| 摘要: |
| 【目的】探索高分一号卫星影像在新疆落叶松林林分郁闭度估测中的应用潜力，为高分一号卫星影像用于林分郁闭度定量估测提供技术方法。【方法】以新疆布尔津林场为研究区，以阿尔泰山西段新疆落叶松林为试验对象，基于高分一号PMS多光谱影像和DEM数据，利用遥感和GIS技术，采用多元逐步回归和BP神经网络2种方法对新疆落叶松林分郁闭度进行估测。【结果】从模型验证结果可以看出，BP神经网络模型（决定系数R2=0.713，均方根误差RMSE=0.082，相对均方根误差rRMSE=0.175，估测精度EA=82.401%）对新疆落叶松林林分郁闭度的估测要明显优于多元逐步回归模型（R2=0.692，RMSE=0.085，rRMSE=0.182，EA=81.680%），且BP神经网络模型建模时R2=0.714，与其精度验证时的R2=0.713非常接近，说明模型的稳定性良好。【结论】2种模型的估测精度均高于80%，这说明高分一号PMS数据在新疆落叶松林林分郁闭度估测方面具有一定的潜力。 |
| 关键词: 遥感 林分郁闭度 多元逐步回归 BP神经网络 新疆落叶松 |
| DOI： |
| 分类号: |
| 基金项目:安徽省属公办普通本科高校领军人才团队项目；安徽省特支计划项目；安徽省高校学科优秀拔尖人才学术培育项目(gxbjZD44)；安徽省高校协同创新项目 |
|
| Quantitative estimation of stand closure density of Larix sibirica by remote sensing based on GF-1 PMS |
|
LIU Saisai,CHEN Donghua,LI Xusheng,et al
|
| Abstract: |
| 【Objective】This paper explored the potential use of GF-1 satellite imagery to estimate stand closure density of Larix sibirica forest and provided technical methods for achieving such estimations.【Method】The Burqin Forest Farm in Xinjiang was selected as research area and the Larix sibirica forest in the west Altay Mountains was selected for analysis.Based on GF-1 PMS multi-spectral imagery and digital elevation model (DEM),remote sensing (RS),geographic information science (GIS),multivariate stepwise regression and back propagation (BP) neural network were used to estimate stand canopy density of the Larix sibirica forest.【Result】The BP neural network model (determination coefficient (R2)=0.713,root mean square error (RMSE)=0.082,relative RMSE (rRMSE)=0.175,estimation accuracy (EA=82.401%) was superior to the multivariate stepwise regression model (R2=0.692,RMSE=0.085,rRMSE=0.182,EA=81.680%) for estimating closure density.The accuracy of the BP neural network modeling (R22=0.714) was very similar to that of validating (R22=0.713),indicating good stability.【Conclusion】The estimation accuracy of both models was higher than 80%,demonstrating that the GF-1 PMS data were useful for evaluating stand closure density of Larix sibirica. |
| Key words: remote sensing stand closure density multiple stepwise regression back propagation neural network Larix sibirica Ledeb |
