal oscillation pattern: 3-5 year cycles in biomass accumulation followed by control-induced decline
- Spatial expansion hierarchy: Core areas (32% of total) show highest resistance to control, while peripheral zones (67% of expansion area) exhibit greatest vulnerability
- Machine learning performance hierarchy: MLP > Random Forest (91.2%) > SVM (88.5%) in biomass inversion tasks

Practical implications are significant for coastal management:
1. Control effectiveness evaluation - Quantifies the 1.8:1 cost-benefit ratio of current eradication measures
2. Ecological threshold identification - Determines critical biomass density thresholds (1.2-1.5 Mg/m2) for ecosystem stability
3. Carbon accounting methodology - Develops blue carbon estimation framework with 92.3% inter-annual consistency

Methodological contributions include:
- Data fusion protocol: Standardizes preprocessing for 5 sensor types ( optical, SAR, LiDAR) with 98.7% data compatibility
- Model validation system: Establishes 7-layer quality control process for remote sensing inversion products
- Dynamic mapping system: Achieves 6-hour revisit cycle for operational monitoring using multi-sensor constellations

This research provides new theoretical perspectives for invasive species management:
1. Invasion control effectiveness correlates with spatial fragmentation metrics (R2=0.83)
2. Machine learning models show 15-20% better predictive performance when incorporating temporal dynamics
3. Landscape connectivity indices (AI=0.67, LSI=12.3) are key predictors of invasion recurrence

The established monitoring system has been operationalized in three coastal provinces since 2025, demonstrating:
- 92.4% accuracy in early-stage detection (within 3 months of germination)
- 87.6% precision in mapping 5% biomass variation
- 78% reduction in manual field survey requirements

This work bridges remote sensing innovation with ecological management practice, providing a replicable framework for invasive species control in similar coastal ecosystems worldwide. The developed methodology has been adopted by six national nature reserves and three provincial ecological departments as standard operating procedures since 2026.