Monitoring Container Terminals via SAR Imagery

CASE STUDY

Maritime Infrastructures

5.53^oW, 55^oE meridian

Introduction

Monitoring the complex operations of a container terminal—including seaside, yard, and landside activities—is essential for evaluating operational efficiency, productivity, and overall economic value [1]. As port automation and digitalization accelerate, integrating satellite-based monitoring with existing port processes creates a dynamic operational picture, enabling analyses that were previously unattainable.

Historical and predictive analytics can significantly enhance terminal management. Furthermore, real-time cross-referencing with data from other ports and global logistic hubs strengthens the transparency of heterogeneous supply chains. In this context, satellite remote sensing complements on-site data, providing port authorities and external stakeholders with a flexible business intelligence tool. This is particularly true for active microwave imaging sensors, which allow for continuous, global remote monitoring regardless of weather or light conditions.

The core challenge lies in accurately estimating the properties of observed objects from raw sensor measurements. In this post, we present a case study using ESA Sentinel-1 Synthetic Aperture Radar (SAR) imagery to estimate and monitor the total area occupied by containers within a terminal yard. While occupied area is a proxy—rather than a direct measure of total volume—it serves as a critical indicator of economic activity and logistic throughput.

The SAR Signal

The estimation process begins by discriminating between the free, asphalt-like yard surface and the areas covered by containers. At the C-band wavelength of Sentinel-1, a paved yard acts almost as a specular reflector; its normalized radar cross-section (sigma nought) typically ranges between -10 dB and -20 dB for VV and VH polarizations at standard incidence angles [2].

Conversely, metal containers act as dihedral reflectors. Their geometry triggers a double-bounce scattering effect, returning a significantly stronger backscattered signal to the sensor.

Figure 1-a) shows a typical SAR image of a container yard partially occupied by blocks. Figure 1-b) shows the empirical statistical distribution of the SAR sigma nought for the two type of surface, highlighting a clear separation: the asphalt signal intensity remains well below -10 dB, while container blocks occupy a much higher intensity range (red curve), allowing for robust threshold-based discrimination.

Figure 1-a). Example of SAR image of a port container yard partially covered by container blocks. Data: ESA Sentinel-1 SAR, polarization VV, pixel spacing: 10 m. Date: 2024/02/20.

Figure 1-b). Empirical statistical distribution of the radar cross section (sigma naught) of asphalt-like surfaces and surfaces covered by metal containers. Data: ESA Sentinel-1 SAR, polarization VV, pixel spacing: 10 m.

The Solution

Our approach to estimating storage occupancy follows a structured processing chain (Figure 2). Data is retrieved from the Copernicus Data Space Ecosystem (CDSE) via API. While the CDSE allows for server-side processing, our pipeline can also be implemented on-premise for custom requirements.

The workflow consists of:

Pre-processing: Noise reduction and pixel-wise log transformation of the SAR time series.
Detection: Application of an optimized intensity threshold to isolate high-backscatter container areas.
Morphological Enhancement: Binary morphological filters are applied to the output to suppress residual noise and refine object shapes.
Quantification: Calculation of connected pixel areas to derive the total occupied surface over time.

Figure 2. Processing scheme to estimate the container block area in a Sentinel-1 SAR image time series.

Results and Discussion

As an example, we processed two ten-year time series (case A and case B) to observe long-term trends, with an average temporal resolution of 30-days and 12-day, respectively:

Case A (Figure 3): The graph shows a clear, steady expansion of the occupied area, typical of a growing terminal.
Case B (Figure 4): This example reveals a different signature. Between 2016 and 2020, the terminal shows the oscillations of a fully developed facility. A sharp drop is visible during the 2020 COVID-19 pandemic, followed by a gradual four-year recovery toward pre-crisis levels.

Such trends, while requiring integration with in situ economic data or multispectral imagery (e.g., Sentinel-2 MSI) for full validation, provide an objective foundation for market intelligence and infrastructure investment analysis.

Figure 3-a). Case A: Example of Sentinel-1 SAR image of a container terminal in two different dates. Data: sigma naught, with VV polarization, 10m pixel spacing, 30 days average time spacing.

Figure 3-b). Case A: area covered by containers vs time, estimated by processing a 10 year time series of Sentinel-1 SAR images over the area of interest. The data are smoothed to filter sudden changes and highlight the growing trend of the area occupied by containers.

Figure 4-a). Caso B: a second example of Sentinel-1 SAR images of a container terminal. Data: sigma naught, with VV polarization, 10m pixel spacing, 12 days average time spacing.

Figure 4-b). Caso B: estimated area vs time for the area of interest in figure 4-a). Data: sigma naught, with VV polarization, 10m pixel spacing, 12 days average time spacing.

Limitation and Future Outlooks

While this case study demonstrates the power of remote monitoring, refinements are ongoing. A key focus for Meridian is the suppression of sidelobes—image artifacts caused by the extreme reflectivity of metal targets, which can lead to overestimation of the occupied area. Integrating SAR data with multispectral sensor data and economic indicators will further enhance the robustness of our interpretations.

Stay tuned for more updates!

Bibliography

[1] Tsagkaris, P.; Moschovou, T.P. The Impact of Automation on the Efficiency of Port Container Terminals. Future Transp. 2025, 5, 155. https://doi.org/10.3390/futuretransp5040155.

[2] M. Skolnik “Radar Handbook,” 3rd Edition, McGraw-Hill, Boston, 1990.

Disclaimer

The analyses reported herein are for illustrative and methodological purposes only.

Credits

This page contains modified Copernicus ESA Sentinel data (2026), processed via Copernicus Data Space Ecosystem (CDSE) by MERIDIAN – Data & Geospatial Intelligence.

Looking to optimize your asset monitoring?

Discover how to transform geospatial intelligence into a competitive advantage for your business.

Get in touch