Weather Effects on Perception Sensors: Understanding Rain, Fog, and Road Spray
- Clemens Linnhoff
- 4 days ago
- 4 min read
Updated: 4 days ago
Reliable perception is one of the fundamental requirements for automated driving systems on-road and especially off-road. While lidar, radar, and camera sensors perform remarkably well under clear weather conditions, their behavior can change significantly when visibility deteriorates. Rain, fog, and road spray introduce additional interactions between sensor signals and the environment that can reduce performance and complicate object detection.
Understanding these effects is essential not only for sensor development and validation but also for creating realistic simulation models that accurately represent real-world conditions.
The Common Physics Behind Adverse Weather
Although rain, fog, and road spray appear very different, they affect active sensors such as lidar through two common mechanisms:
Atmospheric Attenuation
As light travels through the atmosphere, part of the emitted energy is scattered and absorbed by water droplets. Less energy reaches distant objects, and consequently less energy returns to the sensor. The result is a reduction in effective detection range and weaker object returns. Distant objects gradually become harder to detect as weather conditions worsen.
Atmospheric Detections
Weather particles themselves can generate sensor returns. Instead of only receiving reflections from objects in the scene, the sensor also receives reflections from droplets suspended in the air. These detections can contribute to clutter and make scene interpretation more challenging. While the underlying physics are similar, the resulting sensor signatures differ significantly between weather conditions.
Fog: A Wall of Atmospheric Detections
Fog consists of a large number of small water droplets distributed relatively uniformly throughout the atmosphere. For lidar sensors, this often creates a characteristic ring or cloud of detections close to the sensor. Since a significant portion of the emitted laser energy is scattered by nearby droplets, many returns originate from the fog itself rather than from objects in the scene. At the same time, attenuation reduces the visibility range of the sensor. As fog density increases, distant objects gradually fade from the point cloud until they disappear completely. The combination of strong atmospheric detections and reduced range makes fog one of the most challenging conditions for lidar-based perception systems.
Rain: A More Distributed Effect
Rain introduces larger droplets than fog but with a significantly lower spatial density. As a result, atmospheric detections are distributed over a wider range of distances. Instead of the distinct ring pattern often observed in fog, rain produces a broader distribution of weather-related returns throughout the scene. The level of atmospheric clutter depends on rainfall intensity, droplet size distribution, sensor wavelength, and signal processing algorithms. In addition to these detections, rain also attenuates the sensor signal, reducing the intensity of object returns and limiting detection range.
Road Spray: Weather Generated by Traffic
Road spray is particularly interesting because it differs fundamentally from both rain and fog.
Instead of being uniformly distributed in the atmosphere, road spray is generated by vehicle tires interacting with wet road surfaces. The resulting water droplets form dense spray plumes behind moving vehicles. In lidar data, these plumes often appear as distinct clusters of detections following traffic participants. Unlike fog or rain, the atmospheric detections are highly localized and directly coupled to the movement of surrounding vehicles. These spray clouds can partially obscure vehicles, reduce visibility, and introduce additional clutter into the sensor data. They are particularly adverse for object detection algorithms, as they tend to be very sensitive to detection clusters. Since road spray is strongly dependent on vehicle speed, tire geometry, road conditions, and traffic density, it is also one of the more complex environmental effects to model accurately.
Measuring Sensor Performance in Real Weather
To understand the impact of adverse weather, realistic measurements are essential.
At Persival, we use our Weather Test Bench to quantify the influence of rain and fog on lidar, radar, and camera sensors under real-world conditions.
The setup enables long-term measurement campaigns with calibrated reference targets and continuously monitored weather conditions. By correlating sensor performance with measured environmental parameters, we can systematically analyze how perception performance degrades as weather intensity increases.
This results in objective performance benchmarks, such as:
Detection range versus rainfall intensity
Detection range versus visibility distance in fog
Return signal strength under adverse weather
Weather-induced atmospheric detections as detection probability and distance distributions
These measurements provide quantitative evidence of how different sensor technologies behave under challenging environmental conditions.
From Measurement to Simulation
Real-world measurements are valuable on their own, but their impact becomes even greater when used for simulation model development. The performance curves obtained from weather measurements provide a direct basis for parameterizing simulation models. They allow simulation engineers to adjust atmospheric attenuation and atmospheric detection models so that virtual sensors reproduce the behavior observed in reality. The same measurement data can then be used for model validation, ensuring that simulated weather effects remain physically plausible and representative of real sensor performance.
This creates a continuous link between measurement and simulation:
Measure → Parameterize → Simulate → Validate
We have done all these steps in different customer projects. We can also help you to achieve greater confidence that your virtual testing campaigns accurately reflect real-world sensor behavior.
Conclusion
Rain, fog, and road spray all influence perception sensors through atmospheric attenuation and atmospheric detections. However, each phenomenon leaves a distinct signature in sensor data.
Fog often produces a dense ring of detections near the sensor and strong range reduction. Rain creates a broader distribution of atmospheric returns throughout the scene. Road spray forms localized clusters behind moving vehicles, introducing dynamic weather-related clutter.
Understanding these effects requires both accurate measurements and realistic simulation models. By combining real-world weather testing with physics-based simulation, engineers can better predict sensor performance and build perception systems that remain reliable even when the weather is not.
