Application of real-time water temperature prediction system in winter for long-distance water diversion projects

The construction of water diversion projects is an effective way to promote the redistribution of regional water resources and alleviate the water shortage problems in industry, agriculture, and daily life caused by continuous population growth (Langford et al. 2015; Shandany et al. 2018; Long et al. 2020). However, the winter operation and management of water diversion projects located in high-latitude areas still face the threat of ice jams and ice dams. The middle route of the south-to-north water diversion project (MRSNWDP) in China, with a total length of 1,432 km, is a typical water transfer canal system. Due to the large span of the project from south to north latitudes, the water flows from low latitudes to high latitudes, causing significant heat loss along the way. In addition, the air temperature in the canal system north of the Yellow River during winter is low, which makes it easy to produce ice damage problems (Wen et al. 2015). The water flows by gravity, and the main canal has no other online storage reservoirs and limited storage capacity. Also, it adopts the operation mode of the constant water level before the gate. Consequently, it is difficult to dispatch water in winter (Duan et al. 2016). Ice jam is a phenomenon in which a large number of frazil slush accumulates under the ice cover, quickly blocking part of the flow cross-section and causing the upstream water level to rise, mainly occurring in the freeze-up period (Barrette & Lindenschmidt 2023). After the canal system freezes, large-flow water transfer conditions will accelerate the flow velocity of drift ice, forming accumulations at the narrow channels and around hydraulic structures, increasing the risk of ice jam occurrence (Ettema et al. 2009). Water diversion projects often reduce the water flow to suppress the ice jam risk. However, premature and large-scale use of small-flow water transfer methods may cause temporal and spatial losses of water transfer benefits, aggravating the water shortage dilemma in winter. Taking the Beijumahe Controlled Gate at the end of the project as an example, the design flow through the gate is 50 m³/s, while the actual flows in the four winters of 2011–2015 were 11–17 m³/s, accounting for only about 30% of the design flow (Duan et al. 2016). Therefore, it is necessary to finely adjust the time and range of adopting small-flow water transfer based on the water temperature prediction results of the project in winter to improve its safety and efficiency.

At present, the study of water transfer in winter mainly involves ice monitoring, ice prediction modeling methods, and promoting the development of water diversion project information technology. Effective ice monitoring can provide key information for the winter operation of artificial water infrastructure in cold regions (Norvanchig & Randhir 2021; Zhang et al. 2021). Hydrological gauges and pressure recorders are used to monitor the evolution of long-term flow conditions, and empirical models were established to predict ice based on changes in water levels and flows (Beltaos & Burrell 2010; Beltaos et al. 2011). Another research direction is the development of satellite image icing monitoring methods. Liu et al. (2009) reported on the application of high spatiotemporal resolution satellite data in ice observation. Unterschultz et al. (2009) demonstrated the potential of using synthetic aperture radar (SAR) satellite images to identify ice jams during breakup, intact ice, and open water bodies. Chaouch et al. (2014) proposed a method that can distinguish between ice covers and flowing water from remote sensing photography. Both ground sensors and satellite photography have problems with high maintenance costs or insufficient accuracy. Since drones entered the commercial market a few years ago, their prices have become cheaper and have been used in ice monitoring in river and water diversion projects to compensate for these shortcomings (Liu et al. 2019; Chen & Liu 2023).

The hydrological conditions in cold regions, especially in the middle and high latitudes, are sensitive to climate change (Hu et al. 2017). The seasonal accumulation of glaciers and snow become important sources of river flow after they melt (Barnett et al. 2005; Mohammadi et al. 2023). In addition, due to the recent sustained emissions of greenhouse gases, rising air temperatures and changes in precipitation will greatly affect the hydrological conditions in cold regions in the future (Minville et al. 2008; Adam et al. 2009; Huziy et al. 2013; Rasouli et al. 2015). Therefore, hydrological modeling in cold regions has become a research focus for scholars (Cooper et al. 2011; Brown et al. 2014; Gao et al. 2018; Burger et al. 2019; Mohammadi et al. 2023). On the one hand, scholars attempted to reveal the impact of climate change on hydrological conditions and established ice jam occurrence prediction models by collecting hydro-climatic data (Massie et al. 2002; Chen & Ji 2005; Wang et al. 2012; Guo et al. 2018). On the other hand, there are also many studies on the ice jam formation mechanism, numerical simulation, and predicting the probability of ice jam formation. Combined with the prototype observation data of the winter ice regime in the MRSNWDP in the past 5 years, Duan et al. (2016) analyzed the characteristics of the temporal and spatial distribution of the ice conditions. Sui et al. (2002) studied the relationship between water level, ice jam thickness, and ice flow and revealed the empirical relationship between ice thickness and the Froude number. Xu et al. (2010) studied several ice mechanics problems in the water transfer of the MRSNWDP during the ice period, which provided a reference for water transfer scheduling. Guo et al. (2011) conducted a numerical simulation of the whole water transfer process during the ice period in typical years and gave suggestions for safe water transfer scheduling of the MRSNWDP in winter. Shen et al. (1995) proposed a fine river ice model, suitable for river networks, and established a two-dimensional dynamic transport theory, which has been continuously improved to form the DynRICE model, and then validated the model using historical measured data.

The rapid development and widespread application of information technology have provided new management tools for the construction and operation of water diversion projects. The application of new technologies such as GIS (Geographic Information System), BIM (Building Information Modeling), the Internet of Things, digital twin, and big data is conducive to improving the real-time scheduling capability of the project and realizing efficient water transfer, safe water transfer, and intelligent water transfer. BIM + GIS technology has been applied in the operation and management of the Jiaodong Water Diversion Project in Shandong Province and the Dianchi Lake Water Replenishment Project in Yunnan Province, China (Dai et al. 2018; Han et al. 2022). Liu et al. (2021) proposed an integrated visualization framework based on BIM, which integrates data collection, data analysis, early warning release, and emergency response, improving the automation level and emergency management efficiency of water diversion project operation. Meanwhile, the digital twin technology is also an effective tool to ensure the safety and efficiency of water transfer. Liu et al. (2023) proposed a digital twin framework including a real-time self-correcting model that can provide timely fault detection for the daily scheduling of water diversion projects. Ye et al. (2023) proposed a data-driven and physics-based coupled real-time model predictive control (MPC) method for real-time regulation of hydropower plant operation, which has a higher accuracy, faster decision-making efficiency, and real-time feedback correction mechanism compared to traditional models.

In summary, some progress has been made in ice monitoring, ice prediction, and promoting the development of project informatization in water diversion projects. However, a complete and real-time ice prediction method has not yet been formed, and ice prediction and project scheduling have not been effectively combined. Moreover, scholars often establish statistical or intelligent algorithmic models to predict water temperature or the probability of ice jam formation by inputting information such as air temperature and precipitation. Water temperature measurement is an important part of on-site observation and significantly impacts the analysis of the ice evolution in the project (Cook et al. 2006; Turcotte et al. 2012, 2014; Maheu et al. 2016). Researchers use many high-precision temperature sensors to collect water temperature data to study the water temperature change process and then analyze its influencing factors, which are the basis for analyzing ice evolution. This study establishes a water temperature prediction model for long-distance water diversion projects in winter based on the heat exchange principle and river ice hydraulics theory and tests the model's effects on short-term and medium- to long-term water temperature prediction by actual cases. Compared with other studies, the model proposed in this paper is more suitable for water diversion projects, and its accuracy is tested in practical applications (Shen et al. 1993, 1995).

In the model, the unsteady flow equations of an open canal are used to simulate the hydraulic response characteristics of the canal system under the conditions of floating ice cover formation and open canal flow. The governing equations are as follows (Yao et al. 2009; Liu et al. 2011):

When a floating ice cover is formed in the canal, the wetted perimeter and roughness coefficient of the canal are both affected by the ice cover (Wei & Huang 2002; Yang 2018).

The model is set as follows:

• (1)

  An initial ice cover will be formed in the canal cross-section when the proportion of ice covering the water surface exceeds 80% (C_a ≥ 80%) (Liu 2019).

• (2)

  If the Froude number is less than 0.06 and the flow velocity is less than 0.5 m/s, ice floes will form a smooth cover of a single layer of ice floes through juxtaposition. The modeling of the freezing process is described in the literature (Liu et al. 2013).

• (3)

  If the Froude number is greater than 0.06 or the flow velocity is greater than 0.5 m/s, surface ice elements can under turn or submerge at the leading edge to form a thicker ice cover (Ma et al. 2009).

• (4)

  Ice formed in a pool cannot flow through the downstream gate to the downstream pool.

The air temperature and hydraulic data used in this paper are provided by the project management department of the MRSNWDP. The measured water temperature data are provided by the automated water temperature monitoring system along the project. In addition, the Changjiang River Scientific Research Institute is responsible for on-site monitoring to improve the accuracy of the input data through manual field testing and review at some nodes, including Gate 14 (Hutuohe Controlled Gate), Gate 21 (Gangtou Controlled Gate), Gate 24 (Beiyishui Controlled Gate), and Gate 26 (Beijumahe Controlled Gate). This study focuses on numerical simulation and modeling to compare the predictive and measured values of water temperature to test and optimize the model's accuracy. To test the water temperature prediction effects of the model, the water temperature of the canal system was predicted for three periods: 1-day, short-term, and medium- to long-term, and the simulation time is from 1 January to 24 February 2022.

We calculate the root mean squared error (RMSE) of water temperature at four observation stations from 6 to 15 January, as shown in Table 2. Spatially, the RMSE of water temperature at the four observation stations from 6 to 15 January is relatively small, at no more than 0.2. Temporally, except for the RMSE of the water temperature at the four stations on 12 January reaching 0.25, the RMSE at other times does not exceed 0.2. The errors at the Beijumahe Controlled Gate are smaller, which are within the range of ±0.22 °C, and the RMSE is about 0.11.

From the prediction results, the overall water temperature of the canal system in the winter of 2021–2022 is higher than 0 °C, and no icing occurred, which is consistent with the actual situation. Compared with models that use intelligent algorithms, such as adaptive fuzzy inference systems and BP neural networks, this model has physical mechanisms (Wang et al. 2012; Song et al. 2020; Chaplot 2021). However, the numerical solution of the model has accumulated errors in long-distance and long-time simulations. Moreover, the input meteorological conditions are the daily average air temperature data of meteorological stations in prefecture-level cities, and the scarcity of stations leads to poor data representation, affecting the model prediction accuracy. Therefore, the accuracy of the model can be improved by optimizing the model parameters, improving the temperature representativeness of the meteorological stations along the project, shortening the simulation time and range, and improving the monitoring quality of measured data.

This study uses the numerical simulation method to establish a water temperature prediction model, which can predict the future water temperature results by inputting air temperature data and hydraulic data of the project. Then, the icing conditions in the project for a certain period of time in the future can be further predicted. The main conclusions are as follows:

• (1)

  The model predicts the long-distance water temperature for 1 and 7 days with small errors. When the prediction period is the 1-day, the prediction errors of 10 consecutive water temperature predictions from 6 to 15 January at four observation stations can be controlled at ±0.3 °C, with the RMSE ranging from 0.07 to 0.25. When the prediction period is the 7-day, the prediction errors of five consecutive water temperature predictions from 6 to 16 January at four stations can be controlled at ±0.6 °C, with the RMSE ranging from 0.12 to 0.36. For the Beijumahe Controlled Gate at the end of the project, the error ranges are from −0.22 to 0.22 °C and from −0.5 to 0.3 °C, with the RMSE of 0.11 and 0.19, respectively.

• (2)

  The accuracy of 15-day water temperature prediction is greatly affected by the accuracy of the air temperature forecast, and the range of prediction errors is from −2.42 to 0.36 °C. The range of prediction errors for the first 7 days is from −0.59 to 0.36 °C, and the range for the last 8 days is from −2.42 to 0.22 °C. This is because a decrease in the accuracy of air temperature forecasts can lead to a decrease in the accuracy of water temperature predictions. The 21-day water temperature prediction was conducted only once, with the range of prediction errors from −0.35 to 0.67 °C and the range of RMSE from 0.20 to 0.38. It is necessary to organize more water temperature predictions for similar periods in the subsequent studies to test the model's medium- and long-term water temperature prediction accuracy.

• (3)

  This study establishes a real-time water temperature prediction system for water diversion projects, which can use the water temperature prediction model to predict future water temperature changes and then predict the freezing range and freezing time of the project so as to dynamically adjust the water flow by time and space and reduce the waste of water transfer benefits caused by the use of a small flow in the whole project at a fixed time, realizing safe water transfer, efficient water transfer, and intelligent water transfer. However, there is no actual freeze-up that occurred in the verification years, so it is not possible to conduct actual tests on prediction effects of the water temperature under drift ice and freezing conditions, freezing time, and freezing range, and the model parameters also need to be continuously optimized in later applications.

This work was supported by the National Key Research & Development Plan of China (No. 2022YFC3202500), the Open Research Fund of Key Laboratory of River Basin Digital Twinning of Ministry of Water Resources (No. Z0202042022), the National Science Foundation of China (No. 51779196), and the Graduate Innovation and Entrepreneurship Foundation of Wuhan University of Science and Technology (No. JCX2022023).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.