WetSAT-ML (Wetlands flooding extent and trends using SATellite data and Machine Learning) is an open-source R package developed by SEI Latin America. It enables monitoring of wetland flooding dynamics using Sentinel-1 radar imagery combined with machine learning algorithms.
The tool integrates with Google Earth Engine and allows users to:
- Generate wetland flooding extent maps.
- Produce water permanence maps.
- Extract flooded area time series.
- Quantify intra-annual and inter-annual wetland hydrological trends.
Figure 1. WetSAT-ML methodology workflow for generating wetland flooding extent and trends using Sentinel-1 data and machine learning.
WetSAT-ML uses Sentinel-1 Synthetic Aperture Radar (SAR) data to map water extent, overcoming the limitations of optical data, which often fail in cloudy or dense vegetation conditions.
The algorithm combines radar backscatter from VV and VH polarizations with five radar-derived indices:
| Index | Formula |
|---|---|
| PR - Polarized Ratio | σVH0 / σVV0 |
| NDPI - Normalized Difference Polarized Index | (σVV0 − σVH0) / (σVV0 + σVH0) |
| NVHI - Normalized VH Index | σVH0 / (σVV0 + σVH0) |
| NVVI - Normalized VV Index | σVV0 / (σVV0 + σVH0) |
| RVI - Radar Vegetation Index | 4 · σVH0 / (σVV0 + σVH0) |
These indices characterize the scattering behavior of radar signals under different wetland flooding conditions, enabling pixel-level water detection.
The WetSAT-ML package contains four main functions. Each function has practical examples of usage within the documentation:
| Function | Description |
|---|---|
radar_index_stack |
Calculates radar-derived indices (PR, NDPI, NVHI, NVVI, RVI) from Sentinel-1 VV and VH backscatter and extracts median index values within a buffer around reference stations. |
train_rf_model |
Trains a Random Forest classifier using radar-derived indices to detect water presence. Returns the trained model, overall accuracy, and variable importance. |
classify_water_surface |
Applies the trained Random Forest model to classify water and non-water pixels at the image level. Produces wetland flooding maps and water permanence layers. |
performWS |
Generates time series of flooded areas, intra-annual and inter-annual flooding trends, and hydroperiod statistics. |
You can install the development version of WetSAT-ML from GitHub:
# install.packages("devtools")
# devtools::install_github("dazamora/WETSAT")The Everglades region is located in southern Florida, and it extends over an area of 9,150 km2 from the margin of Florida Bay in the south to the Everglades Agricultural Area (EAA) in the north (Figure 1). The area supports a diverse mosaic of different wetlands, including freshwater marshes, swamps, sloughs, and wet prairies (Figure 1a). The area also presents diverse vegetation communities where the sawgrass (especially Cladium jamaicense) is the most abundant, interspersed with patches of shrubs with a mix of swamp and bayhead shrub species, and trees with a mix of swamp, hammock, and bayhead tree species Palomino-Ángel, Wdowinski, and Li (2024).
This is a basic example which shows you how to solve a common problem:
## basic example codeWhat is special about using README.Rmd instead of just README.md?
You can include R chunks like so:
summary(cars)
#> speed dist
#> Min. : 4.0 Min. : 2.00
#> 1st Qu.:12.0 1st Qu.: 26.00
#> Median :15.0 Median : 36.00
#> Mean :15.4 Mean : 42.98
#> 3rd Qu.:19.0 3rd Qu.: 56.00
#> Max. :25.0 Max. :120.00You’ll still need to render README.Rmd regularly, to keep README.md
up-to-date. devtools::build_readme() is handy for this.
You can also embed plots, for example:
Palomino-Ángel, Sebastián, Shimon Wdowinski, and Shanshan Li. 2024. “Wetlands Water Level Measurements from the New Generation of Satellite Laser Altimeters: Systematic Spatial-Temporal Evaluation of ICESat-2 and GEDI Missions over the South Florida Everglades.” Water Resources Research 60 (3): e2023WR035422. https://doi.org/https://doi.org/10.1029/2023WR035422.