Data from: Spatial scaling of environmental variables improves species-habitat models of fishes in a small, sand-bed lowland river

Habitat suitability and the distinct mobility of species depict fundamental keys for explaining and understanding the distribution of river fishes. In recent years, comprehensive data on river hydromorphology has been mapped at spatial scales down to 100 m, potentially serving high resolution species-habitat models, e.g., for fish. However, the relative importance of specific hydromorphological and in-stream habitat variables and their spatial scales of influence is poorly understood. Applying boosted regression trees, we developed species-habitat models for 13 fish species in a sand-bed lowland river based on river morphological and in-stream habitat data. First, we calculated mean values for the predictor variables in five distance classes (from the sampling site up to 4000 m up- and downstream) to identify the spatial scale that best predicts the presence of fish species. Second, we compared the suitability of measured variables and assessment scores related to natural reference conditions. Third, we identified variables which best explained the presence of fish species. The mean model quality (AUC = 0.78, area under the receiver operating characteristic curve) significantly increased when information on the habitat conditions up- and downstream of a sampling site (maximum AUC at 2500 m distance class, +0.049) and topological variables (e.g., stream order) were included (AUC = +0.014). Both measured and assessed variables were similarly well suited to predict species’ presence. Stream order variables and measured cross section features (e.g., width, depth, velocity) were best-suited predictors. In addition, measured channel-bed characteristics (e.g., substrate types) and assessed longitudinal channel features (e.g., naturalness of river planform) were also good predictors. These findings demonstrate (i) the applicability of high resolution river morphological and instream-habitat data (measured and assessed variables) to predict fish presence, (ii) the importance of considering habitat at spatial scales larger than the sampling site, and (iii) that the importance of (river morphological) habitat characteristics differs depending on the spatial scale.

栖息地适宜性与物种独特的迁移能力，是解释和阐明河流鱼类分布格局的核心关键。近年来，河流水文形态学（river hydromorphology）的综合数据已可在低至100米的空间尺度下完成测绘，有望为构建高分辨率的物种-栖息地模型（species-habitat models，如鱼类相关模型）提供支撑。然而，学界对特定水文形态学与河道内栖息地（in-stream habitat）变量的相对重要性，及其影响的空间尺度仍知之甚少。本研究基于河流形态学与河道内栖息地数据，运用提升回归树（boosted regression trees）算法，为某沙质底质低地河流中的13种鱼类构建了物种-栖息地模型。首先，我们针对5个距离梯度（采样点上下游最大4000米范围）计算预测变量的均值，以筛选可最优预测鱼类物种出现的空间尺度；其次，我们对比了实测变量与基于自然参照条件的评估得分在模型中的适配性；最后，我们筛选出可最优解释鱼类物种出现的变量。模型整体平均质量（受试者工作特征曲线下面积（Area Under the Receiver Operating Characteristic Curve, AUC）=0.78）在纳入采样点上下游栖息地条件信息（2500米距离梯度下AUC最高，提升0.049）与拓扑变量（如河流级别）后显著提升（AUC额外提升0.014）。实测变量与评估变量在预测物种出现方面的适配性相当。河流级别变量与实测断面特征（如宽度、水深、流速）为最优预测因子。此外，实测河床特征（如底质类型）与评估得到的纵向河道特征（如河流平面形态自然度）同样为优质预测因子。本研究结果证实：其一，高分辨率河流形态学与河道内栖息地数据（含实测变量与评估变量）可有效用于鱼类出现情况的预测；其二，需考虑采样点以外更大空间尺度的栖息地特征，这一点至关重要；其三，（河流形态学）栖息地特征的重要性随空间尺度的不同而存在差异。