A novel remote sensing method to determine reservoir characteristic curves using high-resolution data

The reservoir characteristic curves mainly refer to the relationship curves of reservoir water level–reservoir capacity (LC curve) and reservoir water level–reservoir water surface area (LA curve). These curves are the main parameters reflecting the scale and characteristics of the reservoir as well as an important basis for the design of the reservoir dam, the determination of the reservoir operation mode, and the flood control review, detection and reinforcement of the reservoir dam (MWR of PRC 2017). Previously, in the field of hydrology-water resources and hydraulic engineering, the methods for determining the reservoir characteristic curve were mostly reckoned based on the topographic survey data (Zhang et al. 2019). But, in fact, there are many disadvantages in these methods, for example, small field of vision (NASA 1998; Ren et al. 2021), heavy field surveying workload (Pal et al. 2018; Cucchiaro et al. 2019; Galicia et al. 2019), and low accuracy (Bossler et al. 2001; Sun et al. 2021). With the development of remote sensing technology, resource satellite data are extensively used in correlative research of reservoirs and dams. Huang et al. (2009) identified river bed for check-dam engineering using SPOT-5 images. Hui et al. (2010) planned and designed the dam system in a grid-based collaborative virtual geographic environment based on resource satellite data. Meanwhile, the watershed underlying surface information was accurately identified by remote sensing data. Boix-Fayos et al. (2008) assessed the impact of land use change and check dams on catchment sediment yield. Liu (2021) extracted the planting structure of crops in Hetao irrigated area based on Sentinel-2. In the domain of water pollution control, Li et al. (2016) and Wang et al. (2019) assessed the non-point source pollution risks in the water protection zone of the reservoir area based on remote sensing images. Also, Yao et al. (2020) studied water pollution control methods in the Yangtze River Basin by artificial intelligence technology based on satellite data. In addition to these researches, the assessment of erosion in the reservoir area was carried out based on resources satellite images (Abedini et al. 2012; Wang et al. 2016; Arabameri et al. 2019). In this paper, a novel method to determine the reservoir characteristic curve was carried out, which is proposed on resource satellite data processing. By using this methed, the accuracy of the characteristic curves of reservoirs will be improved, and the workload will be reduced. As demonstrations, the characteristic curves of Jinshuitan Reservoir and Shitang Reservoir were determined accurately and efficiently.

Jinshuitan Reservoir and Shitang Reservoir are located in Longquan Stream, a Daxi tributary at the upper reaches of the Ou River, the second largest water system in Zhejiang Province, as shown in Figure 1. The catchment area of Jinshuitan Reservoir is 2,773 km², accounting for 77.8% of the total area of the Longquan Stream Basin, and the catchment area of Shitang Reservoir is 467 km². The control catchment area of the two reservoirs is 3,240 km² in total, accounting for 18.1% of the total area of the Ou River Basin. The catchment areas of the two reservoirs are located in the temperate monsoon climate zone with mild climate and abundant rainfall. The average annual precipitation is 1,833.8 mm. The average annual flow of the dam site of Jinshuitan Reservoir is 100 m³/s, and the total average annual runoff is 31.5 × 10⁸m³. The distribution of runoff during the year is extremely uneven. The runoff in the rainy season from April to June accounts for 58.1% of the yearly total, of which the most concentrated is in June, accounting for 24.8% of the yearly total. The runoff in the second flood season from July to October only accounts for 21.64% of the yearly total. The dam site of Shitang Reservoir has an average annual flow of 109 m³/s, and a total average annual runoff of 34.4 × 10⁸m³. There are three hydrology stations, nine precipitation stations, one evaporation station, and three water level stations in the whole area, as shown in Figure 1. The ground-breaking of Jinshuitan Reservoir was made in October 1978. Its main project construction was commenced in October 1981, closure was conducted on October 31, 1983, and impoundment was started on June 24, 1986. The construction of Shitang Reservoir was officially commenced on July 1, 1985, and impoundment was started on December 31, 1988.

The China–Brazil Earth Resources Satellite is a satellite (code CBERS) approved by the joint protocol of the Chinese and Pakistani governments in 1988 and jointly invested and developed by the two countries (Zhang et al. 2021). On October 14, 1999, China–Brazil Earth Resources Satellite 01 (CBERS-01) was successfully launched, and it operated in orbit for 3 years and 10 months. Satellite 02 (CBERS-02) was launched on October 21, 2003. The China–Brazil Earth Resources Satellite has a flying height of 778 km and an operation cycle of 100.26 min, which means that it can cover the entire Earth within 26 days. The CCD camera carried on the satellite can continuously shoot the ground image in the range of 113 × 113 km with a ground resolution of 19.5 × 19.5 m. It has four bands in visible and near-infrared spectral ranges and one panchromatic band. The China–Brazil Earth Resources Satellite flies over Jinshuitan Reservoir and Shitang Reservoir generally at 9:30–10:30 a.m.

Since 1972 when the first land satellite was launched, the United States has successively launched seven similar Earth resources satellites (Liu et al. 2021; Mo et al. 2021). Every new satellite has higher data quality and resolution, and a large number of high-resolution multi-spectral remote sensing data has been provided to users around the world. As the computer technology is developed rapidly, the Earth resources satellite data have been widely used in various fields, such as agriculture, forestry, water conservancy, geography, geology, petroleum, and military fields. The Landsat currently flying at a high altitude has a flying height of more than 700 km and an operation cycle of 98.9 min, which means that it can cover the entire Earth within 16 days. The satellite is equipped with a TM and an ETM+. It can continuously shoot the ground image in the range of 185 × 185 km with a ground resolution of 30 × 30 m (15 × 15 m in panchromatic band). The imaging instrument will automatically turn off in the case of rain and a cloud amount of greater than 70%. The time for the Landsat to fly over Jinshuitan Reservoir and Shitang Reservoir is also generally 9:30–10:30 a.m.

Figure 2 shows the satellite color composite image of the area dated July 24, 2008 where Jinshuitan Reservoir and Shitang Reservoir are located and Figure 3 shows the remote sensing composite satellite image imitating natural color of the water catchment of the two reservoirs.

At present, the types of image data that can be provided by satellite ground receiving stations in China include photo data, negative film data, and digital data. Among them, digital data have the highest accuracy. The satellite image in the form of digital data was used in the research.

Since the construction of Jinshuitan Reservoir and Shitang Reservoir, the water storage in the reservoir area has experienced different water levels. The highest water level of Jinshuitan Reservoir was 186.66 m, and that of Shitang Reservoir was 103.04 m. According to the highest and lowest water levels that have occurred in the reservoirs and in comprehensive consideration of the measured synchronous water level data, a total of 36 satellite images were used in the research, 10 from China–Brazil Earth Resources Satellite, and 26 from Landsat. Twenty-five images were used for Jinshuitan Reservoir and its corresponding water table drop was about 164.8–184.5 m and average interval about 0.79 m. Eleven images were used for Shitang Reservoir, and its corresponding water table drop was about 98.0–102.5 m and average interval about 0.41 m. The satellite data used are shown in Tables 1 and 2.

When the resource satellite scans and images a ground object at a high altitude, the scanning image and the actual position are systematically distorted due to factors such as the flying attitude, height and speed of the resource satellite, the Earth rotation, and the uneven scanning speed. Therefore, the resource satellite data obtained from the ground station must be corrected so that the remote sensing data processing results can be consistent with the actual terrain. Although the remote sensing satellite ground station has done routine correction on the data after reception, it is still insufficient for topographic survey and area measurement with a high-accuracy requirement, and geometric precision correction must be performed, that is, geometric position checking is performed using ground control points to correct the remote sensing satellite image onto a unified geodetic coordinate grid.

The interpolation generates a high-accuracy result. It not only maintains the continuity of the pixel without reducing the resolution, but also has the effect of edge enhancement, and the corrected satellite image can be consistent with the topographic map. Both the fine correction processing of the satellite images and the processing of various other images used in the research were completed on the ENVI platform of the remote sensing image processing system.

The ground object spectrum received by the land satellite is comprehensive information of various objects on the ground, and image enhancement techniques are required to highlight the water bodies. The method used was edge enhancement and filtering. This method scans the entire image using a filter kernel to highlight the water and land boundaries. Then, on this basis, the remote sensing platform is used to identify water bodies.

In order to obtain accurate reservoir water surface area, water body identification on remote sensing images is an important link. The water radiation rate on the near-infrared band is obviously single and lower than other ground objects, thus the water bodies can be effectively determined. Therefore, a suitable near-infrared band is selected to determine the water body threshold, so as to determine the water bodies on the remote sensing images. For the pixel identification at the land–water interface, the ratio calculation method is used for water body identification. To further enhance the accuracy of area calculation, remote sensing images are converted into images of 15 × 15 m pixels using bicubic method.

After water body identification, the water body pixels on each satellite image are counted one by one, and the water surface area value of the reservoir area at different water levels can be obtained. No human errors will be caused to the area obtained using the method and the accuracy is high. The statistical values of the water surface area corresponding to the satellite image maps of Jinshuitan Reservoir and Shitang Reservoir at different water levels are shown in Tables 3 and 4. The water surface shapes are arranged from large to small as shown in Figures 4 and 5.

Tables 5 and 6 show the water surface area and capacity of the two reservoirs at different water levels. Figures 6 and 7 are the characteristic curves of the two reservoirs.

The accuracy of the method of determining reservoir characteristic curves proposed in this paper mainly depends on three factors: (1) the number of images used; (2) the analysis and calculation of water surface area; and (3) the configuration of curves.

For the same water level elevation difference, the more the satellite images collected, that is, the smaller the water level difference between two adjacent images, the higher the calculation accuracy. The total water level elevation differences of Jinshuitan Reservoir and Shitang Reservoir in this research were 20.5 m and 4.5 m and 25 and 11 satellite images were used, respectively. The average water level differences between two adjacent images were 0.79 m and 0.41 m, which could meet the requirements for determining the reservoir characteristic curve. However, weather and other factors created difficulties for the collection of satellite data of some water level segments, resulting in the water level difference between some images of Jinshuitan Reservoir exceeding 1 m and that between some images of Shitang Reservoir exceeding 0.5 m. For example, the water level difference between No. 4 and No. 5 in Table 1 reached 1.11 m, and that between No. 8 and No. 9 in Table 2 reached 0.67 m. After new information is obtained, supplementary information and revision can be made.

In theory, after the Earth resources satellite image data are geographically fine-corrected as per the 1:50,000 topographic map, it is equivalent to the topographic map whether in latitude and longitude, geographic coordinates, or kilometer grid, which can completely guarantee that its span error is kept within one pixel, that is, 15 × 15 m. Therefore, when the water area is counted on such images, the final global error is not more than 15 × 15 m. The water surface area is actually an irregular figure. It seems the total area of the water surface is obtained by placing the water surface area on a 15 × 15 m grid. Of course, for the actual water boundary, some will occupy more than half of the grid, some only occupy less than half of the grid. When the water area is counted using the images, those occupying more than half of the grid are included and those occupying less than half of the grid are rejected. Pursuant to the grid adjustment theory, the final error is a grid, namely, 15 × 15 m.

In this paper, the polynomial curve equation configuration software under the platform of satellite remote sensing image processing system was used for curve configuration, which has avoided the error caused by previous methods. The statistical parameters of curve configuration are shown in Table 7.

In order to test the configuration accuracy of the polynomial curve equation, the water surface area corresponding to the reservoir water level on the 36 satellite images was calculated through the polynomial and compared with the original area, as shown in Tables 8 and 9. It can be seen from the tables that the absolute error of most points of Jinshuitan Reservoir is less than 0.1 km², with an average of 0.012 km², and the relative error is less than 0.5%, with an average of 0.05%, while the absolute error of most points of Shitang Reservoir is less than 0.01 km², with an average of 0.009 km², and the relative error is not more than 0.5%, with an average of 0.154%. Therefore, both the formulation parameters of the equation and the accuracy analysis of the curves show that the reservoir characteristic curve equation determined in this paper is highly accurate.

The existing characteristic curves of Jinshuitan Reservoir and Shitang Reservoir were obtained on April 20, 2007 via collation based on the data available at the time of database creation. The analysis mainly compared the water level–capacity curves in the research with the existing water level–capacity curves of the two reservoirs, as shown in Tables 10 and 11 and Figure 8.

As indicated by the figure, at the same water level, the capacity value of the new characteristic curve is smaller than that of the existing characteristic curve. At the water level of 185 m, the capacity of Jinshuitan Reservoir is reduced by 2.06%, being 1.04758 billion m³, while the capacity of Shitang Reservoir at the water level of 103 m is reduced by 1.47%, being 76.23 million m³. After analysis, the reasons are as follows:

1. After years of operation of the reservoir, soil erosion occurs in the reservoir area. This part of the soil is deposited in front of the dam, resulting in a decrease in reservoir capacity.

2. Through the comparative analysis of satellite images over the years, urbanization has a great impact on the two reservoirs. After field study, part of the spoil from highway construction enters the reservoirs, and the reclamation and development is serious. These are the direct causes of the reduction of reservoir capacity.

Figures 9 and 10 are satellite images of Anren Town and Dayuan Township near Jinshuitan Reservoir and Shitang Reservoir before and after urbanization.

The method has effectively overcome the shortcomings of previous methods, and its advantages are as follows. (1) A large visual field. With a relatively big ground covering extent, the satellite data can fully and accurately describe the whole picture of the reservoir area, which is impossible with previous methods. (2) A fast cycle. The satellite data in the same region can be re-acquired every few days, so the data accumulation in each region is rich, and a series of data can be provided for reservoir dynamic changes under different conditions (e.g., different water levels). (3) Updated data. After the satellite passes the region, the ground station will receive the data. After initial processing, a map will be formed within a short period of time. It can comprehensively and quickly reflect the impact of the profile of the current basin, reservoir, and human activities on the reservoir area. (4) Extensive information. During satellite imaging, not only the visible bands but also the infrared band information that is invisible to the human eye are recorded, and converted to the image visible to human eyes, thus the recognition scope is expanded. If the water body is fully absorbed on the near-infrared band, the image will be black, while the land, vegetation, and other ground objects are strong diffuse reflection objects and they reflect the near-infrared band to different extents, which is enough to form a sharp contrast with the black water surface. This provides favorable conditions for the identification of water area. (5) Fewer constraints. The satellite is not limited by geographical and regional conditions. Even for remote and uninhabited areas, ideal data are also available. In addition, it is also not subject to climate (e.g., bad weather) and ground objects (e.g., vegetation height and density) and a great deal of hard field work is completed indoors. (6) High accuracy. Since the satellite remote sensing image is a high-resolution image, it is quite accurate to calculate water area based on the water range on the satellite image. In the meanwhile, various human errors and instrumental errors caused by the use of a planimeter for survey and calculation on the topographic map are also eliminated. In view of the above advantages, the remote sensing method of determining reservoir characteristic curves using the data of China–Brazil Earth Resources Satellite and Landsat was adopted.

The method of determining reservoir characteristic curve was studied using the CCD image data of China–Brazil Earth Resources Satellite and TM/ETM image data of Landsat with a high ground resolution through image processing technology such as computer classification combined with the accurate water level data of the reservoir area. The precision analysis and the comparison with the existing characteristic curves show that the reservoir characteristic curve determined using the method is highly accurate.

The application of satellite remote sensing technology to determine the reservoir characteristic curve has unique advantages. It can not only quickly and reliably obtain the ground information of an area usually lacking in data or an inaccessible area, but also save a large amount of expense compared with ground survey. Moreover, it can fully meet the needs of reservoir operation, management, and application.

With the enhancement of ground resolution of the Earth resources satellite, the further development of computer classification technology, and the support of GIS, the application of resource satellite data to water conservancy research has wide-ranging prospects.

This work was funded by the National Key Research and Development Program (Grant Nos. 2018YFC0407704 and 2016YFC0401005) and the National Natural Science Foundation of China (Grant Nos. 51779146 and 51879163).

Data cannot be made publicly available; readers should contact the corresponding author for details.