Flood events occur frequently in many regions of SSA due to high climate variability and associated extreme precipitation events, with expectations that floods will become more frequent and extreme with climate change (Asare-Kyei et al. 2015). Growing population along with continued socio-economic changes such as urbanisation and farming expansion increase exposure to flooding (de Moel et al. 2011; Di Baldassarre et al. 2010; McGranahan et al. 2007) and may also exacerbate flood hazards. Flood exposure is predicted to double by 2050 in SSA (Jongman et al. 2012), resulting in large increases in flood risk (Afriyie et al. 2018; Hirabayashi et al. 2013; Winsemius et al. 2016). Flooding poses long-term challenges to livelihoods in SSA, not only through loss of lives, destruction of farmlands and infrastructure (Hirabayashi et al. 2013), but also disease outbreaks and worsened food and water security (Asare-Kyei et al. 2015).
The damaging impacts on livelihoods can be reduced by proper flooding mitigation strategies that are guided by sufficient flood hazard assessment, monitoring and early warning. However, lack of monitoring and information on flood extent and flood exposure hampers spatial targeting of effective mitigation strategies. Flood exposure, defined as population and assets located in flood-prone areas (Muis et al. 2015), however has received little attention to date (Smith et al. 2019). Assessing flood exposure of population and assets is particularly crucial in data scarcity region of SSA, particularly as it is the only global region showing increasing flood mortality rates since 1990 (Tanoue et al. 2016). Accuracte flood mapping and monitoring the location of people and assets exposure to flood can provide a foundation for flood risk assessment and mitigation that can help address the increase in mortality rates (Menoni et al. 2012).
The availability of accurate historical and current information on flood hazard events is particularly limited in SSA (UNDRR 2019). Flood hazard models have been implemented for flood forecasting and monitoring (Perera et al. 2019; Tarchiani et al. 2020). However, the accuracy of model-based flood hazard maps is restricted as it depends on the accuracy of various input data such as meteorological and topography data, thereby leading to greater uncertainty from error propagation (Ward et al. 2015). In addition, these flood hazard models tend to focus on the national or regional scale and are not designed for local-scale estimation where the impacts are experienced and local-level decision are required (Ward et al. 2015). Global scale flood datasets derived from satellite data are avaliable such as MODIS NRT Global Flood Product (Nigro et al. 2014) and the MODIS Global Flood Database (Tellman et al. 2021), The Global Flood Detection System (Kugler and De Groeve 2007). These datasets again have coarse-spatial resolution and validation of them is highly challenging, particulary in data-sparse regions such as SSA (Revilla-Romero et al. 2015) where accurate data is greatly needed for flood risk management. In this context, improving flood monitoring accuracy at local scale is greatly needed (Du et al. 2021; Singha et al. 2020).
Progresses have been made to provide flood monitoring at the local scale and over long term,
particularly with development in Earth Observation (EO) systems of increased revisit frequency and higher spatial resolution that are increasingly used in operational disaster monitoring systems (DeVries et al. 2020). Synthetic Aperture Radar (SAR) is particularly useful for flood mapping since it can provide frequent observations (Alsdorf et al. 2007; Ward et al. 2014) thanks to its capability to monitor land in almost any weather conditions (Cian et al. 2018; Marzano et al. 2012). Flooded areas generate a low backscatter signal and appear to be very dark in SAR images, which makes them distinguishable from other land cover classes such as agricultural land or built-up areas. Several SAR-based flood detection techniques have been proposed (Tsyganskaya et al. 2018), such as histogram thresholding or clustering (Martinis et al. 2009), change detection (Li et al. 2018; Long et al. 2014), and time series analysis (Cian et al. 2018). Defining a robust and objective threshold for classifying flooded area is one of the challenges in accurate flood mapping using the above methods (O'Grady et al. 2011). Currently, most studies are either based on a universal threshold value that may be unsuitable for specific sub-regions or dependent on user-defined empirical analysis, which makes it difficult to apply in different study regions. This study, however, propose an approach that define thresholds that tailed for local case study in an objective day, that is, through using optical satellite data to define thresholds for SAR-based flood mapping techniques. Optical satellite data have been also widely used for flood mapping and especially for long-term flood monitoring (Islam et al. 2010; Qi et al. 2009; Sheng et al. 2001), despite the impact of clouds during the rainy and flood season (Singha et al. 2020). Our study therefore use a combination of optical images and SAR images to improve existing algorithm for flood mapping using SAR data, as well as provide a more complete estimation of flood extent and enable long-term flood monitoring (Tong et al. 2018).
In addition, the lack of local ground data that used to evaluate the accuracy and limitations of satellite data-derived flood extent has been another major drawback in previous studies. Most studies that assess accuracy via inter-comparison of satellite-based flooding maps (DeVries et al. 2020; Singha et al. 2020), for example, optical satellite images which were used to evaluate the accuracy of flood mapping from radar satellite images (Singha et al. 2020), however often underestimate flood extent given high cloud occurrence therefore providing insufficient accuracy assessment. This study instead collected ground data through participatory mapping to assess the accuracy of satellite-based flood maps. Participatory mapping, which engages local knowledge and expertise and allows local communities to delineate flood-affected extent on provided basemaps, has been widely recognized as an effective tool to collect and understand flood extent on the ground (Kienberger 2014). Involving communities’ knowledge of floods through participatory mapping is critical in the data-scarce SSA context as communities experience flooding first hand. Despite the value of local experience and knowledge, very few attempts have been made to combine flood extent derived from satellite data and through participatory mapping (Asare-Kyei et al. 2015; de Andrade and Szlafsztein 2015; Xueliang et al. 2017) for evaluating satellite-derived flood extent accuracy. Such a comparison can not only enable accuracy assessment of satellite-derived flood extent, but also indicate flood-prone areas associated with high impacts for local communities.
This research, for the first time, employed a combination of top-down approaches based on multi-source EO measurements and a consultative approach via participatory mapping to map flooded areas and their dynamics. Further, it investigates the scale and severity of population, infrastructure and farmland exposure to flooding in the White Volta basin in Ghana. Specifically, this study aims to:
Map flood area at fine spatial scale (i.e. 10 m) using multi-source satellite images (including Sentinel-1, Sentinel-2 and Landsat-8) and analyse flood area dynamics through comparison with existing global flood datasets (e.g. MODIS NRT Global Flood Product, The European Commission’s Joint Research Centre (JRC) Global Flood Database, and JRC Global Surface Water dataset) over 2000-2020.
Evaluate the accuracy of satellite-derived flood extent (from this study and the existing global flood datasets), through comparison with participatory mapping outputs.
Estimate flood exposure by combining the satellite-derived flood datasets with socio-economic data including high-resolution population density (100 m and 30 m), land use (30 m) and key infrastructures.