Flood Extent Mapping Using Landsat Multispectral Water Indices in Ehlanzeni District Municipality, Mpumalanga Province, South Africa.

In Mpumalanga Province, South Africa, floods are one of the most devastating natural hazards. The January 2026 event, which was declared a national state of disaster on 17 January 2026, was responsible for the loss of close to R4 billion in Ehlanzeni District Municipality; thus, there is a need for systematic, spatially explicit flood inundation information. In this study, the ability to use freely available multispectral imagery to map flood extent for Ehlanzeni is explored, specifically for the flood event of January 2026. Sentinel-2 Multi-Spectral Instrument (MSI) Level-2A cloud cover <15 % data was acquired for periods when the climate zone was cloudy on 14 January 2026 (pre-flood) and 21 January 2026 (post-flood) at a 10 m spatial resolution.

In Google Earth Engine, three spectral water indices, the Normalised Difference Water Index (NDWI), Modified Normalised Difference Water Index (MNDWI) and Automated Water Extraction Index (AWEI), were computed and compared. The new water-inundated areas were identified by calculating the difference between the post-flood indices and a pre-flood baseline, and by masking with the JRC Global Surface Water v1.4 permanent-water layer with ≥80 % occurrence. 

Preliminary results showed that the indices including the short-wave infrared band (SWIR) provided better discrimination of flooded areas, with an overall accuracy of 82 % and a Kappa of 0.73. But MNDWI incorrectly classified 12 % of urban built-up pixels as water. By combining MNDWI, NDWI, and the Normalised Difference Built-up Index (NDBI), the accuracy of the urban classification was reduced to 4 % urban misclassification, achieving an overall accuracy of 89 % (Kappa = 0.84).

The results were cross-validated with Sentinel-1 SAR imagery (VH polarisation) obtained during the January 2026 event, and it was found that 85 % agreement occurred between delineation by optical index and independent delineation by SAR data. These results are consistent with the importance of cloud-based Google Earth Engine approaches for flood mapping in local government settings in South Africa. The proposed methodology is entirely based on publicly available information and will not involve proprietary software, and can easily be applied to other flood-prone districts, while offering a practical solution for disaster risk reduction and infrastructure exposure assessment.