ABSTRACT

Model-derived relationships between chlorophyll a (Chl-a) and nutrients and temperature have fundamental implications for understanding complex interactions among water quality measures used for lake classification, yet accuracy comparisons of different approaches are scarce. Here, we (1) compared Chl-a model performances across linear and nonlinear statistical approaches; (2) evaluated single and combined effects of nutrients, depth, and temperature as lake surface water temperature (LSWT) or altitude on Chl-a; and (3) investigated the reliability of the best water quality model across 13 lakes from perialpine and central Balkan mountain regions. Chl-a was modelled using in situ water quality data from 157 European lakes; elevation data and LSWT in situ data were complemented by remote sensing measurements. Nonlinear approaches performed better, implying complex relationships between Chl-a and the explanatory variables. Boosted regression trees, as the best performing approach, accommodated interactions among predictor variables. Chl-a–nutrient relationships were characterized by sigmoidal curves, with total phosphorus having the largest explanatory power for our study region. In comparison with LSWT, utilization of altitude, the often-used temperature surrogate, led to different influence directions but similar predictive performances. These results support utilizing altitude in models for Chl-a predictions. Compared to Chl-a observations, Chl-a predictions of the best performing approach for mountain lakes (oligotrophic–eutrophic) led to minor differences in trophic state categorizations. Our findings suggest that both models with LSWT and altitude are appropriate for water quality predictions of lakes in mountain regions and emphasize the importance of incorporating interactions among variables when facing lake management challenges.

摘要

叶绿素a（Chl- a）与养分和温度之间基于模型的关系对于理解用于湖泊分类的水质测度之间的复杂相互作用具有根本意义，但很少有不同方法的准确性比较。在这里，我们（1）比较了Chl -a模型在线性和非线性统计方法中的性能；（2）评价了营养物，深度和温度对湖泊Cha -a的单一和综合影响，如湖面水温（LSWT）或海拔。（3）研究了来自Perialpine和巴尔干中部山区的13个湖泊的最佳水质模型的可靠性。叶绿素-a使用来自157个欧洲湖泊的原水水质数据进行建模；高程数据和LSWT原位数据得到了遥感测量的补充。非线性方法的效果更好，这暗示了Ch1- a与解释变量之间的复杂关系。作为性能最佳的方法，增强型回归树可适应预测变量之间的相互作用。Chl -a与养分的关系用S形曲线表示，总磷对我们的研究区域具有最大的解释力。与LSWT相比，海拔高度（常用的温度替代物）的利用导致了不同的影响方向，但具有相似的预测性能。这些结果支持在Chl -a模型中利用海拔预测。与Chl -a观测结果相比，Chl -a预测最佳湖泊湖泊方法（贫营养－富营养化）导致营养状态分类的细微差异。我们的发现表明，具有LSWT和海拔高度的两种模型都适合预测山区湖泊的水质，并强调了在面对湖泊管理挑战时纳入变量之间相互作用的重要性。