A digital elevation model (DEM) of 1:250,000 scale and a grid resolution of 30 m × 30 m for the study area was obtained from the USGS website. This DEM was projected using Universal Transverse Mercator 36 South (UTM-36S). DEM filling and correction was done using the artificial “sinks” method. A threshold value of 500 pixels was selected. In this method, flow direction and accumulation grids were used to determine the accumulated weight of each pixel on a downslope and a threshold value (500 pixels), beyond which all grid pixels were considered being stream pixels. This approach was also used to map catchment boundary by using contributing up-slope area method. The model catchment boundaries and stream networks were both generated from the DEM using ArcGIS 10.4.1.

The study area was clipped from the FAO Digital Soil Mapof Mozambique. Average physical properties for water holding capacity; hillslope length; hillslope; upslope contributing area; and maximum cover of land that is impervious were assumed to depend mainly on the slope of the basin and considered automatic for the DEM (Figure 2).

The study used Landsat a multispectral image with a 30-meter resolution for land use classification. The Landsat satellite image scene was obtained from the USG archives (https://ers.cr.usgs.gov/) for land use classification (Table 1). Land use/cover map was then reclassified to reflect classes in the basin for use in SWAT.

Image pre-processing and atmospheric correction; supervised classification and maximum likelihood algorithm were done in ArcGIS 10.4.1 and ENVI 5.1 software versions. The image was pre-georeferenced for WGS 84/UTM.

The basin has a sharp elevation gradient (799 m.asl - 277 m.asl) and the Incalaue River drains the catchment with tributaries of Nipatembe, Lulo and Manyanganya (Figure 3). There are 6 vegetation classes of high-density woodland, medium-density woodland, low-density woodland, wooded grassland, mountain forests, and wetland (Figure 4). The rest of the basin area is built-up area, burned vegetation areas and inselbergs. Soil samples were taken at the corners of 5 square meter sampling plots, and samples were uniformly mixed for classification. Soil depth pits were dug in the center of the plot as deep as possible until the hard rock would be reached. These soil sampling plots were put randomly inaccessible vegetation classes in the area (Table 2).

Vegetation and soil sampling would be done simultaneously at the same location during fieldwork. Vegetation in a location that was confirmed to exist in a class was based on the mapping of known classes [27] [28]. The soil pits could not be dug due to wildlife hazard risks in the mountain forest and wetland vegetation classes as these are not accessible because of wildlife concentration in these habitats. In the area, grass, shrub and bush vegetation have been reported to grow roots at least 1 m depth [29].

NASA-POWER satellite-based weather data was used to calibrate the SWAT model for comparison with field data. POWER provides a gridded database of freely available global meteorology and surface solar energy climatology data. The data is available to download with a resolution of 1/2 by 1/2 arc degree longitude and latitude making it potentially suitable for hydrometeorological studies. Data generation is funded through the NASA Earth Science Directorate Applied Science Program. The NASA POWER data has largely been used in agroclimatology modelling [16] [30] - [36]. The model has been used for estimating the renewable energy potential in Africa [37]. NASA-POWER data was downloaded for a center point for the catchment at latitude −12.333 and longitude 37.8125.

SWAT model-generated data was used to generate historical meteorological data for the basin for the model since the basin ungauged. In the SWAT model, there is a WXGEN weather generator model which is used to generate acceptable

climatic data for modelling purposes [18] [38]. Using the Green & Ampt method for infiltration, maximum temperature, minimum temperature, solar radiation and relative humidity, the weather generator independently generates the distribution of rainfall within the day; and wind speed is generated independently. This tool was downloaded from the SWAT website (http://www.brc.tamus.edu/swat/soft_links.html). SWAT WXNGEN data could be available for the area up to 2014 and we depended on its relationship with NASA-POWER to adopt the latter for modelling of the remaining time. In the model, missing weather data were given a negative value (−99.0) in the model which instructs the weather generator of the model to generate weather data for that day.

Flow data is one of the major challenges to water modelling in ungagged catchments. In this study, the SWAT model inbuilt rainfall-runoff model was used to get the flow for use in modelling because there was no nearby catchment that was gauged. There was a good correlation coefficient (0.8) for rainfall-runoff modelled using the FAO dataset for the period 1980-2014 and thus SWAT WXGEN data was adopted for calibration (Figure 5). A coefficient of determination commonly known as R-squared (or R²) is a measure of the amount of variance in the dependent variable that is explained by the independent variable. It shows the strength of a linear relationship between two variables and examines how differences in one variable can be explained by the difference in a second variable.

SWAT data available for the area for the years 1980 to 2014 showed rainfall-runoff with flow peaks at the end of the year and the start of the next year (October to April). The minimum flow levels, when averaged, showed that monthly river flow can be zero during the dry season with river flow peaks for year being January to April with a low minimum for the months of June to October. The strong coefficient of determination (R² = 0.8) shows a good water balance in the basin. This shows that the variance in the river flow is explained by rainfall variation which makes the relationship reliable for the generation of river flow from rainfall data.

Household interviews were used to collect data from the local community knowledge and experiences on seasonal rainfall-runoff, water availability, trends and threats to water availability. This approach was used to gather information verify modelled data since the basin is ungauged. Community consultations were held in the human settlement areas of Ntimbo 1 and Lisongole in the downstream area. Sampling was done by randomly selecting household heads or adult family members who had stayed in the area for >20 years. This was used the back-up remotely sensed data on trends and field-collected data.

The estimated number of households was 123; where 56 were in Ntimbo 1 and 67 in Lisongole. Interviews were held in the dry season timing in the afternoons

when people are not in gardens. This approach of opening up household interviewee was used to avoid bias as members are randomly selected depending on availability in homesteads while ensuring efficiency by sampling adults with experience of the area [39]. The closeness of communities using the river at similar points ensures data reliability and further enhances historical data reliability.

Data analysis was done in Microsoft Excel.

Soils data showed that wooded grasslands had the most uniformly mixed soil among sampling sites in areas with less stony compacted soil (Table 3). In the area, grass, shrub and bush vegetation have been reported to grow roots at least one-meter depth [29].

The shallow stony soil that was not very easy to dig through in the wet soil zone is a sign of compaction that leads to higher rainfall-runoff given that it is a hilly landscape. Soil samples were taken to the laboratory at Eduardo Mondlane University in Maputo for laboratory analysis and these showed soil groups in vegetation classes (Table 4).

The soils were predominantly sandy and sand particle size (Table 5). This kind of soil in this sloping landscape means more rainfall-runoff and sedimentation.

Soils were mainly composed of particle sizes in the classes of <2 mm classes which shows sand soil with smaller granules and this makes it to be prone to erosion that can cause dense sedimentation in the river channel. The uniformity of samples from different sampling sites was tested to estimate deviations in distribution across the landscape using log-log plots (Figure 6). The results above showed that it is only sand content and bulk density that can be related to all the sampling sites. The study results above interestingly show that FAO characterization of the soil misses the top-soil layers and characteristics and these are even most important on river flow generation, composition and quality. The study thus shows a need for soil characterization to support rainfall-runoff studies.

There is no data available on groundwater harvest in the area and only one borehole was observed within the catchment at Lisongole village and another in a nearby town of Mecula. This means that groundwater springs and contribution to river flow cannot be completely ruled out. However, in the bigger landscape and within the basin itself there are dambos which are shallow vegetated areas with wetter vegetation in the dry season and micro-dambos were observed to

have groundwater small streams directly flowing to the rivers through small wetlands that do not dry in the dry season. Dambo areas have been reported in the area as shallow wet areas in a landscape that have wetter vegetation annually in this area with a tropical sub-humid climate characterized by a strong lengthy dry season [24] [40].

NASA-POWER data shows seasonal similarity with gauged rainfall but with much higher values (Figure 7).

NASA-POWER data confirmed additionally that June, July, August and September as the dry moths and this could imply the reduced river flow and river channel sedimentation in this wet season as was confirmed by the SWAT model. The similarity in rainfall trends but a weaker rainfall-runoff relationship for gauged data can be attributed to landscape characteristics such as water losses from catchment storage. The NASA-POWER meteorological data shows a good seasonal trend for temperature, relative humidity and rainfall again in support of reliability for the study (Figure 8).

SWAT model WXGEN rainfall data generates a close rainfall-runoff relationship compared to when NASA-POWER is used (Figure 9).

Data collected by this study also showed a positive trend a good pattern for rainfall and river flow. The peaks for the two years of fieldwork are December

and April in each case the dry season starting in May (Figure 10).

The rainfall relationship over the two years of fieldwork shows a positive trend but with a stronger relationship than observed in SWAT model generated WXGEN data and NASA-POWER data (Figure 11).

Rainfall and river flow data recorded during fieldwork also showed a positive rainfall-runoff relationship with seasons of no rainfall and no river flow periods, and little rainfall and no river flow because of the sandy soil and catchment water storage. The short database collected cannot effectively explain the lower coefficient of variation and the relationship.

Data from the community showed reliance of groundwater springs and confirmed modelling trends. The study showed that area climate variations more than human factors influence of river water availability and trends (Table 6).

The field data collected and observations were similar to community reports and the situation means vulnerability for human settlement community about

water availability and sustainability which requires integrated water resources management and water supply investment. This study found a rainfall-runoff trend as was reported by the community further supporting the potential of the SWAT model to generate rainfall-runoff using NASA-POWER data (Figure 12).

There is a small and more stable change in river flow over the peak seasons in the time peri0d of 2001-2021 studied. This shows increase in water loss possibly by natural landscape processes and changes than human influences causes because no rainfall-runoff storage was observed or reported in the catchment.