BRT carbonate sand model Rogaland, 2015

Region: the county Rogaland Number of field observations: 633 Field sampling year: 1992, 1993, 2012 Prevalence: 43 and 22 for 50% and 85% calcium carbonate content respectively Presence / absences: 275/358 and 142/491 for 50% and 85% calcium carbonate content respectively Method: BRT run with the R-package Dismo. The grid is made up by a combination of three BRT models; two simplified models including 12 and 13 predictors for carbonate sand with 50% calcium content) and one BRT model based on 10 predictor variables for carbonate sand with 85% calcium content; all run with tree complexity equal to 5. The grid constitutes the maximum predicted probability values among these three models, which all are “smoothed” by applying a neighborhood analysis (i.e. mean of 3x3 neighbor cells) in advance of combining the models). Number of predictor variables: 10-13 Information about the predictor variables: depth, curvature at detailed, medium and coarse resolution (i.e. applying a 100, 500 and 1500 m moving calculating window respectively, based on the 25 m resolution DEM), wave exposure, slope of wave exposure, and optimal radiation index, all with 25 m resolution; minimum and average seafloor temperature, average seafloor salinity, minimum and average seafloor current speed, and slope of average seafloor current speed, the latter predictors from a hydrodynamic model with 800 m spatial resolution. AUC internal: 0.98研究区域：罗加兰郡（Rogaland） 野外观测样本总量：633份 野外采样年份：1992年、1993年及2012年 出现率：当碳酸钙含量分别为50%和85%时，对应目标物种的出现率为43%和22% 存在/缺失样本量分布：当碳酸钙含量为50%和85%时，存在与缺失样本量分别为275/358和142/491 建模方法：采用R语言Dismo包运行提升回归树（Boosted Regression Trees, BRT）。研究预测格网由3个BRT模型组合构建：其中2个简化模型针对碳酸钙含量为50%的钙质砂，分别纳入12和13个预测变量，树复杂度均设置为5；另1个模型针对碳酸钙含量为85%的钙质砂，纳入10个预测变量，树复杂度同样为5。在合并三个模型的预测结果前，先通过邻域分析（即对3×3邻域单元格取均值）对所有模型的预测概率进行平滑处理，最终格网取值为三个模型中预测概率的最大值 预测变量数量：10至13个 预测变量信息：具体包含深度、不同分辨率下的曲率（细、中、粗分辨率，分别基于25m分辨率数字高程模型（Digital Elevation Model, DEM），采用100m、500m、1500m移动计算窗口）、波浪暴露度、波浪暴露度坡度、最优辐射指数，以上变量空间分辨率均为25m；还包含海底最低温度、海底平均温度、海底平均盐度、海底最低流速、海底平均流速以及海底平均流速坡度，后一类预测变量来自空间分辨率为800m的水动力模型 内部验证受试者工作特征曲线下面积（Area Under Curve, AUC）：0.98