Example notebook(s) using DEM and related feature engineering

ContinuumIO / elm

Phase I & part of Phase II of NASA SBIR - Parallel Machine Learning on Satellite Data

http://ensemble-learning-models.readthedocs.io

44 stars 23 forks source link

Example notebook(s) using DEM and related feature engineering #186

Open PeterDSteinberg opened 7 years ago

PeterDSteinberg commented 7 years ago

One use case for machine learning in river / estuary 2D or 3D flow and/or sediment transport models is automated derivation of mesh density specifications for inputs into mesh generators. The mesh generators take polygons that specify higher density at changes in slope, e.g. road prisms or levees, and lower density in simple / flat topography.

TODO:

[ ] Get X, Y, Z lidar points data (Use the data from this datashader example - Puget Sound Lidar Consortium)
- [ ] Experiment with informative features from differencing elevation:
- [ ] from scipy.ndimage.filters import laplace
- [ ] np.gradient
- [ ] Build custom estimator / transformer using pywavelet

gpfreitas commented 7 years ago

Possibly useful reference: https://pubs.er.usgs.gov/publication/sir20165141

Also, the data used by the Puget Sound Lidar datashader example mentioned by Peter is here: https://github.com/bokeh/datashader/blob/master/examples/datasets.yml#L52

PeterDSteinberg commented 7 years ago

Use the xarray_filters/notebooks as a basis.

It would be interesting to combine LANDSAT with lidar or DEM elevation feature engineering. See LANDSAT notebooks in datashader and the S3 LANDSAT downloader in earthio

PeterDSteinberg commented 7 years ago

The laplace filter mentioned above as an example relates to the Phase II milestone on zonal statistics and spatial filters.

See numba/examples/kernel_density_estimation.py and other examples in that examples directory for ideas on numba spatial filters. Also, xarray_filters.polygon_tools is a point-in-polygon function from Phase based on numba .

PeterDSteinberg commented 7 years ago

I changed the milestone of this issue for the Zonal Statistics, Filters, and Change Detection (from Promotion / Validation). Making the notebooks will help identify tests / use cases for the zonal stats milestone. See also #188 on how notebooks relate to documentation / auto-testing of code examples.

PeterDSteinberg commented 7 years ago

Another use case for modeling with such DEM/Lidar/LANDSAT input data is natural disaster mapping e.g. fire/flood - https://github.com/ContinuumIO/elm/issues/190.