6  In-class Exercise 6: Dielectric Properties of Natural Media

In this notebook, we will investigate the varying backscatter values associated with different land surfaces like water bodies, forests, grasslands and urban areas. We will use backscatter data from the Sentinel-1 satellite and we will utilize the CORINE Land Cover dataset to classify and extrapolate these surfaces, enabling us to analyze how different land cover types influence backscatter responses.

import numpy as np
import pandas as pd
import xarray as xr
import rasterio
import rioxarray  # noqa
from shapely.geometry import box
from shapely import affinity
import json

from functools import partial
from pathlib import Path
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
from matplotlib.colors import ListedColormap, BoundaryNorm
import holoviews as hv
from holoviews.streams import RangeXY

6.1 Load Sentinel-1 Data

For our analysis we are using sigma naught backscatering data from Sentinel-1. The images we are analyzing cover the region south of Vienna and west of Neusiedlersee Lake. We load the data and and apply again a preprocessing function. Here we extract the scaling factor and the date the image was taken from the metadata. We will focus our attention to a smaller area containing a part of the Neusiedlersee Lake and its surrounding land. The obtainedxarray dataset and is then converted to an array, because we only have one variable, the VV backscatter values.

data_path = Path("~/shared/datasets/rs/sentinel-1/neusiedler").expanduser()


def _preprocess(x, bbox):
    """
    Preprocess file.

    Parameters
    ----------
    x : xarray.Dataset
    xlims: tuple
    ylims: tuple

    Returns
    -------
    xarray.Dataset
    """
    file = x.encoding["source"]

    with rasterio.open(file) as src:
        scale_factor = pd.to_numeric(src.tags().get("scale_factor"))
        time_value = pd.to_datetime(src.tags().get("time_begin"))

        x = x / scale_factor
        x = x.assign_coords(time=time_value).expand_dims("time")

    x = x.rio.clip_box(*bbox, crs="EPSG:4326")

    return x.rename({"band_data": "sig0"}).squeeze("band").drop_vars("band")


bbox = box(16.6, 47.7, 16.75, 47.8)
partial_func = partial(_preprocess, bbox=bbox.bounds)

sig0_ds = xr.open_mfdataset(
    data_path.glob("*.tif"),
    engine="rasterio",
    combine="nested",
    concat_dim="time",
    preprocess=partial_func,
).sortby("time")

sig0_da = sig0_ds.sig0
sig0_da
<xarray.DataArray 'sig0' (time: 8, y: 1221, x: 1230)> Size: 96MB
dask.array<getitem, shape=(8, 1221, 1230), dtype=float64, chunksize=(1, 1, 1230), chunktype=numpy.ndarray>
Coordinates:
  * x            (x) float64 10kB 5.282e+06 5.282e+06 ... 5.294e+06 5.294e+06
  * y            (y) float64 10kB 1.571e+06 1.571e+06 ... 1.559e+06 1.559e+06
  * time         (time) datetime64[ns] 64B 2023-08-05T16:51:22 ... 2023-10-28...
    spatial_ref  int64 8B 0

Let’s have a look at the data by plotting the first timeslice.

sig0_da.isel(time=0).plot(robust=True, cmap="Greys_r").axes.set_aspect("equal")

6.2 Load CORINE Landcover Data

We will load the CORINE Land Cover, which is a pan-European land cover and land use inventory with 44 thematic classes. The resolution of this classification is 100 by 100m and the file was created in 2018 (CORINE Land Cover).

Remember previous notebooks when comparing variables. Lets have a look at some of the metadata of our dataset.

corine_path = Path("~/shared/datasets/rs/corine").expanduser()


def print_raster(raster, name):
    """
    Print Raster Metadata

    Parameters
    ----------
    raster: xarray.DataArray|xarray.DataSet
        raster to process
    y: string
        name of product
    """

    print(
        f"{name} Raster: \n----------------\n"
        f"resolution: {raster.rio.resolution()} {raster.rio.crs.units_factor}\n"  # noqa
        f"bounds: {raster.rio.bounds()}\n"
        f"CRS: {raster.rio.crs}\n"
    )


bbox = affinity.scale(bbox, xfact=1.2, yfact=1.2)

cor_da = (
    xr.open_mfdataset(corine_path.glob("*.tif"), engine="rasterio")
    .rename({"band_data": "land_cover"})["land_cover"]
    .squeeze()
    .rio.clip_box(*bbox.bounds, crs="EPSG:4326")
)

cor_da
<xarray.DataArray 'land_cover' (y: 146, x: 147)> Size: 86kB
dask.array<getitem, shape=(146, 147), dtype=float32, chunksize=(81, 116), chunktype=numpy.ndarray>
Coordinates:
    band         int64 8B 1
  * x            (x) float64 1kB 4.814e+06 4.814e+06 ... 4.828e+06 4.828e+06
  * y            (y) float64 1kB 2.767e+06 2.767e+06 ... 2.753e+06 2.753e+06
    spatial_ref  int64 8B 0
Attributes:
    DataType:                Thematic
    AREA_OR_POINT:           Area
    RepresentationType:      THEMATIC
    STATISTICS_COVARIANCES:  136.429646247598
    STATISTICS_MAXIMUM:      48
    STATISTICS_MEAN:         25.753373398066
    STATISTICS_MINIMUM:      1
    STATISTICS_SKIPFACTORX:  1
    STATISTICS_SKIPFACTORY:  1
    STATISTICS_STDDEV:       11.680310194836
print_raster(sig0_da, "Sentinel-1")
Sentinel-1 Raster: 
----------------
resolution: (10.0, -10.0) ('metre', 1.0)
bounds: (5281990.0, 1558750.0, 5294290.0, 1570960.0)
CRS: EPSG:27704

print_raster(cor_da, "Corine Land Cover")
Corine Land Cover Raster: 
----------------
resolution: (100.0, -100.0) ('metre', 1.0)
bounds: (4813700.0, 2752700.0, 4828400.0, 2767300.0)
CRS: PROJCS["ETRS89-extended / LAEA Europe",GEOGCS["ETRS89",DATUM["European_Terrestrial_Reference_System_1989",SPHEROID["GRS 1980",6378137,298.257222101004,AUTHORITY["EPSG","7019"]],AUTHORITY["EPSG","6258"]],PRIMEM["Greenwich",0],UNIT["degree",0.0174532925199433,AUTHORITY["EPSG","9122"]],AUTHORITY["EPSG","4258"]],PROJECTION["Lambert_Azimuthal_Equal_Area"],PARAMETER["latitude_of_center",52],PARAMETER["longitude_of_center",10],PARAMETER["false_easting",4321000],PARAMETER["false_northing",3210000],UNIT["metre",1],AXIS["Easting",EAST],AXIS["Northing",NORTH],AUTHORITY["EPSG","3035"]]

The long name for CRS corresponds to the grid (EPSG:3035). We will again use rioxarray to reproject_match to convert the CORINE Land Cover map in the desired projection and resolution. The CORINE Land Cover is originally given as integer values. Hence we use the default nearest neighbour method for resampling.

cor_da = cor_da.rio.reproject_match(sig0_da)
cor_da = cor_da.assign_coords({"x": sig0_da.x, "y": sig0_da.y})
cor_da
<xarray.DataArray 'land_cover' (y: 1221, x: 1230)> Size: 6MB
array([[12., 12., 12., ..., 41., 41., 41.],
       [12., 12., 12., ..., 41., 41., 41.],
       [12., 12., 12., ..., 41., 41., 41.],
       ...,
       [ 2.,  2.,  2., ..., 41., 41., 41.],
       [ 2.,  2.,  2., ..., 41., 41., 41.],
       [ 2.,  2.,  2., ..., 41., 41., 41.]], dtype=float32)
Coordinates:
    band         int64 8B 1
    spatial_ref  int64 8B 0
  * x            (x) float64 10kB 5.282e+06 5.282e+06 ... 5.294e+06 5.294e+06
  * y            (y) float64 10kB 1.571e+06 1.571e+06 ... 1.559e+06 1.559e+06
Attributes:
    DataType:                Thematic
    AREA_OR_POINT:           Area
    RepresentationType:      THEMATIC
    STATISTICS_COVARIANCES:  136.429646247598
    STATISTICS_MAXIMUM:      48
    STATISTICS_MEAN:         25.753373398066
    STATISTICS_MINIMUM:      1
    STATISTICS_SKIPFACTORX:  1
    STATISTICS_SKIPFACTORY:  1
    STATISTICS_STDDEV:       11.680310194836

6.2.1 Colormapping and Encoding

For the different land cover types we use the official color encoding which can be found in CORINE Land Cover. The color mapping is stored in a json file under assets.

# Load encoding
with open("../assets/06_color_mapping.json", "r") as f:
    color_mapping_data = json.load(f)

# Get mapping
color_mapping = {item["value"]: item for item in
                 color_mapping_data["land_cover"]}

# Create cmap and norm for plotting
colors = [info["color"] for info in color_mapping.values()]
categories = [info["value"] for info in color_mapping.values()]
cmap = ListedColormap(colors)
norm = BoundaryNorm(categories + [max(categories) + 1], len(categories))

Now we can plot the CORINE Land Cover dataset.

# Get landcover codes present in the image
present_landcover_codes = np.unique(cor_da.values[~np.isnan(cor_da.values)].
                                    astype(int))

# Get colors + text for legend
handles = [
    mpatches.Patch(
        color=info["color"], label=(f'{info["value"]} - ' + (info["label"]))
        )
    for info in color_mapping.values()
    if info["value"] in present_landcover_codes
]

# Create the plot
cor_da.plot(figsize=(10, 10), cmap=cmap, norm=norm, add_colorbar=False).\
    axes.set_aspect("equal")

plt.legend(
    handles=handles,
    bbox_to_anchor=(1.01, 1),
    loc="upper left",
    borderaxespad=0,
    fontsize=7,
)
plt.title("CORINE Land Cover (EPSG:27704)")
Text(0.5, 1.0, 'CORINE Land Cover (EPSG:27704)')

Now we are ready to merge the backscatter data (sig0_da) with the land cover dataset (cor_da) to have one dataset combining all data.

var_ds = xr.merge([sig0_da, cor_da]).drop_vars("band").compute()
var_ds
<xarray.Dataset> Size: 102MB
Dimensions:      (x: 1230, y: 1221, time: 8)
Coordinates:
  * x            (x) float64 10kB 5.282e+06 5.282e+06 ... 5.294e+06 5.294e+06
  * y            (y) float64 10kB 1.571e+06 1.571e+06 ... 1.559e+06 1.559e+06
  * time         (time) datetime64[ns] 64B 2023-08-05T16:51:22 ... 2023-10-28...
    spatial_ref  int64 8B 0
Data variables:
    sig0         (time, y, x) float64 96MB -6.99 -7.32 -8.78 ... -14.32 -14.22
    land_cover   (y, x) float32 6MB 12.0 12.0 12.0 12.0 ... 41.0 41.0 41.0 41.0

6.3 Backscatter Variability

With this combined dataset we can study backscatter variability in relation to natural media. For example we can look at the backscatter variability for water by clipping the dataset to only contain the land cover class water, like so:

# 41 = encoded value for water bodies
waterbodies_mask = var_ds.land_cover == 41
waterbodies_mask.plot().axes.set_aspect("equal")

This gives use backscatter values over water only.

waterbodies_sig0 = var_ds.sig0.isel(time=0).where(waterbodies_mask)
waterbodies_sig0.plot(robust=True, cmap="Greys_r").axes.set_aspect("equal")

To get an idea of the variability we can create a histogram. Radar backscatter from water bodies fluctuates with surface roughness, which changes with wind conditions, creating spatial and temporal variations in signal intensity.

waterbodies_sig0.plot.hist(bins=50, edgecolor="black")
(array([5.0000e+00, 5.0000e+00, 8.0000e+00, 1.4000e+01, 3.5000e+01,
        7.8000e+01, 1.5600e+02, 3.0500e+02, 5.4000e+02, 1.0380e+03,
        1.9550e+03, 3.6850e+03, 6.9910e+03, 1.2269e+04, 1.9088e+04,
        2.6859e+04, 3.6217e+04, 4.5276e+04, 5.0939e+04, 4.9020e+04,
        3.9729e+04, 2.5400e+04, 1.2426e+04, 5.4390e+03, 2.6990e+03,
        2.0500e+03, 2.1330e+03, 2.6890e+03, 3.1490e+03, 3.9510e+03,
        4.2970e+03, 4.1110e+03, 3.6350e+03, 2.5900e+03, 1.6020e+03,
        7.5000e+02, 3.5200e+02, 1.2300e+02, 5.6000e+01, 3.7000e+01,
        2.8000e+01, 2.1000e+01, 1.3000e+01, 9.0000e+00, 1.0000e+01,
        8.0000e+00, 1.0000e+01, 5.0000e+00, 6.0000e+00, 4.0000e+00]),
 array([-37.91  , -36.9202, -35.9304, -34.9406, -33.9508, -32.961 ,
        -31.9712, -30.9814, -29.9916, -29.0018, -28.012 , -27.0222,
        -26.0324, -25.0426, -24.0528, -23.063 , -22.0732, -21.0834,
        -20.0936, -19.1038, -18.114 , -17.1242, -16.1344, -15.1446,
        -14.1548, -13.165 , -12.1752, -11.1854, -10.1956,  -9.2058,
         -8.216 ,  -7.2262,  -6.2364,  -5.2466,  -4.2568,  -3.267 ,
         -2.2772,  -1.2874,  -0.2976,   0.6922,   1.682 ,   2.6718,
          3.6616,   4.6514,   5.6412,   6.631 ,   7.6208,   8.6106,
          9.6004,  10.5902,  11.58  ]),
 <BarContainer object of 50 artists>)

6.4 Variability over Time

Next we will look at the changes in variability in backscatter values over time for each of the CORINE Land Cover types. We do this by creating the following interactive plot. We can spot that backscatter in agricultural fields varies due to seasonal cycles like planting, growing, and harvesting, each of which changes vegetation structure. Changes in backscatter are strongly related to soil moisture content from irrigation or rainfall. Ultimately, phenological stages of crops and canopy moisture dynamics can affect the backscatter signal.

hv.extension("bokeh")

robust_min = var_ds.sig0.quantile(0.02).item()
robust_max = var_ds.sig0.quantile(0.98).item()

bin_edges = [
    i + j * 0.5
    for i in range(int(robust_min) - 2, int(robust_max) + 2)
    for j in range(2)
]

land_cover = {"\xa0\xa0\xa0 Complete Land Cover": 1}
land_cover.update(
    {
        f"{int(value): 02} {color_mapping[value]['label']}": int(value)
        for value in present_landcover_codes
    }
)
time = var_ds.sig0["time"].values

rangexy = RangeXY()


def load_image(time, land_cover, x_range, y_range):
    """
    Callback Function Landcover.

    Parameters
    ----------
    time: panda.datatime
        time slice
    landcover: int
        land cover type
    x_range: array_like
        longitude range
    y_range: array_like
        latitude range

    Returns
    -------
    holoviews.Image
    """

    if land_cover == "\xa0\xa0\xa0 Complete Land Cover":
        sig0_selected_ds = var_ds.sig0.sel(time=time)

    else:
        land_cover_value = int(land_cover.split()[0])
        mask_ds = var_ds.land_cover == land_cover_value
        sig0_selected_ds = var_ds.sig0.sel(time=time).where(mask_ds)

    hv_ds = hv.Dataset(sig0_selected_ds)
    img = hv_ds.to(hv.Image, ["x", "y"])

    if x_range and y_range:
        img = img.select(x=x_range, y=y_range)

    return hv.Image(img)


dmap = (
    hv.DynamicMap(load_image, kdims=["Time", "Landcover"], streams=[rangexy])
    .redim.values(Time=time, Landcover=land_cover)
    .hist(normed=True, bins=bin_edges)
)

image_opts = hv.opts.Image(
    cmap="Greys_r",
    colorbar=True,
    tools=["hover"],
    clim=(robust_min, robust_max),
    aspect="equal",
    framewise=False,
    frame_height=500,
    frame_width=500,
)

hist_opts = hv.opts.Histogram(width=350, height=555)

dmap.opts(image_opts, hist_opts)