Lakes temperature

This repository contains the data, code, and results for “Evaluation of Methods for Estimating Lake Surface Water Temperature Using Landsat 8” article.

Dataset

The data folder contains the following files: - SR_processed.csv - surface reflectance after cleaning - TOA_processed.csv - top-of-atmosphere reflectance after cleaning - hydro_stations.csv - list of hydrological stations (38) with name and ID - lakes_temp.csv - lake water temperature in degrees Celsius - pointsFeatures.txt - location of measurement points as a JavaScript object (this is required by Google Earth Engine) - SMW_LST_L8_Lakes_newEmiss.csv - estimated water temperatures using the Ermida et al. (2020) model - coordinates subfolder - location of measurement points as a shapefile - reflectance subfolder - raw (not cleaned) SR and TOA reflectance - vector/lakes.gpkg - extent of 4 sample lakes (Drawsko, Ełckie, Gopło, Łebsko)

Reproduction

1. Open the geomorph_clustering.Rproj project file in RStudio.
2. Create a JavaScript object with coordinates using 01_create_features.R that will be used in Google Earth Engine.
3. Download reflectance data from Google Earth Engine using 02_Landsat8_SR_download.js (Surface Reflectance) and 02_Landsat8_TOA_download.js (Top-of-Atmosphere Reflectance). You must use the coordinates from the pointsFeatures.txt file.
4. Download data from hydrological stations (water temperature) using 04_hydro_process.R.
5. The main part of the analysis was done in the 05_analysis.R script. It includes training of LM and RF models and validation of all LM, RF, LST and LST-L2 models.
6. 06_LST_calibration.R was used to compare calibration methods for the LST-L2 (USGS) product using empirical data.
7. Entire satellite scenes for spatial prediction can be downloaded using script 07_download_scene.js.
8. Prediction using LM or RF model can be done with script 08_predict.R for individual lakes or the entire scene. The {terra} package was used to process the raster data.

The algorithm to generate the LST product developed by Ermida et al. (2020) is available in the Google Earth Engine repository: https://code.earthengine.google.com/?acceptrepo=users/sofiaermida/landsatsmw_lst

Results

The results of this research are saved in results folder: - lakes_stats.csv- performance statistics of LM and RF models considering training and test lakes - month_stats.csv - performance statistics of LM and RF models by month - predictions_testset.csv - testset with actual measurements and estimated by 4 models (LM, RF, LST, LST-L2) - rf_model.rds - trained RF model in .rds format ({ranger} package is required)

Additionally, in the images/predict folder there are 4 exemplary results of the spatial prediction by the RF model for different terms.

Acknowledgement

The source of the hydrological data is the Institute of Meteorology and Water Management - National Research Institute (https://www.imgw.pl/). Landsat-8 images courtesy of the U.S. Geological Survey (https://earthexplorer.usgs.gov/) and Google Earth Engine (https://earthengine.google.com/).