Iceman.kz has spent significant time dedicated to recognition of ice rubble features from satellite imagery through the history of operations in the Caspian Sea. Stamukhi being vast mounds of ice rubble grounded on seabed are one of the most piculiar features in the region. They reach up to 16m as observed during the last 10 years. Although there are records of features with height up to 25 m observed during seasons before that.
Being a heavily grounded feature stamukha poses navigational threat to marine operations. Let alone the time of the season with ice cover they still persist in open water when majority of ice cover is already gone and light speed boats are already deployed into operation. As it is an unpassable obstacle for such units they become a navigational threat especially in low visibility conditions.
As for offshore structures stamukhi forming pits under their foundation create excess loads on pipelines. Custom case of rubble formation is pile-up on offshore sturctures’ faces that may interact with objects on deck and cause significant damage to equipment.
State of the Art
With huge amount of open access satellite imagery becoming available during the last years Iceman.kz has compiled a database of individual observations of stamukhi in the Caspian region. Each observation contains
- records of dates when they formed and disappeared with accuracy up to 3 days (average gap between images in the archive)
- lifetime duration
- width measured to resolution of the best available image for each observation. Normally this is 15 m confirmed with Landsat and Sentinel-2
During an ongoing season once newly acquired image is available on our servers for ice charting analysis, operator on duty performs stamukhi search utilizing internal data management system with the following algorithm:
- After mobile and stable areas are delineated visual investigation is carried out to spot features containing a point of high response with clear signs of accompanying features indicating presence of an obstacle in moving ice (open water channel left in the drifting ice as shown below, for example). A lot of targets are found during break-up when opening waters contrast out bright points often with tails of rubble floating downwind. Width of such features is then measured and accordingly recorded.
- The record also contains attributes of detection and erosion dates. Detection date is confirmed for each feature by backtracking the stack of season’s images to confirm the nearest in time image, when a high response feature is traceable and ice conditions indicate possibility of stamukhi formation with corresponding ice drift events. In some cases, features found closer to the end of the season are then tracked to the early days of winter, when the first drift events resulted in grounding of ice rubble before the area stabilized for the rest of the season.
- Sentinel-2 and Landsat high resolution cloudless and partly cloudy images through the length of the season are thoroughly investigated to ensure all the uncertainties with interpretation of SAR images over mainly stationary through majority of the season areas of ice cover is removed from the database.
Once each individual season was processed in such way by one operator and cross-checked by another to reduce bias from human perception all the observations were merged into single database.
Raw observations of individual features is published to our common access folder. Note the purpose of sharing extract of the dataset is to get acquainted with the format of the database and to perform studies in a commercially inattractive region.
Visualisation and Analysis
The dataset of individual observations as presented in the chart above has its utilization limits to visualize dependencies from other environmental parameters as well as does not allow showing intensity of the phenomena in a comprehensive and analyzable manner. Concept for stamukhi visualization was taken from report about distribution of Antarctic icebergs by Romanov Y, 2011 and it is generally a regularly spaced grid containing average and cumulative values summarizing point observations for each cell. Based on several sensitivity experiments each grid cell’s size was assigned 7.55 arcminutes along longitude and 5.39 arcminutes along latitude. This cell size corresponds to approximately 10×10 km that is more comprehensive for general public. Grid centroids were positioned geographically to have best compatibility with ERA5 reanalysis and forecast by ECMWF for easier correlation routines in future.
The major rationale behind using the size of grid was operational ice related risks analysis with this distance being sufficient for timely response to hazards within the cell for majority of vessels operating in the region. Figure below illustrates the grid visualized as regularly spaced data with continuously tiled surface. Grid’s coordinates are referencing the centroids of the tiles. Basic summary of observations for the last 5 years is presented in form of PowerBI project in data access section on our site.
Although the method above has proven its reliability through years of running the observation program Iceman.kz continues enhancement of detection algorithms with two different approaches to development to prove which works better and faster.
- InSAR technologies to work of original measurements from radar satellites
- Machine learning algorithms to work with higher resolution optic imagery
- Computer vision applications to process UAV video on the fly
Both methods are targeted to decrease processing time and human interaction with imagery.