欢迎访问林业科学,今天是

林业科学 ›› 2019, Vol. 55 ›› Issue (6): 184-194.doi: 10.11707/j.1001-7488.20190622

• 研究简报 • 上一篇    

基于地理加权回归模型与林火遥感数据估算森林年龄

杜一尘, 李明泽, 范文义, 王斌   

  1. 东北林业大学林学院 哈尔滨 150040
  • 收稿日期:2018-08-10 修回日期:2019-01-17 出版日期:2019-06-25 发布日期:2019-07-11
  • 基金资助:
    国家自然科学基金项目(31470640)。

Estimation of Forest Stand Age Based on GWR Model and Forest Fire Remote Sensing Data

Du Yichen, Li Mingze, Fan Wenyi, Wang Bin   

  1. College of Forestry, Northeast Forestry University Harbin 150040
  • Received:2018-08-10 Revised:2019-01-17 Online:2019-06-25 Published:2019-07-11

摘要: [目的]通过地理加权回归(GWR)模型估算非干扰林龄,利用遥感数据和林火发生历史数据,获取过火区域信息,进而对林火烈度分级,讨论林火烈度与森林类型的交互作用,估算干扰林龄,最终获得黑龙江省森林年龄的空间分布。[方法]以黑龙江森林为研究区域,基于研究区域的多光谱数据结合地面森林资源清查数据,通过逐步回归方法提取了包括遥感因子绿度指数(Greeness)、湿度指数(Wetness)、林分平均胸径(ADBH)、林分平均树高(ASH)及海拔(Altitude)在内的5个显著因子作为自变量,采用GWR模型建立非干扰林龄估算模型。采用全局Moran I来描述模型残差的空间自相关性。绘制研究区非干扰林龄空间分布图并探究林龄的空间分布状态。结合林火位置与面积记录对多光谱数据目视解译提取过火区域,根据dNBR将过火区域火烈度分级。将火烈度图与植被类型图叠加分析,讨论不同森林类型在不同火烈度下的演替情况。定义干扰林龄时,未发生树种更替的森林林龄不变,树种发生更替的森林在林火发生年将其林龄归为0,并在新的优势树种萌发时从1开始累加,以此类推干扰后森林的林龄。[结果]黑龙江省非干扰森林平均林龄为48年,标准差为16年。GWR模型的Radj2为0.68,RMSE为16.171 7。使用Moran I来检验模型的残差,发现GWR模型可很好地消除残差的空间自相关性。研究区林龄整体空间分布状态不均匀,大兴安岭地区林龄普遍高于黑龙江林区。黑龙江省2000-2010年林火主要发生在大兴安岭及小兴安岭地区,根据dNBR将已提取的过火区域林火烈度分为:未过火、轻度过火、中度过火和重度过火4类,总过火面积为527 932 hm2,其中重度29 157 hm2、中度180 268 hm2、轻度318 507 hm2。兴安落叶松林和蒙古栎林在整个研究区中过火面积最大,分别占总过火面积的28.63%和47.23%。根据不同森林类型在不同火烈度下的演替情况,估算干扰森林的林龄并绘制干扰林龄空间分布图。[结论]GWR模型能较有效地估算黑龙江省非干扰林龄,成功地降低了残差的空间自相关性。在估算林龄的过程中加入林火干扰因素,以获取更真实的林龄空间分布数据,可为黑龙江地区森林NPP、NEP以及森林碳储量、森林生物量等相关研究提供数据支持。

关键词: 林龄, 多光谱遥感, GWR模型, 林火烈度, 干扰

Abstract: [Objective] The non-disturbed forest age was estimated by the geographically weighted regression (GWR) model. The information of forest fire severity was obtained by using remote sensing data and history data of forest fire occurrence. Then the forest fire intensity was graded. The interaction between forest fire severity and forest type was discussed, and the age of disturbed forest was estimated. Finally, the spatial distribution of stand age in Heilongjiang province was obtained.[Method] In this study, Heilongjiang forest was taken as the study area, based on the multi-spectral data of the study area and the forest resources inventory data, stepwise regression method was used to extract five significant factors as the independent variables, including the remote sensing factors of Greeness, Wetness, stand average breast diameter (ADBH), stand average tree height (ASH) and Altitude. The GWR model was used to establish the stand age estimation model of non-disturbed forest. The global Moran I index was used to characterize the spatial autocorrelation of the model residuals. The spatial distribution map of non-disturbed forest age in the study area was drawn and the spatial distribution status of stand age was explored. Combining the location and area records of forest fires, visual interpretation of multi-spectral data was used to extract the burned area. Fire severity was divided into four classes according to the dNBR. The ArcGIS software was used to do an overlaying analysis on the fire severity map with vegetation type map. The fire severity map and vegetation type map were superimposed to discuss the succession of different forest types under different fire severities. When the stand age of disturbed forest was defined, the age of forest without tree species replacement remained unchanged. The stand with tree species replacement was classified its age as 0 in the year of forest fire occurrence, and accumulated from 1 at the beginning of germination of new dominant species, so as to deduce the age of forest after disturbance.[Result] The average age of non-disturbed forests in Heilongjiang was 48 years, with a standard deviation of 16 years. The Radj2 of the GWR model was 0.68, and the RMSE was 16.171 7. Moran I was used to test the residual of the model, and it was found that the GWR model was able to eliminate the spatial autocorrelation of residuals well. The overall spatial distribution of forest age in the study area was uneven, and the forest age in Daxing'an Mountains was generally higher than the average level of Heilongjiang forest area. Forest fires occurred mainly in Daxing'an Mountains and Xiaoxing'an Mountains areas in Heilongjiang Province in 2000-2010. According to dNBR, fire severity was divided into four classes:unburned, low, moderate and high. High severity burned area was 29 157 hm2, moderate severity burned area was 180 268 hm2, and low severity burned area was 318 507 hm2. Larix gmelinii forest and Quercus mongolica forest had the largest burned area in the whole study area, accounting for 28.63% and 47.23%, respectively. According to the replacement of different forest types under different fire severities in the burned area, the age of disturbed forest was determined, and the spatial distribution map of disturbed forest age was plotted.[Conclusion] GWR model can effectively estimate the age of non-disturbed forest in Heilongjiang Province, and successfully reduce the spatial autocorrelation of residual. In the process of estimating forest age, forest fire disturbance factors were added to obtain more realistic spatial distribution data of forest age, which provided data support for forest NPP, NEP, forest carbon storage, forest biomass and other related research in Heilongjiang area.

Key words: stand age, multi-spectral remote sensing, geographically weighted regression(GWR) model, fire severity, disturbance

中图分类号: