Forest Potential Productivity Mapping by Linking Remote-Sensing-Derived Metrics to Site Variables
Forest Potential Productivity Mapping by Linking Remote-Sensing-Derived Metrics to Site Variables
复制标题
DOI：
10.3390/rs12122056
复制
发表时间：
2020-06-01
期刊：
影响因子：
5
通讯作者：
Lamb, Sean
中科院分区：
文献类型：
--
作者：
Rahimzadeh-Bajgiran, Parinaz;Hennigar, Chris;Lamb, Sean
高分辨率的全区域森林立地生产力地图是有效的大规模林业规划和管理的基本需要。在这项研究中，我们将Sentinel-2卫星数据纳入基于气候，岩性，土壤和地形指标的森林生产力（生物量增长指数（BGI））的增量测量中，以绘制北美部分地区的改进BGI（iBGI）。最初，几个哨兵-2变量，包括9个单一的光谱波段和12个光谱植被指数（SVI）与森林管理变量相结合，使用随机森林预测树木体积/公顷和高度。结果显示，当Sentinel-2变量与BGI一起纳入体积和身高预测时，袋外（OOB）r（2）增加了10- 12%。后来，选定的哨兵-2变量用于生物量增长预测缅因州，美国和新玩法，加拿大使用的数据从7738省级永久样地。Sentinel-2红边位置（S2 REP）指数被确定为最重要的变量比其他已知的影响网站的生产力。虽然与基础BGI模型相比，iBGI准确性略有改善（类似于2%），但其他变量的系数发生了显著变化，并且当纳入S2 REP时，一些研究中心变量变得不那么重要。
A fine-resolution region-wide map of forest site productivity is an essential need for effective large-scale forestry planning and management. In this study, we incorporated Sentinel-2 satellite data into an increment-based measure of forest productivity (biomass growth index (BGI)) derived from climate, lithology, soils, and topographic metrics to map improved BGI (iBGI) in parts of North American Acadian regions. Initially, several Sentinel-2 variables including nine single spectral bands and 12 spectral vegetation indices (SVIs) were used in combination with forest management variables to predict tree volume/ha and height using Random Forest. The results showed a 10-12 % increase in out of bag (OOB) r(2)when Sentinel-2 variables were included in the prediction of both volume and height together with BGI. Later, selected Sentinel-2 variables were used for biomass growth prediction in Maine, USA and New Brunswick, Canada using data from 7738 provincial permanent sample plots. The Sentinel-2 red-edge position (S2REP) index was identified as the most important variable over others to have known influence on site productivity. While a slight improvement in the iBGI accuracy occurred compared to the base BGI model (similar to 2%), substantial changes to coefficients of other variables were evident and some site variables became less important when S2REP was included.