search

NOAA-OWP / inundation-mapping

Flood inundation mapping and evaluation software configured to work with U.S. National Water Model.
Other
95 stars 30 forks source link

[21pt] Lidar bridges #1242

Open CarsonPruitt-NOAA opened 3 months ago

AliForghani-NOAA commented 2 months ago

We downloaded 982 bridge lines from OSM within HUC 12090301, with 716 of them intersecting the TIFF files generated by Andy. Below is a report on the RMSE distributions of random sample points within these 716 bridge polygons, comparing our lidar-generated elevations with those from Andy's files.

I used only the last return points without excluding any specific classification, as we have not yet found a strong correlation between classifications and bridge decks.

Plot below shows RMSE results for lidar generated TIFF files with a 1-meter resolution, based on the assumption that the bridge deck is 3 meters wide:

image

Increasing the bridges width to 5m would slightly impact the accuracy: image

AliForghani-NOAA commented 2 months ago

@CarsonPruitt-NOAA The table below presents the RMSE (in meters) between the Z elevation of Lidar points within OSM bridge polygons (5m width) and the corresponding sampled elevations from Andy's TIFF files. Notably, 90% of the Lidar points are classified as 'Wire-Guard (Shield)', with a very low RMSE of 0.17m. <html xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:x="urn:schemas-microsoft-com:office:excel" xmlns="http://www.w3.org/TR/REC-html40">

class_code | Class_name | count_Percent | RMSE (m) -- | -- | -- | -- 13 | Wire-Guard (Shield) | 90.22 | 0.17 2 | Ground | 4.64 | 8.72 1 | Unassigned | 1.75 | 7.52 5 | High Veg | 1.56 | 3.93 3 | Low Veg | 1.02 | 8.61 7 | Low Point | 0.34 | 10.72 4 | Medium Veg | 0.33 | 5.25 14 | Wire - Conductor (Phase) | 0.07 | 0.63 9 | Water | 0.06 | 14.49 10 | Rail | 0.01 | 11.32 6 | Building | 0.00 | 6.03

I conducted a new run, this time using only classifications 13 and 2 (wires and ground). The results showed improvement compared to the run that included all classifications:

image

RobHanna-NOAA commented 2 months ago

I don't know if this will help, but I worked with LasTools a few weeks back. One feature they had was to be able to show for sudden elevation changes as well as the ability to see any quick classification change. I was able to use a bunch of tools to quickly isolate out different types of buildings. You may have already hit it but I imagine there is way to easily extract out bridges and possibly multiple ways to do it as well. Let me know if I can help. :)

AliForghani-NOAA commented 2 months ago

For over 600,000 OSM bridges across the US, I downloaded the last return LiDAR points and analyzed the percentage of points in different classification codes for each bridge. The table below presents typical results for some bridges. For instance, for OSM bridge 3144312, 54% of points fall under classification code 17, while 45% are under code 1. image

After identifying the dominant classification code for each bridge, the plot below illustrates the distribution of these dominant codes. The good news is that the majority of bridges belong to classification code 17, followed by class 2 (Ground) and class 1 (Unassigned). image

The results suggest that classification codes 17, 1, and 2 can be used for the final DEM creation. Additionally, classification code 13, the older standard for bridge decks, can also be included.

AliForghani-NOAA commented 1 week ago

I investigated the suitability of using classification code 1 (Unassigned) points from lidar data for the Memorial Bridge in Washington, DC. The distribution of elevations for classification codes 1 (Unassigned) and 17 (Bridge) is shown below, indicating numerous outliers in classification code 1. This suggests that classification code 1 cannot be used reliably in its current form.

image

We have two options in addressing classification code 1:

So, I suggest to mark classification code 1 points as noise. Later we decided to also mark ground points as noise. So, for now we only use classification codes 17 (bridge) and 13 (old bridge). still for all 'noise' points (including class 1 and 2 points) the code can use the elevation of the nearest bridge points (from classes 13, 17) to make sure we will have some points across the bridge domain (only for the sake of helping GDAL to make tif files).

CarsonPruitt-NOAA commented 1 week ago

Bridges like these are not well classified. We decided to leave these out of the LiDAR processing for now and add an enhancement to address these edge cases later.

Image

The bridge pictured here is half "low vegetation" and half "building". The location is on the Middle Oconee River near Athens, GA.