Geographical Features (HPC)

Geographical Features (HPC)#

import pvdeg
import numpy as np

Create Scenario and Populate Location#

The default geospatial datapoints for addLocation are from the Americas satellite. This is the only data we are interested in for this case so don’t specify any location arguments. To speed up calculations we will downsample to get coarser data. The downsampling function is not linear. A downsample factor of 10 takes us from 2018267 entries to only 5109.

features = pvdeg.GeospatialScenario(name="finding-features")

features.addLocation(downsample_factor=10)

Intro to KDTrees#

A K dimensional tree is a datastructure for organizing n points in a k dimensional space. They are often used for nearest neighbors searches and many ML algorithms. Letting k = 2, our dimensions will be latitude and longitude. This is much faster than iterating over tabular data structures to find neighbors. The function to create a kdtree below requires scikit learn (also known as sklearn), a python machine learning library which is not included in the pvdeg dependency list so you will have to install it independently. Scipy also has a kdtree class but it is much slower. Depending on the number of datapoints the tree build time could be quite long but for our purposes, the cell below will be quick.

tree = pvdeg.geospatial.meta_KDtree(meta_df=features.meta_data)

Remaining Points After Downsampling#

Use plot_coords to see all latitude longitude coordinate pairs included in a scenario’s metadata. You can provide corners of a bounding box to change the extent of the matplotlib plot.

features.plot_coords()

Identifying Mountains from Geospatial Data I#

Identify moutains by using the function below. There are currently 2 methods to identify mountains from our data. Both will add a column to our GeospatialScenario.meta_data attribute called mountain containing boolean values that represent if a data point is in/near mountains or not near mountains.

  • classify_mountains_weights uses k nearest neighbors search and calculates local changes in height with different methods, the default is the mean absolute change compared to a points neighbors, then it assigns a weight to each point. If the remaining points are above the threshold then the percentile argument determines sensitivity to classifying points as mountains (higher weight, more mountainous). Much better than the following option.

  • classify_mountains_radii uses 2 nearest neighbors radius searches at each point and compares the average elevation of the inner circle to the avg radius of the circle.

features.classify_mountains_weights(kdtree=tree)

Identifing Mountains from Geospatial Data II#

We do not need to provide a kdtree when we call this method. We can provide a kdtree if we create one first with pvdeg.geospatial.meta_kdtree but it will work without supplying it. GeospatialScenario will create a kdtree internally for us. So the method can simply be called as shown below.

# this will work without running the above cells to create a kdtree or the cell directly above where we utilize the created kdtree.
features.classify_mountains_weights()

Convenience Plotting for Binary Classifications#

the plot_meta_classification method allows you to plot your data with respect to a boolean classifier present in the scenario.meta_data dataframe columns.

features.plot_meta_classification(col_name="mountain")

Identifying mountains within a bounding box.#

Sometimes we only want information from one area. To allow for more control over the selection process. To pick only one area from an existing superset of data we can use a bounding box. Either provide the top left and bottom right (latitude-longitude) pairs or a 2d numpy array of these pairs which will use the most extreme entries to form the bounding box.

features.classify_mountains_radii(
    kdtree=tree,
    elevation_floor=0,
    bbox_kwarg={
        "coords": np.array(
            [
                [47.960502, -115.048828],
                [47.842658, -101.118164],
                [36.738884, -113.686523],
                [36.633162, -100.283203],
            ]
        )
    },
)

features.plot_meta_classification(col_name="mountain")

Identifying Coastlines from Geospatial Data#

To identify points within a search radius of any natural earth features. The following examples show the process of identifying coastline datapoints.

features.classify_feature(
    kdtree=tree,
    feature_name="coastline",
    resolution="10m",
    radius=0.5,  # change this to get a wider band of selected data near the coastline
)

features.plot_meta_classification(col_name="coastline")

Finding Rivers#

Above we saw how to find points near coastlines, now lets do points near rivers. It is very similar

features.classify_feature(
    kdtree=tree, feature_name="rivers_lake_centerlines", resolution="10m", radius=0.2
)

features.plot_meta_classification(col_name="rivers_lake_centerlines")
Contents