DOI: https://doi.org/10.15407/uhmi.conference.01.03

DETERMINING ACCURATE ICE COVERAGE ON DANUBE
BY WEBCAMERAS

 

Gábor Keve 

Faculty of Water Sciences/National University of Public Service, Hungary 

Corresponding author: Gábor KEVE, National University of Public Service/Faculty of Water Sciences,
Hungary 6500 Baja, Bajcsy Zs. u. 12-14., keve.gabor@uni-nke.hu 

 

ABSTRACT

For most Hungarian rivers, especially the Danube, floods and other damages caused by ice have produced and are producing serious problems. Meanwhile, the number of national researches on ice that improve the effectiveness of ice protection is low, and technical development is not significant at this point. The main focus of the research presented in this article emphasizes the advancement of this research and to the further develop of the river ice monitoring methodology.  

The key objectives are listed in the following points:

Develop a fast, automated, cost effective, and continuous ice-collection method based on web camera images with a precision far beyond their manual or estimation procedures. Verification of the developed solution through error analysis. Solutions that do not require specialized software were preferential.

Analyze the time pulsation and daily travel curve of the ice jam coverage ratio of the Danube with the developed high frequency measurement process.

The aim of this paper is to promote modernization of the Hungarian ice-observations and to provide a numerical basis for scientific research related to this topic.

I have demonstrated that the web-based, automated river ice-monitoring system can be used as a detailed hydrographic tool and can provide more accurate results than the currently used estimation or manual image processing methods.  

I have proved that from the images of webcams to determine the rate of ice coverage, it is enough to imagine the

views of the cameras in advance, with a single spatial perspective transformation, it is not necessary to use georeferencing, orthorectification, or complicated form recognition procedures for each frame. From the perspective mapping, the aspect ratio of the pixels (pixels) to the water surface in the image being examined can be calculated, and it is sufficient for the computation of ice coverage in all images with the same viewpoint. By doing this, I've narrowed the task to the grading of the water-ice pixels. A simple numerical method was developed and verified to determine the area ratio of pixels to the surface of the water. I have developed an automatic, adaptable threshold value, which distinguishes between ice and water with appropriate precision as picture points (pixels).

With my method of ice coverage determination, I observed significant temporal pulsation and daily periodicity in the ice movement of the observed Danube reach. I have found that the small number of daily estimates are not representative to determine daily average ice coverage. I recommend continuous webcams monitoring.

The new findings contribute to a more accurate understanding of the spatial and temporal structures of ice floes in rivers, as well as the methodological development of their measurability and reproducibility.

My work creates the basis for the modernization of the Hungarian ice-monitoring network. The operation of such a network provides the condition that in the future on the larger rivers ice floe forecasting and alarm systems may be established. The time series collected over the past decades provide data for national research on river ice phenomenon’s too.

Keywords: fluvial ice, webcamera, ice coverage, ice observation, hyrdometry

 

INTRODUCTION

Despite the processes of global warming, that can be observed from the hydrological data during the past decades, the unfavorable mix of hydrometeorological factors can cause severe ice drift on the Danube in winter, resulting ice-floods. In 2002 I placed a web camera on top of a tall building on the Danube bank of Baja, based on the results of traditional ice photography experiment in the 70's. The success of my initial ice observation in 2008 resulted the construction of 5 additional cameras. The saved images recorded one of the special 30 to 40 km sections of the Danube (130 km) reach managed by the ADUVIZIG (Lower Danube Valley Water Directorate, Hungary). In 2009, 2010 and later during the winter of 2012, ice floes were produced in the examined sections. The recordings helped to improve ice forecasting, reduce the burden of the ice-breaker patrol service, and further research work. Unfortunately, the January 2017 event was only partially recorded due to the lack of maintenance of the camera system.

The revised and possible rethinking of the Technical Guidelines and Water Technical Assistance about the topic of ice detection, is a timely issue for the water resources service agencies. The revitalization is not justified primarily by the age of these documents, but by the fact that the water resources service agencies has been heavily restructured. There has been a lot of change in the number, qualifications, tools and financial background of the water resources industry, which suggest a re-examination of the above listed regulations.

The more important, and obvious reason to study river ice is that the most devastating floods were caused by ice floes in the lower part of the Danube in Hungary. During the last 180 years, extreme ice floods on the Danube occurred on the following occasions: 1838, 1839, 1850, 1876, 1878, 1883, 1891, 1920, 1923, 1926, 1929, 1940, 1941 (Lászlóffy, 1934) and finally the highest level was recorded in 1956.

We do not need to go too far back to look at past events; we only need to look at the damage caused by the recent river ice incidents in 2017 (recent). The ice floe in January 7, 2017 (without any warning) extended for almost the whole Hungarian Danube reach damaged the river signs and caused serious damage. In mid-February, the mass of ice in the river Tisza caused a lot of damage to the water equipment at most riverside towns. Perhaps the most celebrated event in the media, was the incident of the Tiszacsege ferry, where only with a little luck was a fatal accident avoided.

Protection against less frequent natural disasters, even if they cause serious damage, are only briefly given appropriate attention. Therefore, there is not enough interest to reveal the true cause and effect relationship. At these events, typically event-follow-up actions dominate, which often lacks a careful and detailed consideration for decision making. My work cannot solve all river-ice problems, but it tries to show as many aspects of it as possible. The main goal of the work is to modernize ice observation, which can provide a good basis learning the phenomenon more thoroughly and to analyze it scientifically.

This paper is a short outline of my PhD dissertation and that is the reason of using first person singular in the text, I sincerely hope it doesn't hurt anyone.

 

OBJECTIVES

For most Hungarian rivers, especially the Danube, floods and other damages caused by ice have produced and are producing serious problems. Meanwhile, the number of national research on ice that improve the effectiveness of ice protection is low, and technical development is not significant at this point. The main focus of the research presented in my dissertation emphasizes the advancement of this research and to the further develop of the river ice monitoring methodology.

The key objectives are listed in the following points:

The aim of this research was to promote modernization of the Hungarian ice-observations and to provide a numerical basis for scientific research related to this topic.

PRACTICAL APPLICABILITY OF WEB CAMERAS

In 2002, I installed my first webcam for ice-observation in Hungary, followed by several others. By presenting the operating experience and the wide use opportunities of cameras (modeling, flow estimation, ice forecasting, etc.) I explained and demonstrated the justification of the monitoring system.  

The practical applicability of the web camera system is broader than just the ice coverage ratio detection determination. Some example of usage as follows:

In 2018 I have developed an automatic, cost-effective image processing method that has a significant outcome that the system can be used as a hydrographic tool. I have verified by error analysis (Table 1) that the Self-Developed Auto Solution provides more accurate results than the official Hierographic data, any of the Estimating Methods, and even better than Manually Image Processing. Using the individually calibrated, transformed and evaluated camera images as reference, the square medium error (RMSE) was the lowest for the Automatic process:

 

Table 1. Error of ice coverage determination methods

Method:

Hydrographic

Estimation 1.

Estimation 2.

Manual

Automatic

RMSE:

13,15 %

9,03 %

9,59 %

10,18%

5,94 %

 

I have demonstrated that the web-based, automated river ice-monitoring system I have developed in the domestic water management practice can be used as a detailed hydrographic tool and can provide more accurate results than the currently used estimation or manual image processing methods.

I have shown that the practical applicability of the web camera system is broader than just the ice coverage ratio detection determination.

 

SIMPLE AND EFFECTIVE IMAGE TRANSFORMATION

After a thorough reading of the literature (Duguay et al., 2015; Chu et al., 2016; Kraatz et al., 2016; Ansari et al. 2017; Tóth, 2017) a kind of a general solution outlined, but it was not suitable for my problem. Solving the task of image evaluation, the necessary steps were these:

Instead of this process I used the image transformation process of Gálai (2008) for the first time in practice with three independent cameras (Paks, Baja, Mohács). The principle of this process is that I photographed a well-coordinated square grid plate in front of the camera in several positions. From the pictures I read the coordinates of the plane grid square in pixels (Fig. 1). The more pictures I have been able to process, the more accurate I got the internal parameters describing camera distortion. The water surface of the river, where I observed ice floes (in the test section), was approximately horizontal.  

 

 

Fig. 1. Well-coordinated square grid plate 

 

The average slope of the Danube here is negligible. I put my calibration plane horizontally with the help of two water levels perpendicular to each, and the observed area was opened in front of the camera, and in parallel plane I interpreted a parallel coordinate system. By this the necessary transformation (Fig. 2) has become simpler.

 

Fig. 2. Sketch of coordinate transformation task 

 

This coordinate system calculated the transformation function (equation 1) between image and reality (x, y, z) in the pixel of the camera (u, v). The ratio between water and ice can be interpreted in planes parallel to the water surface, so a single determination of the transformation between the horizontal plane plate and the camera image can be used to analyze the ice coverage at any water level. I also used the simplification that, due to the horizontal plane, the height of the spatial coordinate’s “z” is zero

 

                                (1)

Using this function, I calculated the spatial coordinates of each pixel of the camera's images. The procedure was verified by the area computation of any square grids of the plane used for the calibration, where the biggest error was within 3%.

I have proved that from the images of webcams to determine the rate of ice coverage, it is enough to imagine the views of the cameras in advance, with a single spatial perspective transformation, it is not necessary to use georeferencing, orthorectification, or complicated form recognition procedures for each frame. From the perspective mapping, the aspect ratio of the pixels (pixels) to the water surface in the image being examined can be calculated, and it is sufficient for the computation of ice coverage in all images with the same viewpoint.

Instead of expanding the Jacobi determinant proposed by Gálai (2008), I used a more practical, numerical solution of area calculation determined from real-coordinate squares of the pixel corner points. The pixel treated so far as a point can be treated as a square, which is apparent after an appropriate magnification. The square, however, has a planar extension, a surface area, and the area can be calculated based on its corner points. The idea was no matter how elementary I never found to be used for calculating ice coverage. I used the approximated that sides of the rectangles stay straight from transformation from (u, v) to (x, y). This is also true for small pixels.

The corner points of each pixel selected for observation were calculated using the four angular coordinates shown in Figure 3 Starting from the upper left corner, I numbered the corner points clockwise to 1-4. Using this sequence, which I use as the index (i) of the calculated real x, y coordinates, I calculated the actual area of each pixel I want to observe with the following (2) relation.

 

                        (2)

 

Fig. 3. Corner coordinates used for pixel area calculation

The calculated areas are placed in a matrix indexed with the u, v coordinates of the pixels, which provide the actual area of the pixel in question while moving in the area being investigated. By the described process, I have narrowed the task to the grading of the water-ice pixels.

 

 

 

ICE AND WATER SEPARATION

I converted color images provided by cameras to grey shades. Subsequently, I determined an area variable threshold value for distinguishing between ice and water. The use of a single threshold value in the picture has yielded insufficient results.

I have noticed that on the camera images, regardless of the day, the reflection on the water is always in the same direction. Along the direction of reflection, a lower and an upper threshold was linearly distributed, and this way the ice can be separated from the water with enough precision. My further development was to use the interpolation to determine the lower and upper threshold values using the sky from the earlier camera’s images and the clearly identifiable, permanently identifiable surface features. After the ice-water separation and area ratio determination functions were successfully resolved, the examined areas of the cameras were divided into 40 downstream lanes and the coverage ratios in these lanes were defined. The results were saved in the form of pictures (Fig. 4) and data series.

 

Fig. 4. Graphic result of the determination of ice coverage in the band under the Bay of Baja

 

I have developed an automatic, adaptable threshold value, which distinguishes between ice and water with appropriate precision as picture points. Finally, the reliability of the automatic method was verified by error analysis (Table 1).

 

RESULTS AND DISCUSSION

In all examined Danube profiles, the variable amplitudes pulsation of the ice coverage with time was observed. This pulsation is considerably higher than for example at water velocity measurements. Observations that vary this way are characterized by time-averaged values in the water resources industry. That is why I think it is justified to introduce this method at ice jams, along the use of cameras. In the domestic practice, the ice coverage of a river section is currently characterized by one to five estimates per day, although the intra-day period is not negligible. The arithmetic means of the small number of samples, therefore, is significantly different from the mean values derived from accurate steady, high-frequency observations. My results help to determine the optimal observational period of ice jams, whose information density can thus be traced to other hydrological and meteorological data, creating the basis for temporal and spatial analysis of the ice evolution and melting process based on observation.

Figure 5 shows the rapid fluctuation of the instantaneous ice cover and the sensitivity of the mean values to time ranges. Note that webcam measurements only give the daylight hours, and during the nights I approached the curve with linear interpolation.

 

Fig. 5. Ice Coverage Curve during the January 3, 3-week ice-jam of the Danube Baja section.

Legend: calculated by the ice observation camera (red), 1-hour average (blue),
6-hour average (orange) and finally average daily (black) time series

 

With my method of ice coverage determination, I observed significant temporal pulsation and daily periodicity in the ice movement of the observed Danube reach. I have found that the small number of daily estimates are not representative to determine daily average ice coverage. I recommend continuous webcams monitoring.

 

APPLICATION OF THE RESULTS

The new findings contribute to a more accurate understanding of the spatial and temporal structures of ice floes in rivers, as well as the methodological development of their measurability and reproducibility.

István Zsuffa's 1978 pioneering black-and-white industrial camera's continuous ice observation system has been revitalized and upgraded to create an ice-detect service based on ice coverage in real time, which is rarely can be found in the world. This system contributes greatly to the success of the ice floods and ice floe damage prevention work for the water resources agencies. In addition, it creates the possibility of scientific research on ice floes, data supply for in situ numerical modelling.

The research establishes and validates an automated process that can be used to measure the rate of ice coverage and the transverse distribution of the ice surface yield per unit width in consecutive flow cross section. Collection of the in-situ datasets requires serious effort, especially when the measurements are taking place on an icy river. Video recording is safe, but the thickness measurements needed to determine the ice volume flow must be carried out manually on the river. For the manual ice thickness measurements, I created and validated a device that allows simple and quick measurements on icebreaker ships. This result is extremely useful because from the dense timeseries of the transverse ice area ratio further analyses can be conducted.

My work creates the basis for the modernization of the Hungarian ice-monitoring network. The operation of such a network provides the condition that in the future on the larger rivers ice floe forecasting and alarm systems may be established. The time series collected over the past decades provide data for national research on river ice phenomenon’s

 

REFERENCES 

 

Ansari, S., Rennie, C. D., Seidou, O., Malenchak, J., Zare, S. G. Automated moditoring of river ice processes using shore-based imagery. Cold Regions Science and Technology. 2017 Vol. 142 P. 1-16.

Chu, T., Lindenschmidt, K. E. Integration of space-borne and air-borne data in monitoring river ice processes in the Slave river. Canada. Remote Sensing of Environment. 2016 Vol. 181 P. 65-81

Duguay, C. R., Bernier, M., Gauthier, Y., Kouraev, A. Remote sensing of lake and river ice. Remote sensing of the cryosphere. 2015 P. 273-306.

Gálai, A. River Ice Inspection by Webcameras. 2008 ULR: http://www.water.hu/ice/webcam/index.php?en 

Kraatz, S., Khanbilvardi, R., Romanov, P.: 2016. „River ice monitoring with MODIS: Application over Lower Susquehanna River. Cold Regions Science and Technology. 2016 Vol. 131 P. 116-128

Lászlóffy, W. A folyók jégviszonyai, különös tekintettel a Magyar Dunára. Vízügyi Közlemények. 1934. 3. szám [Lászlóffy, W. Ice conditions of rivers, with special regard to the Hungarian Danube. Announcements of  Watermanagement. 1934 Vol. 3. (In Hungarian)]

Tóth, R. R. Jégzajlás tér-idő viselkedésének elemzése videofelvételek alapján. BME TDK dolgozat 2017 [Tóth, R. R. Analyzing the space-time behavior of ice motion based on video recordings. Budapest University of  Technology and Economics Students Scientific Competition 2017 (In Hungarian)]

Zsuffa, I. HIDROLÓGIAI MÉRÉSEK Jégészlelés. Vízügyi Műszaki Segédlet, 1978 VMS 231/6-T/78.01/ G70

[Zsuffa, I. HYDROLOGICAL MEASUREMENTS Ice observation. Technical Guide of Watermanagement, 1978 VMS 231/6-T/78.01/ G70 (In Hungarian)]

Back to CONTENTS