BRT kelp forest model Nordland, 2014

Region: Nordland county Number of field observations: 4331 Field sampling years: 2011, 2012 Prevalence: 13% Presence / absences: 503 / 3828 Method: BRT run in R with the R-package Dismo. (GAMs were also tested and gave similar distribution.) Number of predictor variables: 9 Information about the predictor variables: curvature at coarse and medium detailed spatial resolution with a 1025 and 525 m moving calculating window respectively, based on a 25 m resolution DEM; DEM, slope, and wave exposure (all at 25 m resolution, the predictor variables are described in Rinde et al. 2006); mean salinity, mean current speed, maximum temperature and mean temperature, all with a 800 m resolution, but resampled to 25 m. AUC internal: 0.99.研究区域：诺尔兰郡（Nordland county） 野外观测样本总数：4331 野外采样年份：2011年、2012年 出现率：13% 存在样本/缺失样本：503 / 3828 建模方法：使用R语言及其中的Dismo扩展包实现提升回归树（Boosted Regression Trees，BRT）建模；此外还测试了广义可加模型（Generalized Additive Models，GAMs），其预测分布结果与BRT相近 预测变量数量：9个 预测变量信息如下：基于25米分辨率的数字高程模型（Digital Elevation Model，DEM），分别以1025米、525米的移动计算窗口生成粗、中两种空间分辨率的曲率因子；包含25米分辨率的DEM、坡度与波浪暴露度因子（相关预测变量的详细说明可参见Rinde等人2006年的研究）；另有平均盐度、平均流速、最高温度与平均温度变量，初始分辨率为800米，后经重采样统一至25米分辨率 内部验证曲线下面积（Area Under the Receiver Operating Characteristic Curve，AUC）值：0.99