Assessing water availability using remote sensing in South Rukuru and North Rumphi River Basin, Malawi Open Access

Hydrological modelling is now frequently used to assess water availability as it provides reliability streamflow estimates and trends over long periods (Speed et al. 2013). One of the hydrological models that can be used to estimate the current status and trends of water availability in an area over a specific period of time is the rainfall–runoff soil moisture method approach in Water Evaluation and Planning (WEAP) modelling software (Yaykiran et al. 2019).

WEAP is a modelling software for integrated water resources planning, developed by the Stockholm Environment Institute (SEI). It provides a comprehensive, flexible and user-friendly framework for water resources management (SEI 2005). WEAP's soil moisture method takes into account a one-dimensional, two-layer (‘bucket’) soil moisture dynamic accounting system that uses empirical functions to partition water into evapotranspiration, surface runoff, sub-surface runoff (i.e., interflow), and deep percolation for a sub-catchment unit at the root zone (SEI 2022a). However, to effectively and efficiently assess water availability, there is a need for good quality hydrological time series data (Mwale et al. 2012).

Globally, there is a dense decline in the network density of operational hydro-meteorological field stations, which is an obstacle to the effective development of hydrological models (Karimi & Bastiaanssen 2014) that can be used to estimate water availability trends over long periods (Speed et al. 2013). In Malawi, more than 300 hydrological stations historically existed, but now most of these stations do not have a continuous record of data and some of the stations have been abandoned or closed. This decline in the network density of hydrological data is mainly due to low priority and limited financial resources for hydrological services (GoM 2014).

Limited availability of hydro-meteorological data in river basins worldwide increases the value of alternative data sources, such as remotely acquired data in hydrological modelling (Hulsman et al. 2020). The recently released Global Land Data Assimilation System (GLDAS) 2.0 version generates a variety of hydro-meteorological variables, including rainfall, air temperature, wind, specific humidity, soil moisture and many more at a spatial resolution of 0.25° (from 2000 to present) (Ji et al. 2015).

Several hydrological studies have been conducted worldwide to evaluate the use of remote sensing products for assessing water availability using simple conceptual water balance models. Moreira et al. (2019) investigated the potential of assessing the water availability using a water balance model in South America using satellite remote sensing precipitation and evapotranspiration datasets, terrestrial water storage changes from Gravity Recovery and Climate Experiment (GRACE), and discharge measurements. They concluded that there is indeed potential for assessing terrestrial water availability. Armanios & Fisher (2014) also conducted a study to assess the feasibility of using an entire satellite remote sensing based hydrologic budget model to measure water availability in the Rufiji basin, Tanzania. Their results showed that the hydrologic budget model had a good performance, and it was recommended that future researchers have to use newly up-to-date remote sensing tools that can provide near-real-time retrievals and also at good spatial and temporal resolution to a scale closer to that of interest of a water resource manager.

Only a few studies have used integrated hydrological and water resource models, such as the WEAP modelling software to assess water availability trends with remotely acquired hydro-meteorological data. Furthermore, most of these studies have been carried out in continents other than Africa. The studies were conducted in large river basins and used remote sensing datasets that had low spatial and temporal resolutions. GLDAS datasets have been widely used, and several studies have been conducted to evaluate their applicability (Ji et al. 2015; Bi et al. 2016; Zhang et al. 2018; Khasmakhi et al. 2020), and only a few of these studies evaluated GLDAS climatic data to develop a rainfall–runoff model in WEAP modelling software.

Since most of these studies have been carried out in continents other than Africa, there is a need for specific studies to be conducted in Africa, particularly in developing countries such as Malawi, to evaluate and validate the applicability of GLDAS datasets, considering that, in these developing countries, the hydro-meteorological data are needed the most (García et al. 2016), as the hydrological monitoring networks are in a poor state and have the most deficient infrastructure for gathering data needed to monitor and predict them (Sheffield et al. 2018). The hydro-meteorological networks have deteriorated over time, at present providing only limited hydro-meteorological information for developing effective and efficient hydrological models (Karimi & Bastiaanssen 2014).

Therefore, the purpose of this study was to evaluate the potential use of remote sensing datasets in the integrated water resources modelling software (WEAP), to develop a rainfall–runoff model (soil moisture method) that will be used to assess water availability within South Rukuru and North Rumphi River Basin. The study's specific objectives were two-fold: (a) developing a remote sensing rainfall–runoff model using the WEAP modelling software and (b) conducting a performance analysis of the remote sensing rainfall–runoff model.

The remote sensing datasets in this study are commonly and widely used and freely available on a global scale. Elevation data at 90 m (3 arc seconds) spatial resolution was derived from Hydro SHEDS digital elevation data obtained from the Space Shuttle flight for NASA's Shuttle Radar Topography Mission (SRTM). For climatic data, GLDAS climate datasets of precipitation, relative humidity, temperature, and wind speed on a monthly basis at a spatial resolution of 0.25° × 0.25° were used for a period from 2000 to 2021 (22 years). This climatic data was accessed from the GIOVANNI website (https://giovanni.gsfc.nasa.gov/giovanni/). Land cover data from the European Space Agency (ESA) Climate Change Initiative (CCI) project was used. It has 22 Land Cover Classification System (LCCS) classes of land cover with a spatial resolution of 300 m (SEI 2022a). The elevation and land cover data were automatically downloaded in the WEAP modelling software.

Streamflow data was obtained from Rumphi District Water Office for four hydrometric stations within the South Rukuru and North Rumphi River Basin for a period of 22 years. The first 16 years (2000–2015) were used for calibration, and the last 6 years (2016–2021) were used for validation. Observed streamflow data contained gaps or missing values. For this study, where streamflow data are missing, infilling of missing streamflow data were done using the spline interpolation methods in Microsoft Soft Excel.

The WEAP hydrological model was calibrated to estimate land use and soil-related parameters using the manual (trial-and-error) method until a good fit is observed between the measured and simulated streamflow. The calibration process was done for a period of 16 years (2000–2015). The WEAP embedded soil moisture method involves estimating seven soil and land use-related parameters (Ingol-Blanco & McKinney 2009; Abera Abdi & Ayenew 2021). Table 1 shows the seven soil and land use-related parameters that are used to re-calibrate the hydrologic model. A validation data set of 6 years (2016–2021) was used to assess the adequacy of the model. The adjusted calibration parameters of the hydrological model were now used in the validation data set and then the model was run. The hydrological model was regarded as a success when the resulting output at the validation stage was reasonably close to the observed measurements for the model.

The WEAP model performance was evaluated using the following objective functions: The coefficient of determination (R²) (Krause et al. 2005), Nash–Sutcliffe coefficient of efficiency (NSE) (Nash & Sutcliffe 1970), Index of Agreement (IA) (Willmott 1981), Percent Bias (PBIAS) (Gupta et al. 1999), Kling and Gupta Efficiency (KGE) (Gupta et al. 2009). The current version of WEAP is integrated with the R-programming language, which includes these objective functions. These objective functions have been widely used in WEAP evaluation studies and have been successfully implemented (Ingol-Blanco & McKinney 2009; Tena et al. 2019; Yaykiran et al. 2019; Abera Abdi & Ayenew 2021) for measured and simulated flows over the calibration and validation periods. For visual inspection, the monthly simulated and observed hydrographs were plotted.

In the North Rumphi sub-basin, the precipitation is around 2,873 mm/year, with higher values in December, January, February and March, while surface runoff and evapotranspiration reach 2,060 and 740 mm/year, respectively. Similarly, with the South Rukuru sub-basin, the peak precipitation and surface runoff occur each year from December to March. The precipitation, surface runoff and evapotranspiration reach around 1,175, 381, and 756 mm/year, respectively. For both sub-basins, the water availability is highest in the months of peak precipitation, with lowest water availability between June and September. In some months, the evapotranspiration exceeded the amount of precipitation in May, June, July, August, September and October. On the contrary, in months ranging from November to March, the evapotranspiration was less than the precipitation amounts. The baseflow, which is another crucial component of the streamflow, is also available throughout the year with 65 mm/year in North Rumphi sub-basin and 34 mm/year in South Rukuru sub-basin.

Over 60% of global precipitation is consumed by evapotranspiration and it is the largest withdrawer of water from the systems (Abera Abdi & Ayenew 2021). For this study, the evapotranspiration shows steadiness throughout the year, with the highest values in January for North Rumphi sub-basin, while for South Rukuru sub-basin the values occur in September and October. It clearly shows that even during the dry season, when there is low water available in the sub-basins, there is high evapotranspiration, which means there is a continuous decrease in the total water storage of the system. In summary, the North Rumphi sub-basin presents the highest precipitation and surface runoff, with evapotranspiration being lower compared to South Rukuru sub-basin in the hot dry seasons for the entire period.

The performance of the model was tested using statistics from the simulated streamflow and observed streamflow. Table 2 shows general criteria for performance evaluation in a statistical test at the watershed scale.

According to Santhi et al. (2001) and van Liew et al. (2007), the values of R² that are greater than 0.5 are acceptable with higher values indicating less error variance. The closer the NSE values get to 1, the better the correspondence between observed and modelled data (Fernández-Alberti et al. 2021). Like NSE, KGE = 1 indicates perfect agreement between simulations and observations (Gupta et al. 2009). For IA a computed value of 1 indicates a perfect agreement between the measured and predicted values, and 0 indicates no agreement at all (Willmott 1981). PBIAS with a smaller magnitude shows better model performance (Gupta et al. 1999).

For comparison between the measured and simulated streamflow, goodness of fit metrics presented in Table 3 showed a stronger agreement for both stations with R² of 0.87, NSE of 0.86, IA of 0.97, KGE of 0.93 and PBIAS of −0.2 for station 7G18. Similarly, station 7H3 had an R² of 0.93, NSE of 0.92, IA of 0.98, KGE of 0.84, and PBIAS of −10 during the calibration period. For the validation period, there was a reasonably good agreement between the measured and simulated streamflow for the two stations with R² of 0.73, NSE of 0.72, IA of 0.92, KGE of 0.79 and PBIAS of −10.9 for station 7G18. Likewise for station 7H3, R² of 0.95, NSE of 0.94, IA of 0.98, KGE of 0.87 and PBIAS of −7.7 during the validation period.

According to Moriasi et al. (2007), for the calibration period, the model performance was achieved to simulate streamflow with NSE, R², IA, PBIAS, and KGE values that have shown a strong agreement between measured and simulated streamflow in both catchments of the model. Likewise, for the validation period, there was a very good agreement between the measured and simulated flows in both catchments (see Table 3). However, the PBIAS values for all the catchments and periods were negative, which indicates that the model was slightly underestimating the streamflow values. In general, the WEAP model was able to maintain a very good agreement in reproducing the overall streamflow characteristics. The model shows an outstanding representation of both catchments, with the best fit to the North Rumphi Catchment. Thus, these results indicate a more than reasonable ability of the WEAP hydrological model to simulate the flows using remotely acquired data.

Similarly, previous studies have confirmed the capability of the WEAP hydrologic model in reproducing catchment hydrology processes in different parts of the world (see Table 4). Among these, Abera Abdi & Ayenew (2021) reported that the WEAP hydrologic model attained the R², NSE, and IA values of 0.82, 0.80, and 0.95, respectively, between monthly measured and simulated streamflow in the Ketar River Basin, Ethiopia, for the calibration period. During the validation period, the model showed a reasonably good agreement between the measured and simulated flows with R² of 0.91, NSE of 0.91, and IA of 0.98. The performance of the WEAP model simulation in the Chongwe River Catchment, Zambia, showed a positive strong relationship using the R² of 0.97 and the NSE of 0.64 (Tena et al. 2019). Ingol-Blanco & McKinney (2009) developed the WEAP model to assess the hydrologic processes in Rio Conchos Basin, Mexico. Six gauging stations were used in the basin for model performance evaluation. The goodness of fit metrics values range as follows: NSE from 0.65 to 0.87, R² from 0.92 to 0.97 during calibration and for validation NSE from 0.60 to 0.88 and R² from 0.92 to 0.97 between measured and simulated flows, respectively. Finally, to characterize and evaluate the water supply and downstream demands of the hydrological basins associated with the Conguillío National Park, located in the Andes Mountains in central-southern Chile, the WEAP hydrologic model was developed by Fernández-Alberti et al. (2021) which has demonstrated the capability from four gauging stations with values ranging as follows: NSE (0.80–0.86, 0.77–0.84), R² (0.86–0.94, 0.88–0.96), and BIAS (0.04–0.19, 0.08–0.23) at calibration and validation periods between measured and simulated flows data, respectively.

The rainfall–runoff model was developed in WEAP modelling software using GLDAS datasets of precipitation, air temperature and wind speed. The hydrological model was able, not only to estimate water availability trends over longer periods for the South Rukuru and North Rumphi River Basin, but also to monitor nearly all components of the hydrological cycle. In a temporal context, the WEAP model hydrological response showed that in months at which the precipitation was at the peak in all the sub-basins, the surface runoff was also at peak. The lion's share of available water returns to the atmosphere via evapotranspiration in both sub-basins. However, the annual stream discharge is mainly composed of interflow and baseflow which contribute a major part of flow during the dry season, and during the wet season the main contributor is surface runoff.

Therefore, the sustainability of water availability in the South Rukuru and North Rumphi River Basin throughout the year is largely dependent on surface runoff in wet seasons and baseflow in dry seasons. Anything that alters the hydrologic behaviour of the sub-basins may impact the amount and the sustainability of water availability. The WEAP model developed has shown to be a reliable hydrological tool for water resources management and decision making for South Rukuru and North Rumphi River system especially in applications such as water allocation planning. This could help all the interested parties in the system in taking necessary steps to balance water supplies with demands, particularly to manage the natural variability of water availability, and to avoid frequent or unexpected water shortfalls.

The WEAP model performance was evaluated using various commonly used goodness of fit metrics and joint plots of simulated and observed streamflow. The observed streamflow data was also used during the calibration and validation of the remote sensing WEAP model. The WEAP model was calibrated using a trial-and-error method over 16 years (2000–2015) and validated for an independent 6-year period (2016–2021). Model calibration was performed to estimate the soil and land use-related parameters until a good fit was observed between the measured and simulated streamflow. Model validation followed to check the adequacy of the remote sensing WEAP model. The model was accepted only when the goodness of fit metrics showed a strong agreement between the measured and simulated streamflow. The WEAP model was able to maintain a very good agreement in reproducing the overall streamflow characteristics within the sub-basins. Investigations in this thesis is based in Northern Region of Malawi and only a single basin. It is therefore necessary that future work has to be done on the usage of GLDAS datasets to assess water availability in other river basins in Northern Region, Central and Southern of Malawi. In addition, the model requires the use of up to date GLDAS climate data inputs, and it is recommended that before using the model, the GLDAS climate data should be up to date.

We thank Rumphi District Water Office, particularly Mr Philip Kondowe, for the support rendered during data collection.

This work did not receive any funding.

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

The authors declare there is no conflict.