A WebGIS-based flood control management system for small reservoirs: a case study in the lower reaches of the Yangtze River

To modify the uneven distribution of water resources in both time and space, more than 847,000 reservoirs have been constructed globally in the last 100 years, and approximately 95% of these are small reservoirs (Song et al. 2015). They not only provide water, hydroelectric energy and irrigation, but also stabilize extreme inflows to mitigate floods or droughts. Due to the natural uncertainties, public concern about the small reservoirs and the specific efforts for flood control management have increased in recent years (Chang & Chang 2006; Rodrigues et al. 2012).

Reservoir flood control management is a historic issue regarding the use of structural and non-structural measures to optimize flood discharge and reduce flood risks. It requires simultaneous considerations about the hydrologic, geotechnical, environmental and behavioural aspects. Among them, the flood forecast, including volume, peaks and duration, is one of the key factors. Many flood-prone countries have employed forecasting systems since the 1990s. These include the National Weather Service River Forecasting System (NWSRFS) for 13 main rivers across the USA (Burnash & Singh 1995), the Delft Flood Early Warning System (Delft-FEWS) for several European countries (Werner et al. 2009), the integrated flood control management system for 69 large reservoirs of China (Cheng & Chau 2004), etc. These early examples provide flood control decision support by establishing chart-based user interfaces around the hydrological and hydraulic models used. However, their application and usage requires a good understanding of flood processes, which is feasible for national operational centres (Werner et al. 2013; Cools et al. 2016), but difficult for small reservoir managers with uneven levels of hydrological knowledge.

Recent development in information technology has provided substantial opportunities to enhance early warning and flood control at different spatial scales. There are global systems for upcoming floods in large world river basins (Alfieri et al. 2013; Wu et al. 2014), and continental systems that complement the national systems with medium- and long-range forecasts for transboundary rivers and lakes (Roo et al. 2011; Werner et al. 2013; Thiemig et al. 2015). Flood control systems for local areas have widely incorporated web-based technologies. Demir & Krajewski (2013) provided an integrated online platform with flood forecasts and inundation maps. Horita et al. (2015) combined monitoring data with volunteered geographic information for river flood risk management. The above systems combined with desktop- and web-based geographic information system (GIS) have proved to be integrative and cooperative environments for river basins and urban watersheds, so that researchers, decision-makers and the general public are better involved. However, practical flood control management usually requires targeted and deterministic data to help formulate discharging schemes, especially for the numerous and widely distributed small reservoirs. In order to keep the results concise and effective for reservoir managers, the system has to cover multidisciplinary knowledge within, but present as user-friendly interfaces.

Many rainfall–runoff models have been developed to generalize hydrologic processes and offer flood forecasts. Some are lumped conceptual models, such as the Sacramento model, the Xinanjiang model and the tank model (Zhao 1992; Lü et al. 2013). They are applied in many regions but typically have more than ten parameters that require gradual adjustment to the measured stream-flow data (Chang & Chang 2006). In particular, the Soil Conservation Service Curve Number (SCS-CN) model is designed for small watersheds (Mishra & Singh 2003). Due to the simple structure and clearly stated assumptions, relatively small amounts of data are required when using this model. However, the spatial analysis for watershed characteristics, such as elevation, soil type and land use, is essential to obtain more accurate CN values, rather than depending on the tables developed for US conditions.

To represent the spatial conditions of underlying surface and meteorological data, GIS has enabled the models to have physically distributed characteristics. The LISFLOOD model is an example that simulates the spatial and temporal pattern of river discharge in large basins. The TOPKAPI model is applied to medium-sized river basins, and transforms rainfall–runoff routing processes into non-linear reservoir differential equations. The HEC-HMS model supports numerous infiltration loss parameterizations to calculate overland flow runoff (Vieux 2001; Ciarapica & Todini 2002; Van Der Knijff et al. 2010). The selection and application of rainfall–runoff models is one of the most important tasks for reservoir flood control management, since the specific flood forecast relies on not only the practical hydrological measurements but also the available time-range to guide flood discharge.

The Yangtze River has been subjected to flooding throughout history, especially for the provinces and municipalities located in its lower reaches, which are of great economic importance for China (Cheng & Chau 2004). Due to the intensive rainfall events and humid climate during the flood season, inflow floods rapidly raise the water levels and cause dam failure risks (Zhao et al. 2014). Although flood control management is available for large reservoirs, it is inadequate for small reservoirs. Instead of using advanced forecasting and decision-support tools, these reservoirs primarily rely on the long-term field experience of the operation personnel who prefer to maintain a higher capacity due to economic benefits, but this approach introduces a significant amount of uncertainties regarding dam safety. Meanwhile, the hydrological measurements are mostly limited to water level logged by days or hours, and rainfall in or near the catchment, which limits the direct usage of existing rainfall–runoff models.

To improve this situation, a WebGIS-based flood control management system has been developed since 2014, and the Hengshan Reservoir in Jiangsu Province selected as the study area for the typical hydrologic and reservoir project conditions in the lower reaches of the Yangtze River. To date, it has completed a trial operation and will be gradually applied to other small reservoirs in the vicinity.

In order to provide useful reference to the flood control and water resource utilization in humid regions, this study focuses on the design and implementation of the WebGIS-based system. It tries to use multidisciplinary knowledge to establish user-friendly tools to bridge the gap between large-scale flood control systems and practical management of small reservoirs. It also investigates the capability to present concise flood forecasts based on limited hydrological measurements to guide the flood discharge during intensive rainfall in the flood season. Typical rainfall events in the Hengshan Reservoir are chosen to estimate the model parameter values, and approximately 100 rain days are used to analyse the forecast accuracy. After describing a real flood discharging example, future improvements and recommendations for the system are discussed.

There are two classes of basic materials used in this study. One is necessary for flood level forecasts, as shown in Table 1, and the other is optional to enrich the map browse and water level monitoring, which contains an online geographic base map, 30-m digital elevation model (DEM), dam construction designs, photos taken at key locations and video surveillance.

Providing accurate and efficient inflow flood forecasts during intensive rainfall to support flood discharge is the main purpose of the system. Meanwhile, some characteristics of small reservoirs have to be considered, such as limited hydrological measurements, fast rainfall–runoff processes and uncertainties of geospatial data composition. Therefore, the system design and implementation requires a combination of multidisciplinary knowledge from different fields, such as reservoir management, flood discharge, hydrological modelling, map development and software integration.

Agile software development methods are currently adopted in many cross-regional and cross-domain practices to cope with crises, such as rising complexity and extended cycles (Brhel et al. 2015). Compared with traditional planning-based methods, they value a progressive and iterative approach to achieve effective results. Following the principles of agile methods, this study uses scenario description, visual design and rapid prototyping to keep the system available during the entire process of design, implementation and application. Thus, different participants are initiatively involved, and make suggestions to improve the system based on their respective expertise.

The detailed concepts are as follows:

1. The Thiessen polygons of the rainfall gauges are established to obtain total rainfall amounts in the catchment.

2. The double-layered soil moisture model is used to estimate the rainfall–runoff. The upper layer receives rainfall and quickly gathers surface runoff, while the lower layer receives infiltration from the above layer and gradually forms subsurface runoff.

3. The evaporation assessment depends on the rain conditions. On a rainy day, it is ignored due to the humid conditions. On a non-rainy day, it assumes that there is a constant consumption in the upper soil layer.

4. Flood inflow contains the surface and subsurface runoff. The reservoir water consumption, such as drinking, irrigation and leakage, is generalized as flood discharge. Thus, there is a minimum daily discharge amount.

The model contains eight parameters, which are shown in the first three columns of Table 2, and the overall forecasting procedure is presented in Figure 4(b) with detailed descriptions as follows:

1. Retrieve the evaporation constant and daily rainfall amounts during N days before the forecasting day.

2. Estimate the soil moisture of the two layers from N day before to 1 day before.

3. Calculate the accumulated inflow flood volume based on the surface and subsurface runoff.

   

   1

   where IF is the accumulated inflow flood volume (m³) and R_l is the total runoff volume (mm).
4. If the forecasted flood level based on reservoir stage–storage curve is higher than the FLWL during the flood season, flood discharge is required immediately. The total outflow volume can be calculated as follows:

   

   2

   where OF is the total outflow volume (m³), v_i is the flood discharge during a particular period (m³/s) and t_i is the corresponding duration (s).

5. Finally, the discharged water level is obtained by the actual change of storage capacity.

According to the theories of the SCS-CN model, the antecedent soil moisture can be approximated as zero if intensive rainfall occurs over the entire watershed after several days of drought (Shi et al. 2009). Thus, the value of the Hengshan Reservoir catchment is analysed by choosing typical rainfall events based on three empirical principles: (a) there are 10–20 non-rainy days before the rainfall; (b) the rainfall lasts more than 1 day with a total amount exceeding 50 mm; (c) the surface runoff is almost zero during and after the rainfall process, which is estimated by the change of measured water level and the stage–storage curve.

The resulting rainfall events are shown in Table 3 with an average total amount of 90.4 mm, indicating that the conditions are optimal for the antecedent soil moisture without surface runoff following non-rainy days (W_m). It is also found that the water level rises at an average rate of 5.3 mm per day during 6 days after the rainfall events. Therefore, based on the speculated subsurface runoff volume from previous soil moisture, R_b is 5.3 mm, W_Lm is 31.8 mm (5.3 × 6) and W_Um is 58.6 mm (90.4 − 31.8). After choosing the closest meteorological stations in the China Daily Ground Climate Dataset (http://data.cma.cn/), E is regarded as 4.2 mm by averaging the evaporation consumption on non-rainy days from June to September between 2006 and 2012. Thus, regardless of how much it rains, the remaining soil moisture in the upper layer gradually evaporates over the next 14 days (58.6/4.2), and N is finally obtained. Combined with the basic information of the Hengshan Reservoir, the values of the model parameters are shown in the last column of Table 2.

According to the national standard of hydrological forecasting (MWR 2008), the accepted range of water level forecast errors is within 20% of measured variations, and takes 0.1 m as a minimum value. Thus, 0.1 m is regarded as the maximum absolute error in this study. Based on the above results, although there is some error for the heavy rain class, the forecast is acceptable for rainfall events whose daily amount is above or close to the heavy rainstorm level (Figure 10(d)).

Following the previously described methodology, the WebGIS-based system is implemented and applied in the Hengshan Reservoir. The system functions are presented from the perspective of a reservoir manager, and a real example, when Typhoon Chan-hom hit southeastern China, is used to verify the decision-support for flood discharge during intensive rainfall in the flood season.

The numerous and widely distributed small reservoirs are characterized by the uneven levels of management and limited hydrological measurements. Their storage capacity is normally maximized for multiple purposes, but discharged for flood control and conservation during the flood season (Rodrigues et al. 2012; Song et al. 2015). In such circumstances, this study presents the WebGIS-based system as an effective tool to guide the flood control management particularly for reservoir managers. It is different from the existing flood early warning systems for large river basins and reservoirs (Roo et al. 2011; Wu et al. 2014; Thiemig et al. 2015), which focus on flood routing simulation for different lead times, and also from the flood assessment systems for local areas (Shivakoti et al. 2011; Demir & Krajewski 2013; Horita et al. 2015), which currently adopt web-based technologies to involve different communities. Since several institutional and social conditions have to be fulfilled when applying sophisticated flood control systems (Cools et al. 2016), a targeted and adaptable tool could play a more useful role in practical flood discharge of small reservoirs.

In order to make the result system concise and practical, this study introduces the agile development methods and tools during the entire system development process (Figure 2). They help to embody multidisciplinary knowledge from different fields, such as reservoir management, flood discharge, hydrological modelling, map development and software integration. Since different experts usually prefer popular methods within their respective domains, a loosely coupled structure of different tools and languages is used (Figure 3). The reservoir map browse is composed of three independent widgets, considering the uncertainties of spatial materials possessed by different reservoirs.

As the core of the system, the forecasting model is mainly based on rainfall in the catchment and water level in front of the dam. Aiming at concise information for flood discharge, it is a simplified transformation of lumped conceptual hydrological models, and adopts a horizontal water level to generalize the fluctuating water surface caused by flood routing (Zhao 1992; Shi et al. 2009; Lü et al. 2013). Meanwhile, the model currently uses daily lead time, considering the actual formulation mode of flood discharging scheme in the study area, and the applicability to other similar reservoirs that possess only daily hydrological measurements. A shorter forecast range, e.g., hourly, is also available using this method, but the corresponding flood discharge may have to include several results to make an overall decision.

According to the verification based on historical rainfall events, the forecast errors gradually decrease as the total rainfall amount increases (Figure 10(d)). In addition, the forecasted water levels are more likely to be greater than measurements in lighter rainfall, but smaller during heavier rainfall (Figure 9). The explanation and the analysis are as follows:

1. When acquiring the parameter values of the study area, typical rainfall events are chosen based on the empirical principles that surface runoff does not form during and after the rainfall process, but the water level gradually rises in the following days. Then, the total rainfall amount of each typical event is averaged to estimate the value of W_m; however, the evaporation consumption is ignored during the calculation process. Therefore, the resulting soil moisture value is slightly over-estimated, which is presumably the reason why the forecasted inflow flood volume tends to be smaller during heavier rainfall.

2. Runoff usually forms quickly after heavy rainfall, owing to the saturated soil moisture in the flood season. However, during the non-flood season, the rainfall infiltration from the upper to the lower soil layer requires further refinement, and the actual evaporation consumption should be greater than the flood season (Vieux 2001; Lü et al. 2013). Thus, the forecasted runoff volume tends to be greater than the actual value. This explains why the forecasted water level is more likely to be greater after lighter rainfall.

Therefore, detailed evaporation analysis is recommended for further improvement of the model. When estimating the antecedent soil moisture, the evaporation consumption in the upper layer should be classified by months and weather conditions, and deducted from the value of W_m. In order to acquire more accurate forecasts during the non-flood season and lighter rainfall, slightly more complex hydrological models are worth trying when distributed characteristics with relatively high resolution of the catchment are available. For example, the applicable CN values of the SCS-CN model could be calibrated combined with water level variations, so that the resulting runoff volumes are comparable (Mishra & Singh 2003; Shi et al. 2009). In addition, the usage of the WFlow model (http://wflow.readthedocs.io/) with rainfall interception and kinematic wave modelling could give the runoff process for each stream channel.

Due to the relatively sufficient flood discharging capacity and water storage requirements of the study area, the system currently relies on occurred rainfall, rather than numerical weather predictions, such as the Weather Research and Forecasting model (Bartholmes & Todini 2005; Skamarock et al. 2005), and satellite- and radar-based imagery data (Park & Hur 2012; Wu et al. 2014). However, meteorological forecasts are important to provide additional lead time to flood control preparedness, especially for the small reservoirs affected by flash floods. In addition, user interfaces of water level comparison are used to help formulate discharging schemes. This method is concise, but still requires empirical judgement by reservoir managers. Based on the above runoff process simulation, stage-wise flood control operation rules and optimization algorithms currently used in cascade reservoirs and river–reservoir systems (Che & Mays 2015; Chou & Wu 2015) have good reference value to balance flood prevention and water storage, as well as to flood discharge of small reservoirs.

Small reservoirs and their current status of flood management are garnering increased attention from both researchers and decision-makers. In this study, a WebGIS-based flood control management system with a single targeted forecasting model is outlined to provide decision support for flood discharge during intensive rainfall in the flood season. The features and contributions are summarized as follows:

1. In order to make the result concise and effective for reservoir managers, agile development methods are adopted during the entire implementation process, which helps to combine multidisciplinary knowledge from different fields. A loosely coupled structure of different tools and languages is used to integrate reservoir map browse, flood level forecasts and discharging schemes comparison, which enables the experts to use their popular methods from different domains.

2. Based on the hydrological measurements of rainfall in the catchment and water level in front of the dam, a flood level forecasting model with daily lead time is established by estimating the antecedent soil moisture and accumulated inflow flood volume. The forecast result is acceptable for the rainfall amount above or close to the heavy rainstorm level, according to the national standard of hydrological forecasting of China.

Currently applied in the Hengshan Reservoir in the lower reaches of the Yangtze River, the system is validated by historical rainfall events and a trial operation during the typhoon season. It is characterized by the usage of limited hydrological data that a small reservoir possesses, and the user-friendly interfaces for routine management and flood discharge for reservoir managers, which make it adaptable to other small reservoirs in humid regions. Since the optimization for flood control management is a complex and multifaceted issue, further improvements to the flood level forecasting model and study on the stage-wise flood control operation is recommended to better support the flood discharge of small reservoirs.

This study was sponsored by the National Natural Science Foundation of China (No. 41471460 and 41130750), Science and Technology Service Network Initiative (No. KFJ-SW-STS-174), ‘One Hundred Talents Program’ of the Chinese Academy of Sciences, and Postdoctoral Research Funding Programs of Jiangsu Province (No. 1601038B). We appreciate the editor and three anonymous reviewers for their valuable suggestions and comments.