Abstract The bio-chemical and physical characteristics of a river are directly affected by water temperature, which therefore affects the overall health of aquatic ecosystems. In this study, long term variations of river water temperatures (RWT) in Kupa River watershed, Croatia were investigated. It is shown that the RWT in the studied river stations increased about 0.0232–0.0796ºC per year, which are comparable with long term observations reported for rivers in other regions, indicating an apparent warming trend. RWT rises during the past 20 years have not been constant for different periods of the year, and the contrasts between stations regarding RWT increases vary seasonally. Additionally, multilayer perceptron neural network models (MLPNN) and adaptive neuro-fuzzy inference systems (ANFIS) models were implemented to simulate daily RWT, using air temperature (Ta), flow discharge (Q) and the day of year (DOY) as predictors. Results showed that compared to the individual variable alone with Ta as input, combining Ta and Q in the MLPNN and ANFIS models explained temporal variations of daily RWT more accurately. The best accuracy was achieved when the three inputs (Ta, Q and the DOY) were included as predictors. Modeling results indicate that the developed models can well reproduce the seasonal dynamics of RWT in each river, and the models may be used for future projections of RWT by coupling with regional climate models.

中文翻译：

---

河流温度的长期变化以及气温和河流流量的影响：克罗地亚库帕河流域的案例研究

摘要 河流的生化和物理特性直接受水温影响，进而影响水生生态系统的整体健康。在这项研究中，调查了克罗地亚库帕河流域河流水温 (RWT) 的长期变化。结果表明，所研究河流站点的 RWT 每年增加约 0.0232-0.0796ºC，这与其他地区河流的长期观测报告相当，表明有明显的变暖趋势。过去 20 年的 RWT 上升在一年中的不同时期并不是恒定不变的，站点之间 RWT 上升的差异随季节而变化。此外，还实施了多层感知器神经网络模型 (MLPNN) 和自适应神经模糊推理系统 (ANFIS) 模型来模拟日常 RWT，使用气温 (Ta)、流量 (Q) 和一年中的某一天 (DOY) 作为预测变量。结果表明，与单独以 Ta 作为输入的个体变量相比，在 MLPNN 和 ANFIS 模型中结合 Ta 和 Q 更准确地解释了每日 RWT 的时间变化。当三个输入（Ta、Q 和 DOY）被包括作为预测变量时，实现了最佳准确度。建模结果表明，所建立的模型可以很好地再现各河流RWT的季节性动态，该模型可与区域气候模型耦合用于RWT的未来预测。在 MLPNN 和 ANFIS 模型中结合 Ta 和 Q 更准确地解释了每日 RWT 的时间变化。当三个输入（Ta、Q 和 DOY）被包括作为预测变量时，实现了最佳准确度。建模结果表明，所建立的模型可以很好地再现各河流RWT的季节性动态，该模型可与区域气候模型耦合用于RWT的未来预测。在 MLPNN 和 ANFIS 模型中结合 Ta 和 Q 更准确地解释了每日 RWT 的时间变化。当三个输入（Ta、Q 和 DOY）被包括作为预测变量时，实现了最佳准确度。建模结果表明，所建立的模型可以很好地再现各河流RWT的季节性动态，该模型可与区域气候模型耦合用于RWT的未来预测。