Modelling future flood events under climate change scenarios in the Pungwe River Basin

Climate change has direct impacts on ecosystems and the services they provide to society, thus influencing human well-being, health (Kalantari et al. 2018), and increasing water-related extremes such as floods events, with significant impacts on the economy and environment (Di Baldassarre et al. 2017). Flood occurs due to rapid population growth, land degradation, and climate change (Desalegn & Mulu 2021). To reduce the effects of water-related natural hazards and their societal effects, it is imperative to create more efficient strategies, methods, and tools for incorporating water-based analyses, such as flood mapping into spatial planning, especially in the most vulnerable countries (Kalantari et al. 2018; Korah & Cobbinah 2019);

The latest IPCC report claimed that extreme environmental conditions, including extreme floods, will become unprecedented in magnitude, frequency, and timing in high-risk areas and new regions not typically prone to natural disasters (IPCC 2021). This report claim call for hydrological research and assessment of climate change impacts as a central issue at all levels (Korah & Cobbinah 2019). In fact, several climate change impact studies have been done, but the gap is that they characterize African River Basins and cities therein as homogenous, experiencing the same climate events and requiring the same planning solutions (Gaisie & Brandful 2023). However, African River Basins and cities are diverse and different across regions, scales, and contexts, despite some resemblance of their settlement's problems (Ramayanti et al. 2022; Gaisie & Brandful 2023). The Pungwe River Basin is vulnerable to the adverse effects of climate change such as flooding on a yearly basis due to its downstream characteristic, coastal proximity, and exposure (Williamson et al. 2023), representing a high risk of damage to the community, infrastructure, destruction of livestock and loss of life. For example, tropical Cyclone Idai struck the coastal city of Beira, on 14 March 2019. The cyclone brought heavy rain accompanied by strong winds that lasted for more than a week (Government of Mozambique 2019; Ramayanti et al. 2022). As a result, the Pungwe and Buzi Rivers Basins overflowed and inundated low-lying areas (Guo et al. 2021). This disaster caused 4,000 houses to be damaged or inhabitable, about 1,600 people were injured, and 603 people were killed in Mozambique, Zimbabwe, and Malawi (UNDP 2019).

Therefore, flood hazard maps are an essential tool to help planners and decision-makers to better understand the spatial distribution of hypothetical flood characteristics more directly and easily (Grežo et al. 2020; Chen et al. 2022) and provide necessary information for many strategies for mitigating, managing flood and setting policies for land development and insurance (Liu et al. 2021). Scientifically, two main approaches can be used for flood mapping and risk assessment (Chen et al. 2022). The first one involves using indicators to assess flood risk with the application of geographic information systems (GIS) and multi-criteria analysis (MCA) to estimate current and future flood risk maps (Rahmati et al. 2016; Chakraborty & Mukhopadhyay 2019). This approach is limited by the need for exploiting experts' knowledge in assigning weights, which can be considered a source of bias (Rahmati et al. 2016). The other approach simulates flood event variables, such as peak discharge and flood inundation, using hydraulic and hydrologic models under varied conditions. Hydrologic models such as HEC-HMS (Gumindoga et al. 2016; Nharo et al. 2019), and TOPMODEL (Januário et al. 2022) were used for streamflow simulation. Gumindoga et al. (2016) applied HEC-HMS for ungauged runoff simulation in the Upper Manyame Catchment of Zimbabwe. Different authors have also used hydraulic models such as HEC-RAS (Chipere et al. 2020) and SOBEK (Yue et al. 2022) for flood inundation mapping. Desalegn & Mulu (2021) used HEC-RAS for mapping flood inundation areas at Fetam River, Upper Abbay Basin, in Ethiopia. The study discovered that flooded areas in the upstream and middle parts of Fetam River are high as related to the downstream parts, which aided in mapping high-hazard areas.

That said, the serious impact of floods on the environment, the extensive damage to infrastructure, and human well-being in the cities located in and along the Pungwe River Basin is an important reason to assess flood mapping extent risk. Flood mapping identifies areas most vulnerable to flooding based on physical characteristics (Ramayanti et al. 2022). The maps of flood extent should be accurate and rapidly obtainable because it is essential for flood prevention and mitigation (Zare & Schumann 2021). However, the limited time available to map all areas, impractical fieldwork, or traditional hydrological methods is the challenge to generating an accurate map (Rahmati et al. 2016; Ramayanti et al. 2022). Therefore, since high-impact events such as floods often span spatial scales larger than those covered by traditional regional monitoring, integrating remotely sensed data with flood modelling, especially via satellite, is an appealing approach to the quick, fair, and accurate assessment of flood events, real-time monitoring and design of early warning systems (Schumann et al. 2018; Nharo et al. 2019).

This study advocates for using GIS, Earth observation, hydrological, and hydraulic modelling techniques with HBV and HEC-RAS to map the spatial extent of floods and assess the impacts of these floods in the context of climate change. To the author's knowledge, this is the first study to attempt the integration of all those models in generating flood extent maps. It is therefore significant to improve knowledge of mapping floodplains areas, and knowledge of catchment precipitation patterns in the basin in the face of climate change and variability to inform development pathways, and land and water management options (Nyabeze 2021). Specifically, the objectives of this study are: (i) to project future changes in climatic variables for the mid-2050 and 2080s; (ii) calibrate the HBV model and estimate future changes in extreme and seasonal streamflow due to climate change; (iii) estimate the flood inundation extent for various return periods (50, 100 and 1,000 years) in the context of future climate. The results of this study are important for the development of policies and strategies for flood management, map areas for urban and social infrastructure development, development of early warning systems, and technical assistance for water and not water-related specialists. The final products of maps for inundation extents herein presented can be used by institutions such as the Regional Water Administration – Centre of Mozambique (ARA-Centro), the National Disaster Management Institute (INGD), National Meteorological Institute (INAM).

Observed Hydrometeorological data were obtained from Administração Regional de Águas do Centro – ARA-Centro (Regional Water Administration – Centre) and Instituto Nacional de Meteorologia – INAM (Meteorology National Institute) of Mozambique, respectively. The data were from nine stations in the basin (Table 1).

Regional Climate Models (RCMs) used in this study were downloaded from the CORDEX-Africa (https://esg-dn1.nsc.liu.se/search/cordex/) runs downscaled from 10 GCMs of the Climate Model Inter-comparison Project Phase 5 (CMIP5) (Chipere et al. 2020). The latest version of the regional climate model by the Swedish Ross Centre's regional Atmospheric model – RCA4 was used for this study. It features an improved physical and energy flux parametrisation, and its modified radiation scheme has considered the absorption and concentration of carbon dioxide. Based on the recommendations of the land-use information, a quadrilled land surface scheme is proposed. The scheme consists of one to three key tiles, and the spatial resolution of the simulations for the CORDEX-Africa domain is 0.44 × 0.44. The projections are for 1951–2100 for both historical and future periods, 1951–2005 and 2006–2100, respectively. The projections were for RCPs 4.5 and 8.5 for two periods near the future (2022–2060) and far future (2061–2099). These models have been evaluated in multiple studies on Africa, and have been found to perform well in simulating rainfall and temperature (Worku et al. 2018; Dibaba et al. 2019). The RCA4 simulations used and evaluated in this study are described in Table 2.

RCMs are used for forecasting future climate, including rainfall and temperature characteristics (IPCC 2014). Therefore, to effectively quantify the impacts of climate change at the local scale, it is necessary to select appropriate RCMs (Dibaba et al. 2019). Each model, however, exhibits some instances of bias therefore the use of a single RCM for climate change impact analysis gives room for further disputed uncertainty (Chipere et al. 2020). To reduce uncertainty, this study made use of the multi-model ensemble approach. As such, using the observed baseline daily data, the performance of the selected climate models in mimicking past observed temperature and rainfall was assessed using the performance metric utilised in several past studies (Dibaba et al. 2019; Demissie & Sime 2021) namely Nash–Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), coefficient of determination (R²), and index of agreement (d).

Computer models are frequently used for catchment hydrology studies. The HBV model (Bergström 1976) belongs to the conceptual class of hydrological models, characterised by attempts to cover the most essential runoff-generating processes using a simple and robust structure and a small number of parameters. The model consists of three main modules: snow accumulation and melt, soil moisture accounting, river routing, and response modules. This model is often used in discharge simulation, flood forecasting, and assessment of environmental change (Poméon et al. 2017; Hegdahl et al. 2023). The Shuttle Radar Topographic Mission 30-m resolution digital elevation model downloaded from https://earthexplorer.usgs.gov/ was used to delineate the watershed and extract terrain-specific characteristics. The DEM was used for DEM hydro-processing in the Integrated Land and Water Information System (ILWIS) to obtain hydrological features. Maathuis & Wang (2006) procedure for DEM hydro-processing has been used in the Pungwe River Basin for flow determination, drainage network and catchment extraction. From the process, eight sub-basins have been defined for further study analysis in the basin.

A PTQ-file (ptq.txt) that contains time series of daily precipitation (mm/day), temperature (°C), and discharge (mm/day) was created. One PTQ-file was created containing all the sub-basins. Using ILWIS GIS, the slicing algorithm was applied to the Pungwe River Basin raster map to create the elevation zones. From the filled DEM, the elevation ranged from 206.13 to 1,649.04 m. From the DEM, eight classes were divided at a minimum elevation interval. Three vegetation zones in HBV were selected. Land-use images were acquired from CCI land cover – S2 prototype land cover 20 m map of Africa 2016 with a spatial resolution of 20 m × 20 m and imported to ILWIS. Initially, the image had more than three vegetation zones reduced to three using an ‘IFF’ function. The vegetation and the elevation zones maps were crossed, and the product was crossed again with the sub-basins map. For each sub-basin 24 elevation, vegetation zones were created. Before crossing, the maps were resampled to ensure they were in the same geo-reference system.

The calibration is the estimation of the parameters that enable the model to match the real system's behaviour (Van Tiel et al. 2020). In this study, manual calibration was carried out by searching the parameter values that maximise the model's reliability (Hegdahl et al. 2023). The calibration period was from 2000 to 2010. Before calibration, 1999–2000 was used for model warmup. The validation period was from 2011 to 2015. The calibration and validation were done using the dataset for catchment 1, on which the data series had consistency and reliability for modelling and analysis. The calibrated and well-optimised model parameters were kept constant to generate streamflow for the baseline and future periods. To analyse the performance of the simulation, the NSE, coefficient of determination (r²), mean difference (MD), efficiency on peak flow (EPF), and standard deviation (σ) were used.

The Hydrologic Engineer Center's River System Analysis System (HEC-RAS) software version 6 and the QGIS were used for hydraulic simulations and flood inundation mapping (Brunner 2008). The Pungwe River Basin 30 m SRTM DEM downloaded from https://earthexplorer.usgs.gov/ was used to create Geometric layers such as the stream centerline, bank lines, flow path centerlines and cross-section for the Pungwe, Nhazonia, Urema, Metuchira, Nhacarangue, Mucombezi, Messambize, Nhancangale, Vunduzi, and Nhadungue rives were created from the DEM. Manning's coefficients were defined as 0.02 for the channel and 0.03 for the riverbanks based on the field observation of the land-use practices. The discharge peaks for the near and far future at different timesteps derived from the HBV model and statistically predicted for 50-, 100- and 1,000-year return periods using Gumbel Extreme value were used for future flood inundation modelling under climate change scenarios. Inundation extents were computed for the historical and future periods. The flood extent was validated using the Cyclone Idai-induced flood on 19 March 2019.

To project changes in rainfall and temperature, first, the performance of RCMs were assessed and to unbound uncertainty associated with climate models, 10 models have been studied on simulating the rainfall and temperature of the Pungwe River Basin and a multi-model ensemble approach have been adopted for further analysis (Dibaba et al. 2019). This approach has been used by Chipere et al. (2020) to study future flood inundations under Climate Change in Middle Zambezi Basin in Zimbabwe. On the results of the study, the authors verified on the analysis of the historical GCM ensemble data against station data daily, that the GCM's ensembles were able to capture most of the peak precipitation periods for each catchment. This approach is crucial for the purposes of establishing flooding events in the past. Table 3 shows the results of the performance of the climate change models analysed at a daily timestep. The overall analysis shows that the models performed well on temperature analysis compared to rainfall; this happened because rainfall has greater spatial and temporal than temperature (Masimba et al. 2019).

The results also show similarity with the one observed by Masimba et al. (2019) when assessing the baseline and downscaled projected climate variables in the Upper Manyame sub-catchment of Zimbabwe and concluded that the projected rainfall did not show a consistent increase or decrease in all the projected periods. However, the author verified that there will be a decline in the amount of rainfall during the onset of the rains that is October and an increase in the duration of the mid-season dry spell as shown by the decline in the amount of rainfall received in January. This indicates that the rainfall distributions do not show linear projections during the summer periods when most flood events occur and will impact negatively on the planning mitigation measures and call for the need of preparedness actions such as flood susceptibility mapping due to the shift in seasonal rainfall distribution that can lead to flash floods in the Pungwe River Basin (Ramayanti et al. 2022).

From Table 4, it can be depicted that the HBV model performed well in reproducing the peaks and baseflows during the calibration period (1999–2010), although there are some overestimations on peak flow during the simulation period. For the Ponte Gorongosa sub-catchment, the Nash–Sutcliffe efficiencies (NSE) obtained were 0.88 and 0.83, 0.89 and 0.82 for the coefficient of determination, 4.04 and 7.07 for mean difference, 0.91 and 0.84 for efficiency of the peak flow, and 3.67 and 6.41 for standard deviation during the calibration and validation period, respectively. These indices obtained suggest good model performance (Wubneh et al. 2023). The most sensitive parameters are the percentage of imperviousness and other parameters less sensitive but calibrated are: initial deficit, maximum deficit, and constant rate. Ramesh (2012) asserts that the proper characterisation of topography plays a vital role in runoff generation. Thus, the obtained objective function values indicate that the topographic-index distribution function for the catchment is adequate.

However, thou the performance of the model during calibration and validation periods is good, it must be noted the drawback caused by the lack of long series of data which led to the use of one single station with reliable runoff data for simulating the rest of the sub-basins, which limits the use of the model for future runoff simulation. The other drawback is that due to the scarcity of runoff data, it was not possible to calibrate and validate the model at the estuary of the basin and some of the flooded areas are located at downstream of the calibration and validation point. Hamutoko et al. (2022), in his study of the challenges associated with climate modelling in Africa concluded that the lack of seamless access to available data and the poor quality/missing data, often associated with the long-term measurement of climatic variables poses a great limitation.

Changes in future hydrological extremes flows were assessed by analysing changes in the high flow from baseline and the percent of change in 2050 and 2080s. Table 5 summarises changes in peak flow in different RCMs. An increase in extreme flow is projected under RCP4.5 for Honde, Vanduzi, and Bue Maria sub-basin ranging from 10 to 61% for 2050 and 2080s. Under the same emission scenario, there is an expected decrease from 5 to 30% in extreme flow for Nhazonia and PC Inferior sub-basin for both periods and for PC Inferior and Urema a range of 8–12% decrease in 2050s. For RCP8.5, an increase in extreme events is projected for Honde, PC Superior, Vanduzi, Muda, Urema, and Bue Maria sub-basins. The range of increase varies from 11 to 104% for all future periods. The increase in extremely high flows, which is associated with floods, will impact negatively in many parts of Central Mozambique.

As the projected situation illustrates, it is expected that future flood inundation extension be greater if compared to the actual period, this will cause damage to infrastructure as compared to the damage that took place during the cyclone Idai-induced flood (Ramayanti et al. 2022). It is projected that there will be great negative impacts on schools, hospitals, houses, and livestock mainly of riverine people. Water and infrastructure planners need to look for strategies for mitigation of the effects of climate change by looking at future inundated areas as indicated by the map in Figure 7, this includes the construction of flood structures (Afzal et al. 2022), such as dams and impoundments.

This study has demonstrated the integrated application of a series of tools with the objective of mapping the future extent of flood under climate change using the Pungwe River Basin as an example. While the assessment of the climate change impact is not demonstrated in full, the results demonstrate, however, the significant vulnerability that the basin is facing. The analysis of the historical and projected hydrological and meteorological series is important and even more relevant when considering the localised effects of global climate change. Therefore, to detect climate change and analyse the climatic parameters. The main conclusions of our work are:

• The multi-model ensemble performed better in simulating the projected climate variables of the Pungwe River Basin in mimicking rainfall. There is no consistency in the projected signal of rainfall and temperature compared to the baseline. In general, there is an increasing projected change in temperature time series up to 15% and an inversely decreasing projection of change in rainfall time series in the stations located in the Pungwe River Basin. Results on projected rainfall, temperature and streamflow accretion would help to develop effective adaptation measures for reducing the impacts of climate change and to work out long-term water resource management plans in the river basin.

• The HBV model performed well in reproducing the extreme and seasonal flow during the calibration (1999–2010) and validation (2011–2015) periods with the Nash–Sutcliffe efficiencies (NSE) of 88 and 83%, respectively. In peak flow change analysis, it was observed that in the future is projected an overall increase in peak flow is up to 100%, leading to the occurrence of more frequent and intense flood events. There is a clear indication that climate change will impact in river basin flow, reducing the seasonal flow and increasing the extreme flows. This will reduce water availability and also influence the occurrence of flood events.

• In flood extent, an increase of the inundation area is projected with the return period of 100 years for 2080s, giving the overall maximum expected flood for the Basin under RCP8.5 extension of 5,041 km². The cyclone Idai-induced flood map from Sentinel 1 matched perfectly with simulated flood mapping, making the approach useful to validate map future flood events. While these results are obtained, there is a need of need for the integration of other factors that trigger floods such as land-use, and population density (urbanisation). Moreover, these results may change if new developments are seen in the Basin such as the construction of flood structures such as dams and impoundments. The results suggest that the population that live in river basins that might inevitably be affected by increased flood events during climate change. The last conclusion is associated with the uncertainty bound to climate change models. It is, however, well documented in the literature, while the aim of this paper is not to discuss this matter, it must be noticed that there is a chance that these results may be affected by this uncertainty and decision makes the need to consider when making plans with outputs from this research and the idea was giving a guideline for analysis of climate change impacts in extremes, mainly floods.

There is a need of development of more research analysing climate change, landuse, and urbanisation in the Pungwe River Basin. Modelling of extreme events requires the existence of long-term hydroclimatic quality data. Therefore, the ARA-CENTRO needs to improve the strategy of data manipulation from the collection methods, processing, and storage to avoid over- and under-estimating water resources and their flows and aid in proper data-driven solutions.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.