| 基于可解释性机器学习下哈尔滨市景观生态风险时空演变及驱动力 |
| 摘要点击 268 全文点击 6 投稿时间：2025-09-04 修订日期：2025-10-25 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 XGBoost-SHAP 景观生态风险 时空演变 轨迹分析法 驱动因素 |
| 英文关键词 XGBoost-SHAP landscape ecological risk spatiotemporal evolution trajectory analysis method driving factors |
| DOI 10.13227/j.hjkx.202509060 |
|
| 中文摘要 |
| 深入研究哈尔滨市景观生态风险对东北老工业基地区域生态保护及可持续发展具有重要意义. 基于2003~2023年土地覆盖数据，运用景观生态风险评价模型、轨迹分析法与XGBoost-SHAP可解释性机器学习，分析了哈尔滨市景观类型的时空变化、生态风险分布与动态特征及驱动机制. 研究发现，哈尔滨以耕地、林地为主，建设用地增加859.14 km2成为推动景观格局变化的主要因素. 全市景观生态风险整体偏低，低和极低风险区面积占比大于96%，而高和极高风险区仅占0.79%且主要集中于道里、南岗等城市核心区. 风险转移分析显示，风险保持不变的区域始终占主导，而风险升高与降低的区域则表现出动态变化. 轨迹分析揭示风险动态的多样性，稳态维持型占61.8%为主要模式，升高性波动和降低性波动分别占22.9%和14.6%，波动回归型与跨越性波动仅占0.6%和0.1%. 驱动因子分析显示高程（SHAP值0.014 3）和NDVI（SHAP值0.004 19）是影响景观生态风险格局分布的主要自然因素，人口密度（SHAP值0.009 05）和GDP（SHAP值0.003 4）在道里区、南岗区和香坊区等城市核心区呈负向驱动效应，体现了基础设施建设对生态风险的缓解. 研究结果可为哈尔滨市生态保护与国土空间规划提供决策参考. |
| 英文摘要 |
| Investigating landscape ecological risk in Harbin is of great significance for ecological protection and sustainable development in the northeastern old industrial base. Based on land cover data from 2003 to 2023， this research employed landscape ecological risk assessment models， trajectory analysis， and XGBoost-SHAP interpretable machine learning to analyze the spatial and temporal variation of landscape types， risk distribution and dynamics， and driving mechanism. The results showed that cropland and forest land were predominant， with an increase of 859.14 km2 in built-up areas being the primary factor of landscape pattern changes. The overall ecological risk level in Harbin remained low， with more than 96% of the area classified as low or very low risk； high and very high-risk zones accounted for only 0.79%， mainly concentrated in core urban districts such as Daoli and Nangang. Risk transfer analysis indicated that regions with stable risk levels dominated， whereas areas experiencing risk increase or decrease demonstrated dynamic shifts. Trajectory analysis revealed diverse risk development modes， with the stable maintenance type accounting for 61.8%， while risk-increasing fluctuation and risk-decreasing fluctuation accounted for 22.9% and 14.6%， respectively. Fluctuation-reversion and cross-over fluctuation accounted for only 0.6% and 0.1%. The analysis of driving factors showed that elevation （SHAP value 0.014 3） and NDVI （SHAP value 0.004 19） were the primary natural factors influencing the distribution of landscape ecological risk patterns， whereas population density （SHAP value 0.009 05） and GDP （SHAP value 0.003 4） exhibited negative effects in core urban areas such as Daoli， Nangang， and Xiangfang， reflecting infrastructure development’s role in alleviating ecological risks. These findings provide scientific references for ecological conservation and territorial spatial planning in Harbin. |
