Land Mapper Data Pipeline

The purpose of the Land Mapper Data Pipeline is to streamline the workflow for fetching, processing, and updating geospatial data layers used in Land Mapper's mapping platform. It consists of Jupyter notebooks and python scripts for fetching and processing raster and vector data from various public repositories, as well as the code to refresh the map layers with new data. The pipeline integrates code to update the following layers available in Land Mapper: Forest Community Type, Tree Size Class, Canopy Cover, Forest Stocking, Soil Type, and Taxlots.

The repository is organized as follows

---

├── LICENSE
├── README.md          <- The top-level README. You are here.
|
├── docs               <- A default Sphinx project; see sphinx-doc.org for details.
│
├── models             <- Trained and serialized models, model predictions, or model summaries.
│
├── notebooks          <- Jupyter notebooks for data pipelines.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis in various formats.
│
├── environment.yml    <- Python dependencies for reproducing the environment.
│
├── setup.py           <- Makes project pip installable so src can be imported.
|
└── src                <- Source code for data pipelines.
    |
    ├── datafactory    <- Scripts to download and process data.
    ├── composition    <- Scripts to generate the Forest Community Types layer.
    └── structure      <- Scripts to generate the Forest Structure layers.

---

Installation

git clone https://github.com/Ecotrust/landmapper_data_pipeline.git

cd landmapper_data_pipeline
mamba env create -f environment.yml
mamba activate landmapper-dpl

Run the following commands to setup the environment variables in a .env file in the src directory:

echo DATADIR=/path/to/data > .env
echo GRID=/path/to/usgs_grid >> .env

DATADIR is the path to the directory containing the input data and the destination for predictions. The following directories and datasets are expected:

DATADIR
|── tiles
|   ├── climatena       <-- PRISM 1971-2000 gridded monthly climate data
|   ├── 3dep            <-- Elevation data from the USGS 3D Elevation Program
|   ├── sentinel2sr     <-- Sentinel-2 surface reflectance imagery with 10-bands
|   ├── landtrendr      <-- LandTrendr forest change data
|   └── dynamic_world   <-- Forest/non-forest mask from Dynamic World
└── predictions         <-- Output directory for model predictions

GRID is the path to the USGS quarter quads grid geojson file.

---

Usage

Download new input data

datafactory/download_dem.py - Fetchs DEM from the USGS 3D Elevation Program (3DEP) and calculates the following layers: Slope, Aspect, Flow Accumulation, Topographic Position Index (TPI300 and TPI2000), Slope Position Class (SPC300), and Land Form Class. All layers are saved in a single GeoTIFF file with 8 bands.

python download_dem.py

datafactory/download_s2.py - Fetches Sentinel-2 surface reflectance imagery from Google Earth Engine. The bands fetched are: B2, B3, B4, B5, B6, B7, B8, B8A, B11, and B12.

python download_s2.py --year <year>

datafactory/download_dw.py - Fetches a forest/non-forest mask from Dynamic World land cover classification available in Google Earth Engine.

python download_dw.py --year <year>

datafactory/download_ltr.py - Fetches land cover change data genered with the LandTrendr algorithm available in Google Earth Engine.

python download_ltr.py --year <year>

Note: We will add instruction to download and process PRISM climate data soon.

Refreshing the Forest Community Types layer

To generate the Forest Community Types map we use a pre-trained UNet model. The model was trained on forest composition data produced by the LEMMA research group at Oregon State University for Oregon and Washington State. The input data for the model consists of a stack of Sentinel-2 surface reflectance imagery, an 8-band raster with a DEM and 7 topographic metrics, and an 8-band raster with climate variables from PRISM.

The model predicts the following community types:

Class	Description
1	Shrub
2	Riparian
3	Lodgepole pine
4	Ponderosa Pine
5	Mixed Conifer
6	Western Juniper
7	Mixed Oak - Conifer
8	Quaking Aspen
9	Mixed Hardwood - Conifer
10	Coastal Spruce Cedar or Redwood
11	Douglas-fir - Western Hemlock
12	Silver fir - Mountain Hemlock
13	Spruce - Subalpine Fir

Run the script composition/predict.py

python predict.py

Refreshing forest structure map layers

Three forest structure map layers, Tree Size Class, Canopy Cover, and Forest Stocking are generated using a pre-trained Gradient Boosting regression model. The model was trained on forest attributes data produced by the LEMMA research group at Oregon State University. The input data for the model consists of a stack of Sentinel-2 imagery, a DEM, and LandTrendr data.

Run script structure/predict_structure.py

python predict_structure.py

Refreshing the Soil Type layer

Instructions coming soon.

---

Refreshing the Taxlots layer

Instructions coming soon.

---

Project based on the cookiecutter data science project template. #cookiecutterdatascience