notebooks

Demonstration Notebooks

We demonstrate the RABASAR multitemporal denoising on ALOS-1 and UAVSAR Images.

Outline

• Download data according using instructions
• Update config.json.
• Run through notebooks 0 - 4.

UAVSAR at Waxlake.

1. Follow the directions in the uavsar_waxlake/readme.md. Once completed, there should be uavsar_waxlake/data_original_tiff with the RTC images from UAVSAR. Check with QGIS or another GIS viewer.

2. Make sure config.json is properly configured.

3. Run the notebooks in order.

4. Inspect the products in out/uavsar_waxlake_tv/ or out/uavsar_waxlake_bm3d/ depending on the regularizer.

Note this could in theory be adapted at other sites at the UAVSAR Data Portal that have the newly added *.rtc file (this indicates the pixelwise multiplicative factor for the radiometric and terrain correction). However, note we had to manually download numerous files and organize them. This was quite time-intensive. Moreover, the UAVSAR data is not distributed with gdal-readable metadata so we needed to reformat these images so that we can use our GIS tools as usual.

ALOS-1 at Waxlake.

1. Follow the directions in the alos1_waxlake. Once completed, there should be alos1_waxlake/data_original_tiff with the RTC images from ALOS-1. Check with QGIS or another GIS viewer.

2. Make sure config.json is properly configured as below.

3. Run the notebooks in order.

4. Inspect the products in out/alos1_waxlake_tv/ or out/alos1_waxlake_bm3d/ depending on the regularizer.

Note this could more easily reproduced (relative to UAVSAR) because of the asf search tool which produces a python script for downloading time series determined via this GUI.

Example Config Files

TV

• ALOS-1 @ the Waxlake

  {
    "sensor": "alos1",
    "site": "waxlake",
    "regularizer": "tv",
    "spatial_weight": 1.0,
    "temporal_average_spatial_weight": 1.0,
    "ratio_weight": 1.0
  }
  

• UAVSAR @ the Waxlake

  {
    "sensor": "uavsar",
    "site": "waxlake",
    "regularizer": "tv",
    "spatial_weight": 1.0,
    "temporal_average_spatial_weight": 1.0,
    "ratio_weight": 1.0
  }
  

BM3D

• ALOS-1 @ the Waxlake

  {
    "sensor": "alos1",
    "site": "waxlake",
    "regularizer": "bm3d",
    "spatial_weight": 0.05,
    "temporal_average_spatial_weight": 0.005,
    "ratio_weight": 0.05
  }
  

• UAVSAR @ the Waxlake

  {
    "sensor": "uavsar",
    "site": "waxlake",
    "regularizer": "bm3d",
    "spatial_weight": 0.1,
    "temporal_average_spatial_weight": 0.03,
    "ratio_weight": 0.1
  }
  

Remarks about BM3D

The bm3d regularizer is quite complex and thus required many more computational resources. In the notebooks, we only apply the RABASAR with the bm3d regularizer to a 1000 x 1000 box otherwise such an application would likely require days to run using the current implementation. Moreover, we also note the weights across the different despeckling tasks (e.g. despeckling the ratio image and the temporally averaged reference had different weights). We suspect that bm3d is more sensitive to noise signatures than tv. Moreover, we used different parameters for ALOS-1 and for UAVSAR.