Estimation of flow in various sizes of streams using the Sentinel-1 Synthetic Aperture Radar (SAR) data in Han River Basin, Korea

https://doi.org/10.1016/j.jag.2019.101930Get rights and content

Highlights

  • A novel method based on SAR images is proposed for stream discharge estimation.

  • Water surface area was related to stream discharge using an optimization algorithm.

  • Fifteen streams with varying sizes (mean discharge of 2–305 m3 s−1) were tested.

  • Power-law relationship was identified between water surface area and discharge.

  • The correlation coefficient ranged around 0.97 (calibration) and 0.89 (validation).

Abstract

This study proposes a novel approach of estimating stream discharge using the European Space Agency Sentinel-1 satellite data. Fifteen hydrological stations with a mean discharge ranging between 2 m3 s−1 and 305 m3 s−1 in the Han River Basin in Korea were chosen as test sites. A series of Sentinel-1 Synthetic Aperture Radar (SAR) images and observed streamflow data were selected to develop and validate the methodology. The methodology relates the stream discharge to the water area that is extracted from the satellite image analysis. The satellite image analysis involved the following sequential steps: (1) A series of SAR images for a given test site is pre-processed for thermal noise, radiometric calibration, speckle filtering and terrain correction; (2) The histogram matching technique is applied to all images to unify the backscatter distribution; (3) The polygonal area including the stream is manually delineated along the reach and the image filter with the shape of the polygon is applied to all images so that only the water area of the stream is extracted; (4) A single backscatter threshold value is applied to all images to extract a series of water area; (5) the power-law relationship between the series of the water area and the corresponding stream discharge is established and the correlation coefficient of the relationship is calculated; (6) The process of (4) and (5) was repeated to identify the optimal backscatter threshold that maximizes the correlation coefficient of the relationship. A clear relationship was developed for the 13 stations except for the two at which flow is highly influenced by hydraulic structures such as dam. At the 13 stream locations, the R value of the power-law relationship varied between 0.68 and 0.98 with a mean value of 0.89. The relationship was influenced by the geometric properties of the stream such as the size and side slope.

Introduction

Fresh water availability is a major concern of survival for one third of the world’s population (Vörösmarty et al., 2000). It is becoming more critical due to climate change, population growth and urbanization (Döll et al., 2015). In order to develop efficient management plans and to minimize the potential risks associated with freshwater, it is crucial to understand the flow regime of streams and rivers (Marsh, 2002). Hence, different types of streamflow observation networks have been developed which operate at local, national and regional scales throughout the world (Runge et al., 2005). The World Meteorological Organization (WMO) has stressed upon increased cooperation from its member states to strengthen the Global Runoff Database (GRDB) by contributing the national streamflow data. However, due to economic constraints and rationalizing priorities of water resources development organizations, it has become a challenging task to maintain and operate a large number of hydrological observation stations particularly in the inaccessible and remote reaches of watersheds (Alsdorf and Lettenmaier, 2003). This has led to a decrease of up to 75 percent in the annual streamflow monitoring globally since 1970 (WWAP, 2009; Sneeuw et al., 2014). Besides this, the commercialization of water distribution networks, proprietary rights and trans-boundary water agreements have also reduced free access to streamflow data for research purposes (Hannah et al., 2011). As a result, the worldwide decrease in streamflow monitoring has increased the uncertainty in disaster assessment, prevention, flood forecasting and watershed modeling (Dai and Trenberth, 2002; Hunger and Döll, 2007).

In this scenario, there is an increasing interest in using remote-sensing as a cost-effective and technically plausible tool for filling the gaps in streamflow estimation at ungaged locations (Wulder et al., 2012; Bjerklie et al., 2003; Barrette et al., 1988). This is because remote sensing has unmatched advantages to monitor hydrological processes on local, regional and global scales. The repetitive observation mechanism of satellites enables to study the dynamics of these processes on several spatial and temporal scales. The satellites can observe different physical properties of streams in time and space such as depth, width, velocity and surface area that can be related to stream discharge (Chu et al., 2008; Bjerklie et al., 2005).

Remote sensing of stream flow regime can be categorized into the techniques based on satellite altimetry, Synthetic Aperture Radar (SAR), and optical images. First, Satellite altimetry has been widely applied to monitor water surface elevation. Tarpanelli et al. (2013) and Jarihani et al. (2013) used multi sourced altimeters data to estimate change in water levels in large rivers and lakes. Koblinsky et al. (1993) used the GeoSAT satellite altimeter data to estimate the water level in the rivers and found a root-mean-square error of 70 cm when compared with in-situ observations. Morris and Gill (1994) used GeoSAT data to estimate the lake water level with a root-mean-square deviation of 11 cm. Birkett (1994) also used the GeoSAT data to estimate the relative change in water level with an accuracy of 10 cm in some of the world’s largest lakes and wetlands. Frappart et al. (2006) and Da Silva et al. (2010) estimated the water level of several rivers with a root-mean-square error of 30 cm using the ENVISAT altimeter data.

Second, the Synthetic Aperture Radar (SAR) images are obtained from active sensors onboard satellites, so they can consistently monitor spatial extent of water bodies regardless of weather conditions. Bjerklie et al. (2005) proposed a method of flow estimation for a single reach of a river based on remote-sensing and ancillary ground data, the stream width was determined from the Synthetic Aperture Radar (SAR) images and digital orthophoto quadrangles, the channel slope was obtained from the USGS 1:24,000 topographic map, while the surface velocity was estimated using the doppler interferometry technique of Goldstein et al. (1989). Smith et al. (1995) proposed another method of estimating flow rate in a braided channel, using the relationship between the effective width of the river extracted from the SAR satellite image and the flow rate measured at the ground station. Based on the same method, Smith et al. (1996) developed a relationship between the effective width and the flow rate for three rivers using the ERS-1 SAR imagery. They concluded that the accuracy of flow width-discharge relationship method is largely dictated by the length of selected reach and explicit parameterization of the channel slope and vegetation cover of the braided streams. SAR interferometry is a viable tool to determine changes in land surface elevation, but it has limited applicability to estimate changes in water level due to low coherence and specular reflection from open-water surfaces. However, Alsdorf et al. (2000) showed that the existence of floating vegetation on flood water can contribute to increased coherence. Hence this property was exploited to estimate a relative change of 2–5 cm in flood level. Hydraulic parameters derived from SAR satellite imagery can likely be used to run hydraulic models. Giustarini et al. (2011) showed that the periodic assimilation of ERS-2 and ENVISAT-derived water levels in a 1-D hydraulic model has reduced the uncertainty and greatly improved the model prediction. Similarly, Domeneghetti et al. (2014) showed that the ERS-2 and ENVISAT-derived water levels in combination with in-situ hydraulic data can be used to improve the calibration of a quasi-2D hydraulic model. The advantage of remotely-sensed data assimilation in hydraulic models is that the calibrated model can be used to estimate discharge at ungaged locations.

Third, optical satellite images have been widely applied to monitor earth’s surface processes. Xu et al. (2004) estimated the river flow rate using the stream width-water level, and water level-flow relationship obtained from the 1-m resolution optical images of IKONOS and QuickBird-2 satellites, their estimated flow agreed with the flow measured at five ground stations with an error of 5%. Smith and Pavelsky (2008) used the effective width of stream obtained from the 250 m resolution images of Moderate Resolution Imaging Spectroradiometer (MODIS) to estimate the river flow rate with an R2 value of 0.81. Pavelsky (2014) used the relationship between observed flow rate and effective stream width for a 62 km long reach of river and found that the estimated flow rate using this relationship has a comparatively low root-mean-square error of 7%.

The accuracy and frequency of monitoring the physical parameters of streams using satellites is largely influenced by the satellite spatial resolution, cloud cover and the satellite revisit time (Smith et al., 1996). The early stage SAR altimeters had an average orbital error of tens of centimeters (Koblinsky et al., 1993) and the spatial resolution of several hundred meters that were not suitable to observe the dynamics of small size streams, this error has now been reduced due to continuous orbital corrections and improved satellite technology (Birkett, 1998; Smith, 1997; Le Traon et al., 1995). However, the unexpected cloud covers in optical imagery and the relatively longer revisit time of 10–35 days for most of the operational satellites could still be an obstacle to detect sudden changes in stream width and water levels due to flash flood.

Although, several studies have attempted different approaches of streamflow estimation using satellite images. However, most of them have been used infrequently on large streams due to their low temporal and spatial resolution where the adopted approaches are less susceptible to errors (Roy and Mistri, 2013). This study proposes a novel methodology for discharge estimation on streams with a wide range of sizes and can address the weaknesses of the previous studies. First, to overcome the problem of limited remote-sensing data with irregular long intervals, we used the Sentinel-1 SAR images which are provided potentially at 6-day fixed interval and free of cost by the ESA regardless of the weather condition. Secondly, the relationship between water-surface area and observed flow rate is utilized due to the relatively low spatial resolution of Sentinel-1. The method is applied to fifteen hydrological stations on the Han River, Korea. The influence of the geometric properties of stream such as the size, side slope and the sensitivity of the water area to the flow rate is assessed.

Section snippets

Study area and observed flow data

This study is conducted in the Han River Basin, South Korea, which covers an area of ˜35,000 km². Han River runs for a total length of 495 km and is the fourth longest river in the Korean peninsula. Its three main tributaries are the Bukhan River, Namhan River and Rimjin River. Bukhan and Namhan Rivers originate in the north-eastern and south-eastern mountains, respectively, and converge together at Yangpyeong to form the Han River as shown in Fig. 1. The Rimjin River originates in the northern

Water area - flow relationship

Fig. 7 shows the relationship between the water area and the flow rate for fifteen study sites. The parameters of the relationship, the correlation coefficient for calibration and validation periods, the Nash-Sutcliffe Efficiency (NSE) index and the Root-Mean-Square Error (RMSE) are summarized in Table 2. Based on the mean observed flow, the streams were categorized into small (<30), medium (31–60) and large streams (>61 m3 s−1). The resulting rating curves corresponded well with the measured

Need of field observation

This paper relates the water area extracted from SAR images using the OTC procedure to the discharge values observed at the stream gaging stations. Therefore, the campaign to observe flow rate exactly at the time of satellite image acquisition is indispensable. Regarding this the following two points should be carefully considered: (1) The observed flow data obtained from the field campaign should include a wide range flow conditions; (2) The time difference between streamflow observation and

Conclusion

A satellite - based streamflow estimation model is presented and applied to fifteen hydrological stations on different tributaries of Han River basin, South Korea. The SAR images of ESA Sentinel-1 satellite and the ten-minute flow data of ground observation stations is used as input. The histogram matching of the series of SAR images improved the consistency and helped in the accurate identification of water area using the OTC technique based on a single threshold intensity. As a result of

Acknowledgement

This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (Project No. NRF-2017R1C1B2003927).

References (52)

  • D. Alexakis et al.

    Soil moisture content estimation based on Sentinel-1 and auxiliary earth observation products. A hydrological approach

    Sensors

    (2017)
  • D.E. Alsdorf et al.

    Geophysics. Tracking fresh water from space

    Science

    (2003)
  • D.E. Alsdorf et al.

    Interferometric radar measurements of water level changes on the Amazon flood plain

    Nature

    (2000)
  • P. Ashmore et al.

    Prediction of discharge from water surface width in a braided river with implications for at‐a‐station hydraulic geometry

    Water Resour. Res.

    (2006)
  • E.C. Barrette et al.

    Satellite remote sensing requirements for hydrology and water management from the mid-1990s, in relation to the Columbus Programme of the European Space Agency

    Hydrological sciences journal

    (1988)
  • C. Birkett
    (1994)
  • C.M. Birkett

    Contribution of the TOPEX NASA radar altimeter to the global monitoring of large rivers and wetlands

    Water Resources Research

    (1998)
  • G.R. Brakenridge et al.

    Orbital microwave measurement of river discharge and ice status

    Water Resour. Res.

    (2007)
  • Y. Chu et al.

    Monitoring Level Fluctuations of the Lakes in the Yangtze River Basin from Radar Altimetry. Terrestrial

    Atmospheric & Oceanic Sciences

    (2008)
  • J.S. Da Silva et al.

    Water levels in the Amazon basin derived from the ERS 2 and ENVISAT radar altimetry missions

    Remote Sensing of Environment

    (2010)
  • A. Dai et al.

    Estimates of freshwater discharge from continents: latitudinal and seasonal variations

    J. Hydrometeorol.

    (2002)
  • S.G. Dellepiane et al.

    A new method for cross-normalization and multitemporal visualization of SAR images for the detection of flooded areas

    IEEE Trans. Geosci. Remote Sens.

    (2012)
  • P. Döll et al.

    Integrating risks of climate change into water management

    Hydrological Sciences Journal

    (2015)
  • N.B. Duy

    Automatic detection of surface water bodies from Sentinel-1 SAR images using Valley-Emphasis method

    Vietnam. J. Earth Sci.

    (2015)
  • O. Elmi et al.

    River discharge estimation using channel width from satellite imagery

    Geoscience and Remote Sensing Symposium (IGARSS), 2015 IEEE International

    (2015)
  • L. Giustarini et al.

    Assimilating SAR-derived water level data into a hydraulic model: a case study

    Hydrol. Earth Syst. Sci.

    (2011)
  • Cited by (24)

    • Sentinel-1 based Inland water dynamics Mapping System (SIMS)

      2022, Environmental Modelling and Software
      Citation Excerpt :

      Rather than using a single shapefile in the above-mentioned applications, utilizing the temporal collection of shapefiles shall allow researchers to capture the realistic dynamic nature of the water body, thereby providing a vital understanding of the actual scenario. Combining the derived temporal information with complementary datasets like satellite altimetry, which measures water level, opens up huge possibilities of inland water monitoring such as estimation of surface water volume dynamics, river discharge dynamics, etc. (Frappart et al., 2005; Ahmad and Kim, 2019). With the futuristic SWOT mission, potential of altimetry would improve dramatically considering its capability to observe smaller and narrower inland water bodies.

    • Channel responses to flooding of Ganga River, Bihar India, 2019 using SAR and optical remote sensing

      2022, Advances in Space Research
      Citation Excerpt :

      These phenomena have usually taken place on a large scale particularly in monsoonal river catchments (Agnihotri et al., 2019). To assess the potential risk associated with these large fluvial systems and to develop efficient management plans to mitigate the risk of these environmental calamities, accurate information regarding the formation or removal of geomorphic features, pre-flood and peak flood information, in general, is indispensable (Ahmad and Kim, 2019). Several techniques are developed to assess the changes in river planform by acquiring the referenced and concurrent data to the flooding time (Thakur et al., 2012; Wang et al., 2018; Khan et al., 2018; Ali et al., 2019), including field surveys, remote sensing, physical and numerical modelling.

    • Estimation of river flow using CubeSats remote sensing

      2021, Science of the Total Environment
      Citation Excerpt :

      As the deluge of big data continues to impact practically every commercial and scientific domain, geosciences have also witnessed a major revolution from being a data-poor field to a data-rich field (Karpatne et al., 2019; Reichstein et al., 2019). The use of remote sensing by satellite for streamflow analyses can be categorized into techniques based on satellite altimetry, Synthetic Aperture Radar (SAR), and optical images (Ahmad and Kim, 2019). To accurately detect river flow regimes at a field scale from space, a new high spatial and temporal resolution remote sensing source is necessary to improve water-level time series (Bogning et al., 2018).

    • Precipitation threshold for urban flood warning - an analysis using the satellite-based flooded area and radar-gauge composite rainfall data

      2020, Journal of Hydro-Environment Research
      Citation Excerpt :

      The SAR images are obtained by sensing the electromagnetic waves shot from the satellite penetrating the clouds, so they contain the information on the water body not only during the incremental weather condition but also during the night time. For this reason, SAR images have been widely used to detect the water body area (Ahmad and Kim, 2019), the riverine floodplain (Brivio et al., 2002; Matgen et al., 2004; Manjusree et al., 2012; Schumann et al., 2012; Schlaffer et al., 2014) and the urban flooded area (Matgen et al., 2010; Giustarini et al., 2012; Pulvirenti et al., 2016; Mason et al., 2009, 2014, 2018; Tanguy et al., 2017). Flooded areas are typically identified by applying a threshold backscatter value to the image.

    View all citing articles on Scopus
    View full text