/lfmc_from_sar

Map live fuel moisture using remote sensing and deep learning

Primary LanguagePython

SAR-Enhanced Mapping of Live Fuel Moisture Content

Seasonal variation of live fuel moisture content estimated by our deep learning model.

Seasonal variation of live fuel moisture content estimated by our deep learning model.

This repository contains analysis performed for the paper SAR-Enhanced Mapping of Live Fuel Moisture Content in the Journal Remote Sensing of Environment by Rao et al., 2020. You can view the live fuel moisture content (LFMC) maps produced by the deep learning algorithm in this study in a web-app here.

Earth Engine Web-app user guide

The web-app allows users to interactively explore the LFMC maps produced in the paper. The slider bar at bottom controls the time. The blue point on the map controls the location for which LFMC time series is produced from 2016 - 2019.

Interested in creating your own web-app similar to the Live Fuel Moisture Viewer? Here is the source code for the web-app.

FAQs

  1. The web-app seems broken? Allow few seconds for the web-app to load. Make sure you are using Google Chrome, Microsoft Edge and Mozilla Firefox. It does not work on Microsoft Internet Explorer. If the web-app still doesn't work, raise an issue in this repository.
  2. Can I access more recent LFMC maps? As of July 2020, maps from Jan 2016 to Jul, 2020 are available. Moving forward, the project team plans to update the maps directly in the web-app. A fixed update frequency (or a fixed latency) cannot be guaranteed at the moment. For requests related to updating maps, please contact the corresponding author of the manuscript. Do not raise a Github "issue" for this purpose.
  3. Why are there dark green or dark brown patches on some days? On some occassions, the LFMC maps may appear patchy. The patches are caused by incorrect cloud or snow masking. The algorithm relies on the in-built quality assessment flags in the Landsat-8 product to mask "snow", "cloud", or "cloud shadow". For more information on how these quality assessment flags were developed refer to Vermote et al., 2016.

Download LFMC maps

The LFMC maps are hosted on Google Earth Engine (GEE) which is a free platform for largescale image visualization and analysis. The maps can be found in an ee.ImageCollection() object as a public asset at the following link: https://code.earthengine.google.com/?asset=users/kkraoj/lfm-mapper/lfmc_col_25_may_2021. You can use the maps in the following ways-

  1. Directly on GEE by importing the collection or
  2. Downloading the maps to your local computer

Both options need a GEE account signup here. It is free.

Use/analyse maps on GEE

  1. Once you have your GEE account, open this script
  2. The script will import the LFMC maps as an ImageCollection and display the mean for 2019. You can then proceed with your analysis with the imported image collection.

Download maps to your computer

Option 1: Code Editor-

  1. Once you have your GEE account, open this script
  2. Modify the start_date and end_date to suit your needs
  3. Modify scale to set pixel resolution of output maps. The native resolution of the maps are 250m but you can rescale to whatever resolution you want to suit your analysis.
  4. Click on Run button at the top
  5. Click on the Tasks panel on the top right. Verify the maps that you need are set in staging. If ok, click on the Run button beside each map. The maps will be downloaded to your Google Drive in a folder called "lfmc_folder".

Option 2: Python API-

If you want to download many maps, consider using GEE's python API. It will let you download the maps without having to click the Run button for each map. In the link referred, follow the download instructions.

  1. Once you have the python API installed, open this script.
  2. Modify the start_date and end_date to suit your needs
  3. Modify scale to set pixel resolution of output maps. The native resolution of the maps are 250m but you can rescale to whatever resolution you want to suit your analysis.
  4. Run the script. The maps will be downloaded to your Google Drive in a folder called "lfmc_folder".

Option 3: Manual-

250m resolution GeoTiff maps from April 2016 to Oct 2020 have been archived at Google Drive.

Repository details

The rest of this Readme file pertains to reproducing the analysis and sharing the algorithms associated with the paper.

Scripts:

The repository consists of scripts in the "scripts" folder to perform the following-

  1. Prepare input data and train an long-short term memory model to predict LFMC in LSTM.py
  2. Make plots from the manuscript using plot_functions.py
  3. Make LFMC maps using make_map_features_and_predict.py

Rest of the scripts are not needed. They were used for development of the model and preliminary investigation only.

Training data

The training data along with the labels can be found on Radiant MLHub. The documentation corresponding to the training data and meaning of the column names can be found here.

Training labels were scraped from the National Fuel Moisture Database hosted by the United State Fores Service. We are grateful to them to make this data public.

Trained model

The training model saved using best model checkpoint on keras can be found in trained_model folder.

Prerequisites

  1. Python 3.6
  2. keras v. 2.2.2

Reproducibility guide

  1. Clone the repository using git clone https://github.com/kkraoj/lfmc_from_sar.git
  2. Change the directory addresses of dir_data and dir_codes in dirs.py
  3. Run plot_functions.py by uncommenting any of the functions at the end of the script to reproduce the figures you wish

License

Data and scripts presented here are free to use under CC BY-NC-ND 4.0 license. Please cite the following paper if you use any data or analyses from this study:

Rao, K., Williams, A.P., Fortin, J. & Konings, A.G. (2020). SAR-enhanced mapping of live fuel moisture content. Remote Sens. Environ., 245.

Issues?

Check the Issues tab for troubleshooting or create a new issue.