Earthdata Forum

Hi, all,

Recently I find that SWOT Pixel Cloud Data Product (SWOT_L2_HR_PIXC_D) misclassifies cropland (may be characterized with high soil moisture) as water area, which leads to incorrectly wider river channel width estimation if cropland is contiguous with river channel in this situation (see attached figures).

Specifically, I directly utilize classified results from SWOT pixel cloud data product. Only pixels with with a water fraction above 0.1 and not bright land that were classified as open water, water near land, or dark water ("water_frac" >= 0.1, "bright_land_flag" = 0, "classification" ranges from 3 to 5), which is slightly modified from Simoes‐Sousa et al. (2025)

Does any experience this issue before?

Reference:
Simoes‐Sousa I T, Camargo C M L, Tavora J, et al. The May 2024 flood disaster in southern Brazil: Causes, impacts, and SWOT‐based volume estimation[J]. Geophysical Research Letters, 2025, 52(4): e2024GL112442.

Hi, would you please share the exact .nc file name that corresponds to the image that you shared?
E.g., SWOT_L2_HR_PIXC_042_158_157R_20251128T072513_20251128T072524_PID0_01.nc.

Also, how did you apply water_frac_uncert dataset in your water classification?

Thanks for your response. SWOT pixel cloud data utilized here is "SWOT_L2_HR_PIXC_019_200_086L_20240806T215721_20240806T215732_PGD0_01.nc".

As for "water_frac_uncert", I did not apply it for open water classification because I find "water_frac_uncert" value between open water (river) and cropland is trivial (see attached figure "water_fraction_uncertainty_comparison.png"). Spatial distribution of selected pixel cloud data with respect to open water and cropland is shown in attached figure "river_pixel_cloud_data.png" and "cropland_pixel_cloud_data.png", respectively.

Thank you for providing the exact file name.

Would you please tell me lat/lon index to subset the region?
I tried to plot the data but I could not locate the region that shows the same pattern from the file.

Here are the 2 images that I got - one without uncertainty and another with uncertainty < 1.0.

Thanks for your detailed description. The specific spatial extent is at the end of Daqing River, whose specific longitude range is from 116.875035 to 116.931715 and latitude range is between 39.046942 and 39.065448. This chosen region is not large because it is only used as a demonstration of misclassifying neighbouring cropland (potentially with high soil moisture) as open water.

Thanks for your time.

Hello, what do you think about this misclassification issue? In my opinion, it may be a research direction related with river channel geometry estimation (i.e., classification, top width) derived from SWOT Pixel Cloud Data Product.

Hi, thank you for sharing lat/lon bounds.

I tried to limit the plot for the specific bounds.
The above plot doesn't show any meaningful pattern that matches your plot.

How did you plot your dots and overlay image?
Is the image from Google maps?
Are we looking at the same data?

Thanks for your help and detailed description. Take a deep look into your all illustrations, I find that your last attchment plot (the bottom one) is filtered by "water_frac_uncert" < 0.1, not "water_frac_uncert" < 1.0 (mentioned words in your figure is right but your description words is wrong).

1) The dots is directly retrived from abovementioned specific SWOT pixel cloud data product "nc" file. The background overlay image includes Sentinel-2 true color image (Band4-3-2 composition) and high spatial resolution google earth image. I also have plotted a new figure without clipping (see attachment figure) to better compare with your displayed plots.

2) Both google maps and Sentinel-2 images (Sentinel image observation time is in the same day with SWOT pixel cloud data) are displayed in different plots.

3) Definitely we do.

Based on my further analyses, I find that recommended filter condition ("water_frac_uncert" < 0.1 or "water_frac_uncert" < 1.0) is not suitable for open water detection in this situation. If we choose to filter more noisy pixel clouds with "water_frac_uncert" < 0.1, a large portion of correct open water pixel clouds has also been removed mistakenly. But if we choose a loose filter condition with "water_frac_uncert" < 1.0, some noisy pixel clouds still exist (see attachment figure).

To summarize, a fixed threshold value with respect to "water_frac_uncert" parameter is not applicable for automatic open water detection in the large region. Perhaps, I believe, auxilliary data should be utilized for a more automatically accurate open water detection by SWOT pixel cloud data product and I also notice that there are some recent papers have made some progress about this.

Hi, thank you for your extra plots and explanation.
I apologize for my mistake in my last diagram.
I forgot to apply the same filters. The diagram is purely based on lat/lon bounds.

Here's what I got after applying filters & uncertainty.
For some reason, I don't get any matching data points in 116.50'E ~ 116.56'E as shown in your plot.
Data points start from 116.699'E.
I'm using Python. What software are you using for plotting data?

Thanks for your explanation. I just utilized python matplotlib package for plotting, the same with yours.

I think the underlying reason that you cannot find matching pattern with mine is incorrect recognization of geographic coordinate unit for visualization in the bottom figure. Please note that the top figure is displayed with unit of decimal degree with respect to longitude and latitude range, whereas it changes to degree-minute-second format in the bottom figure. In a word, the correct longitude range is 116.833-116.933 degree east and latitude range is 39-39.1 degree north in the bottom figure.