Speaker
Description
Wetlands are multi-function ecosystems which store surface water and regulate water flows, they drive evatransporation and precipitation, and purify and replenish groundwater aquifers. Wetlands moreover constitute biodiversity hotspots for terrestrial, avian and aquatic life, and consequently for the livelihood of people. – Where there are wetlands, there is life.
Wetlands are characterised by their inundation dynamics, where inundation maximum and minimum extent and duration are key parameters that define their composition. Information on forested wetlands, where inundation occurs partly or fully under a closed forest canopy is however scarce and recognised as an information gap by the Ramsar Wetlands Convention. In the Amazon Basin, forest wetlands constitute around 15% of the basin.
JAXA’s L-band SAR missions (JERS-1, ALOS, ALOS-2) have a long track record of mapping and monitoring wetland inundation on a large scale, with the long L-band wavelength being particularly effective due to its ability to operate under cloud cover and penetrate vegetation canopies. Under JAXA’s systematic observation strategy, PALSAR-2 ScanSAR data have been acquired over the pan-tropical belt on a regular (6-week) basis from launch in 2014 until present.
In a collaboration between JAXA, soloEO, Aberystwyth University and Wetlands International, we have used the complete archive of PALSAR-2 ScanSAR data over the northern part of South America, including the Amazon and Orinoco basins, and Central Africa, to create time-series maps of inundation extent on a 6-week basis from late 2014 to present (September 2026) at 50 m pixel spacing.
The ALOS-2 data used have been processed by JAXA to radiometrically terrain corrected (RTC) datasets according to CEOS Analysis Ready Data Normalised Radar Backscatter (CEOS-ARD NRB) format standards. The workflow was implemented on the CEOS Analytics Lab cloud server and includes the following steps:
- Reprojection and speckle filtering
- Residual antenna pattern calibration
- Corrections for incidence angle dependency and scene bias
- Decision tree image classification
- Mosaicking
- Gap-filling and temporal filtering
- Generation of inundation duration maps
- Uncertainty assessment
For each of northern South America and Central Africa, 9 inundation extent maps have been generated for every year, resulting in some 200 maps for the period 2015-2025 for the two regions. Each map consists of two bands – band 1 is the wetland map (6 classes: inundated vegetation, seasonally submerged vegetation, open water, mud banks, other land cover, and masked areas). Band 2 is a metadata layer which shows the classification provenance, such as whether a pixel was derived from one or two SAR observations, through gap-filling or temporal filtering.
Additionally, annual maps of inundation duration, which indicate how many times a pixel has been classified as Inundated Vegetation in the time-series for each year were generated. With nine wall-to-wall PALSAR-2 coverages per year, pixel values range between 0-9, where DN=0 thus indicates classified as non-flooded and DN=9 indicates classified as flooded during all acquisitions during the year in question. 11 inundation duration maps were generated for each of northern South America and Central Africa.
The data are being made available under a CC BY 4.0 license from JAXA EORC and on Zenodo.