Okavango AdDSWE

Code accompanying the manuscript "A 41-year satellite record reveals abrupt redistribution of the Okavango Delta's flood pulse since 2012."

This repository provides the Google Earth Engine and Python code used to (1) generate the Adapted Dynamic Surface Water Extent (AdDSWE) inundation record for the Okavango Delta, Botswana, and (2) reproduce the analyses and figures in the manuscript.

AdDSWE is a 41-year (June 1984 – December 2025), 30-m, monthly, multiclass surface-water record derived from the Landsat Collection 2 archive. It augments the standard Dynamic Surface Water Extent (DSWE) confidence framework (Jones 2019) with a dynamic shortwave-infrared (SWIR2) threshold test ("Test 6") that detects water beneath vegetation, classifying every pixel into five confidence classes:

Class	Meaning
0	Not water / dry land
1	Low-confidence inundation (vegetated / partial)
2	Partial / water under canopy
3	Moderate-confidence open water
4	High-confidence open water

Open water = classes 3–4; total inundation = classes 1–4.

Data availability

The AdDSWE products and all derived datasets are archived on Dryad:

• Dryad: https://doi.org/10.5061/dryad.msbcc2gct
• Interactive viewer: https://okavango-water-mask-viewer.streamlit.app/

Note: the Dryad dataset is currently under review and not yet publicly released, so the DOI above will not resolve until the dataset is published (typically upon publication of the associated article). The identifier itself is final and will not change.

The analysis/ notebooks are designed to run against the Dryad file layout — download the archive and point each notebook's input paths at your local copy. See the Dryad README for file-level descriptions, variable definitions, units, and coordinate reference systems.

Repository layout

addswe/          Core AdDSWE algorithm (importable package)
  classification.py   spectral indices, 5 DSWE tests, dynamic SWIR2 Test 6, decision tree, morphological filter
  composites.py       Landsat monthly compositing + iterative temporal gap-filling
generation/      GEE notebooks that produce the inundation and ancillary products
analysis/        Notebooks that reproduce the manuscript figures/tables from the archived data
environment.yml  Conda environment specification

Script → figure / table map

Notebook	Manuscript output
generation/landsat_addswe_monthly_30m.ipynb	AdDSWE + original DSWE products (inputs to Figs. 2, 4, 7)
generation/chirps_precipitation.ipynb	Precipitation series (Fig. 7, Supp. Fig. 4)
generation/pekel_watersurface.ipynb	JRC/Pekel comparison series (Supp. Fig. 4a)
analysis/monthly_record_fig2.ipynb	Figure 2 + LS_AdDSWE_monthly_summary
analysis/distributary_total_inundation_fig3.ipynb	Figure 3, Table 1 (total inundation)
analysis/distributary_open_water_suppfig1.ipynb	Supplementary Fig. 1, Table 1 (open water)
analysis/mean_difference_map_fig4.ipynb	Figure 4 (7-band difference raster)
analysis/centroid_migration_fig5.ipynb	Figure 5 + centroid_weighted_all.csv
analysis/open_water_validation_supp.ipynb	Fig. 7a–d, Supp. Fig. 5, Supp. Table 1
analysis/qc_mask_stats_suppnote2.ipynb	Supplementary Note 2 (QC statistics)

Reproducibility & requirements

The two stages differ in how they can be re-run:

• Generation (generation/, and the product-generation sections of the Fig. 4 and Fig. 5 notebooks) requires a Google Earth Engine account and project, authentication, and long-running export tasks. These notebooks document exactly how the products were produced; they are not turnkey.
• Analysis (analysis/) reproduces the manuscript figures and tables from the archived Dryad data and is runnable by anyone once the archive is downloaded.

The core AdDSWE algorithm is in the addswe package so it can be read, cited, and reused independently of the notebooks. Call ee.Initialize() before using the Earth Engine functions.

Installation

conda env create -f environment.yml
conda activate okavango
pip install -e .

The editable install makes addswe importable from anywhere (including the generation/ notebooks), so from addswe import ... works without adjusting sys.path.

Then, for Earth Engine functionality, authenticate once with earthengine authenticate.

Note on Study_Compare_MaxInundation.csv

The cross-product comparison table (Dryad inundation/Study_Compare_MaxInundation.csv; used in Fig. 7 and Supp. Fig. 4) is compiled from multiple sources rather than produced by a single script. Its columns are drawn from:

Column	Source
AdDSWE Total Inundation (JASO / MJJASO)	this study (generation/ + analysis/)
AdDSWE Open Water (MJJASO)	this study
Landsat-SWIR threshold (JASO)	Inman & Lyons (2020)
MODIS	Thito et al. (2016)
NOAA AVHRR	Gumbricht et al. (2004); McCarthy et al. (2003)
MODIS-SWIR threshold	Wolski et al. (2017); Hofmann et al. (2024)
JRC Global Surface Water	Pekel et al. (2016) — via generation/pekel_watersurface.ipynb
Total Precipitation (mm)	CHIRPS — via generation/chirps_precipitation.ipynb
Annual Peak Discharge at Mohembo	Global Runoff Data Centre (GRDC)

Citation

If you use this code or the AdDSWE products, please cite:

• Manuscript: Rees, H., Crompton, O., Lourenço, M., WinklerPrins, L., Larsen, L., Caylor, K., Ganti, V. A 41-year satellite record reveals abrupt redistribution of the Okavango Delta's flood pulse since 2012. (in preparation; DOI to be added upon publication).
• Data: Dryad, https://doi.org/10.5061/dryad.msbcc2gct
• Code: this repository (archived release DOI to be added via Zenodo).

License

Code is released under the MIT License (see LICENSE). The archived data on Dryad are released under CC0 1.0.