2nd Stage

The workflow for the 2nd and 3rd stages with exposed parameters is illustrated below.

_images/temporal_stacking_windowing.png

In case of overlapping tiles (deadzone_x > 0 and/or deadzone_y > 0), the image below shows the tile arrangement. The example only considers the neighboring tiles on the left/bottom of the blue tile. The same applies for neighboring tiles in the other directions.

Note: The overlap of tiles is always deadzone_[x|y] * 2 due to symmetries.

_images/tiles_overlap.png
class rsdtlib.Stack(sar_asc_path, sar_dsc_path, opt_path, sar_bands_name, sar_mask_name, opt_bands_name, opt_mask_name, tf_record_path, starttime, endtime, delta_step, tile_size_x, tile_size_y, bands_sar, bands_opt, deadzone_x=0, deadzone_y=0)

Class for stacking, assembling and tiling. The stacking is independent of each observation type. Assembling combines all observations over time, of types optical multispectral, and Synthetic Aperture Radar (SAR) in ascending and descending orbit directions.

Note: Even though the starttime might be later, earlier observations are still considered for stacking. That is, earlier observation pixels are carried forward if masked at starttime. However, only observation time stamps are effectively written, that are within the time frame of starttime and endtime.

Parameters:

  • sar_asc_path (str) – Directory of SAR observations (ascending)

  • sar_dsc_path (str) – Directory of SAR observations (descending)

  • opt_path (str) – Directory of optical multispectral observations

  • sar_bands_name (str) – Name of the SAR bands

  • sar_mask_name (str) – Name of the SAR mask

  • opt_bands_name (str) – Name of the optical multispectral bands

  • opt_mask_name (str) – Name of the optical multispectral mask

  • tf_record_path (str) – Tensorflow Record path to store results

  • starttime (datetime.datetime) – Starting time of the time series to consider

  • endtime (datetime.datetime) – End time of the time series to rconsider

  • delta_step (int) – The step value (\(\delta\)) to avoid redundant observations temporally close together

  • tile_size_x (int) – Tile size in x dimension

  • tile_size_y (int) – Tile size in y dimension

  • bands_sar (int) – Number of SAR bands for each orbit direction (\(b_{SAR}^{[asc\mid dsc]}\))

  • bands_opt (int) – Number of optical multispectral bands (\(b_{OPT}\))

  • deadzone_x (int) – Overlap tiles on x-axis so that neighboring tiles share 2*deadzone_x pixels per side (default = 0)

  • deadzone_y (int) – Overlap tiles on y-axis so that neighboring tiles share 2*deadzone_y pixels per side (default = 0)

Example:

Define the stacking, assembling, and tiling with different parameters. Observations are taken from sar_asc_path/sar_dsc_path for SAR data, and opt_path for optical multispectral data. The SAR and optical bands (“L1_GND” and “L1C_data”), and their masks (“dataMask” each) are explicitly specified. The timeframe to consider for stacking is 01-01-2017 till 01-07-2017. The step value \(\delta\) is set to two days, and y and x sizes of the tiles to 32 pixels each. The number of bands \(b_{SAR}^{[asc\mid dsc]}\) is two (for each oribt direction), and \(b_{OPT}\) is 13.

import rsdtlib
from datetime import datetime

sar_asc_path = "<SOURCE PATH OF S1 ASC OBSERVATIONS>"
sar_dsc_path = "<SOURCE PATH OF S1 DSC OBSERVATIONS>"
opt_path = "<SOURCE PATH OF S2 OBSERVATIONS>"
tf_record_path = "<DESTINATION PATH OF TFRECORD FILES>"

stack = rsdtlib.Stack(
                sar_asc_path,
                sar_dsc_path,
                opt_path,
                "L1_GND",
                "dataMask",
                "L1C_data",
                "dataMask",
                tf_record_path,
                datetime(2017, 1, 1, 0, 0, 0),
                datetime(2017, 7, 1, 0, 0, 0),
                60*60*24*2,   # every two days
                32,           # tile size x
                32,           # tile size y
                2,            # bands SAR
                13)           # bands opt
process()

Start the stacking, assembling and tiling process.

Return type:

None

Example:

Start the stacking, assembling, and tiling process as defined by the object stack. This will require significant compute resources depending on the amount of stacked observations.

stack.process()