The study area is a 4,278 km² watershed with the 200 km Songwe River forming the physical boundary between the United Republic of Tanzania and the Republic of Malawi. It is part of the semi-arid Eastern and Southern African Great Rift Valley and lies between latitudes 9°6′23″–9°56′17″ South and longitudes 32°44′34"–33°56′31" East. Slightly over half of its total area falls in the districts of Ileje, Mbozi, Mbeya and Kyela of the Mbeya Region on the Tanzanian side and the rest in Karonga and Chitipa districts on the Malawi side. Three physiographic sub-basins (Figure 2) can be identified, with the upper sub-basin containing the headwaters and major tributaries at elevations ranging from 1,200 to 2,450 m above sea level. The middle sub-basin is comprised of mountains and tributaries that fall from elevations of 2,100-1,800 m down to 800-700 m at the foot of the escarpments within a distance of 20-30 km [7]. The lower sub-basin begins on the foothills down to Lake Malawi at 476 m above sea level. It is this part that dominates the Songwe river floodplain at elevations of mostly less than 1,000 m above sea level.

The Pan-European Soil Erosion Risk Assessment (PESERA), a simplified hydrological process-based approach [19], was used in this study to combine hydrological surface runoff factors to estimate sediment yield. By attempting to use advances in the understanding of runoff processes as opposed to sediment transport, it was only sensible for a forecast of runoff and soil erosion in PESERA to be built on a hydrological core [19]. While this paper concentrates on calibration, validation and its scenario application in the tropical region of the study area, a development account of the PESERA model is provided by [19].

Erosion by running water, resulting from precipitation [15], is the most severe hazard threatening the protection from soil loss [17,20]. Over time, such erosion-induced drainage basin changes affect river discharges [21]. With sedimentation being sensitive to both climate and land use, as well as to detailed conservation practices, temperature and precipitation proxies were used to estimate the climatic parameters for the model. Evapo-transpiration was partitioned proportionally to vegetative crown cover and its potential determined using the Hargreaves model.

The monthly rainfall was derived from averaging monthly daily rainfall data for the years 1998 through to 2006 obtained from seven gauging stations, namely: Itungi, Songwe, Kyela, Karonga, Isangati, Chitipa and Idiwili (Figure 2). Monthly mean rain per rain-day was obtained by averaging all days with a rain recording of greater than zero for each station to capture the daily storm events that drive infiltration overland flow and soil erosion. Mean monthly rains were also indicators of the saturation levels of the soil. It was manually observed that rainfall data for the seven gauging stations varied with elevation, latitude and longitude; as such, to obtain a realistic monthly distribution covering the entire study area, regression functions for each month were fitted to interpolate the rain values.

Infiltration excess overland flow runoff was estimated from storm rainfall and soil moisture. A Hortonian process of point hydrological balance was then used to estimate sediment transport from the excess overland flow. This was coupled to a vegetation growth and soil model to: (i) budget for the living biomass and organic matter subject to the constraints of land use and cultivation choices; and (ii) estimate the required hydrological variables from moisture, vegetation and seasonal rainfall history, respectively [19].

Runoff thresholds determine near-surface water store and soil properties constrain its upper limit [22]. The model parameters of available water storage capacity, crustability and erodibility were, therefore, obtained from the available soil properties. For agricultural land, full water storage after ploughing decays exponentially with time, reducing to minimum for vegetated areas [23,24]. This allows for seasonal response in runoff thresholds, and therefore in infiltration excess overland flow [19].

Vegetation cover reduces with real time processes of fire, plant gathering and grazing [19]. This relates empirically to the seasonal cover cycle and/or above-ground biomass of the land use classes [25]. Therefore, the PESERA model estimated subsurface flow, which influences infiltration excess overland flow, using the TopModel [26]. Topographic properties were estimated from local relief.

Sediment transport was then estimated with sediment yield, Y (kg m^-2 year^-1), defined as the sediment transported to the slope base, averaged over the slope length [19]. That is: (1) Y = k L Λ B ∑ r 2

Where k is the empirical erodibility value,

With the catchment experiencing seasonal rainfalls, the derived vegetation cover was pegged at 30% for all the months, except the wetter months of February (50%), March (70%) and April (90%) where the rainfall peaks [27]. The analysis incorporated a 30% [27] reduction in vegetative land cover due to fire, plant gathering and grazing. The model was corrected for snow fall and melt. Table 1 summarizes the 93 climatic, land use, topographic and physiographic parameters assembled and used to simulate the 2002/2003 season cycle on a monthly time scale with all parameters prepared at or re-sampled to 90 m resolution.

Basin storage of sediment is an important variable in reservoir sedimentation management practices as not all the sediment eroded from uplands is immediately delivered downstream [18]. Drainage density, which [16] defines as the capability of a stream network to discharge total rainfall and hence sediment from its watershed, is one indicator of basin storage quantified as: (2) D s = L s / Awhere D_s is the drainage density, L_s is the total stream length of the system and A is the total area of the watershed.

A multi-criteria evaluation (MCE) was then run to identify the critical areas requiring prioritized conservation management on the following factors: (i) distance to the river channel; (ii) slope; and (iii) the estimated PESERA outputs of runoff as a delivery medium [14] and sediment load. The estimated monthly sediment was also used to quantify the amount of estimated sediment input into the proposed reservoir in an assimilated flow accumulation. Using the Statistical Package for Social Scientists (SPSS), the statistical correlation of the increasing flooding events and the rainfall trends was determined. Figure 3 below summarizes the procedure.

Estimated mean annual sediment generation varied considerably in magnitude, but showed no significant spatial variability. Worrying levels are confined to the upper catchment (Figure 4) with most of the areas having an estimated sediment potential of over 25 tonnes per hectare per year (t ha⁻¹yr⁻¹). With high topographic percent rises in the upper catchment and extreme magnitude variability of the rainfall both in space and time, infrequent but heavy rains tend to be responsible for moving much of the sediment generated in the tropical catchment.

It is important to verify whether the assumptions made this far are reasonable and whether they adequately reflect the essential features and behavior of the real system. In modeling, it is important that input variables can be satisfactorily measured in the field or otherwise and, as such, PESERA avoided difficult variables, such as the zone depth DF, as much as possible in the ANSWERS model. Though developed for the temperate regions, the strength of its hydrological base and flexibility of the model code for corrections and adaptation allowed the PESERA model to be satisfactorily fitted to the tropical environment for which factors like snow fall and melt are non-existent. With much of the data on the distributions of the parameters held in raster-based GIS formats, the smooth integration of PESERA to raster GIS proved very useful in this analysis apart from providing the detailed spatial output displays that are useful to policy makers.

However, a limited number of measured values were used. This necessitated interpolation to fill the entire study area and it is understood that this may add uncertainties to the measured values due to the interpolation scheme. Uncertainties are also introduced in the choice of derived vegetation cover for each month that was based on the author's knowledge and experience of the study area given the lack of available monthly land cover data. Again, the data requirements for PESERA are substantially large to suffice the simulation process. While maintaining to be as physically based as possible, some empirical relations were inevitable (the Hargreaves and TopModel models) in deriving some of the parameters.

Sensitivity analysis demands alteration of each individual parameter and observation of the model output. However, with coded values in the land use and soil parameters (Table 1), increasing or decreasing such values by some factor would render them meaningless to the model. Again the substantial computation and time requirements hindered alteration of each of the climatic parameters to observe its impact on the output. However, assuming 20% [15,27] of the generated sediment is deliverable to the proposed dam site (see Figure 2), the model response was tested for the real time processes of fire, plant gathering and grazing. This was considered since cut-and-burn agricultural practice has intensified [5] recently due to the demand for more agricultural land [4]. Figure 5 compares the control 5% vegetative cover reduction against the model estimates at 30% vegetative cover reduction. The model responds well to changes in vegetative cover and, as shown, it is sensitive to the same.

The monthly distribution of the sediment deliverable to the reservoir follows the rainy season (Figure 5). Observed sediment concentration data for Mwandenga, a gauging station in the lower sub basin just below the proposed reservoir site, is shown in Table 2. At an average annual maximum discharge of 713.92 m³/s [8] the observed data was translated into tonnes per year. No vegetative cover loss (0%) would have been an ideal control but that would not have been realistic. Though the model has under-estimated the sediment yield (see Figure 5), the results compare very well considering that the observed sediment yield translation used observed monthly maximum discharge values. The actual observed sediment is expected to be less than depicted, thereby falling closer to the model estimates. Despite some observed sediment concentration data missing for the dry months (June to November), the available data sufficed in this analysis as it was for the critical window period—the rainy season. The model estimates monthly series translate into an average annual estimated sediment input of 9,000,000 t yr⁻¹, an equivalent of 3.4 Mm³ yr⁻¹, which raises concerns of sedimentation. Assessed against Malawi's largest 19 Mm³ Kamuzu II dam, a catchment size of 1,800 km² (approximately half that of Songwe), it would represent a substantial 7.5% annual reduction of its current capacity. These results point to a high sedimentation risk facing the life expectancies of any reservoirs planned for construction in the catchment, for flood control or otherwise, and the results compare well with relative sedimentation estimates of similar tropical catchments of India and Philippines [18].

Sustainable watershed rehabilitation and management does not only depend on quantified magnitudes of sediment generated but also on awareness of the sediment residence time in the catchment [18]. The drainage density was calculated to be very low at 0.1126, signifying that the available sediment in the watershed has a long residence time. This makes the sediment generated in the catchment available for transport even after further erosion has been prevented upstream.

Furthermore, the determination of whether sediments are from human or geological origin is of particular importance to reservoir capacity and storage sustainability [18]. This determines the expected type of sediment, large from landslides and/or fine from agricultural and built up lands, and provides for a focused catchment conservation strategy.

Dams are built with a provision for a portion of the reservoir to hold incoming sediment loads projected to occur during the economic life of the dam, referred to as “dead storage” [2,32]. This storage is in the deepest portion of the reservoir, usually close to the dam, between the level of the water intake and the streambed. “Live storage”, on the other hand, refers to the stored water that will actually be used [18]. While fine sediment is carried as suspended load farther into the reservoir and settles into the planned dead storage area, coarse sediment is carried as bed load and is deposited at the reservoir inlet taking up live storage area [33].

For the study area, a large proportion of the sediment originates from built up village land (Figure 6). Dunne [34] observed that, despite little study of paths and village settlements, for which the estimates of actual land area covered tend to lack, erosion from these features generate a large part of the sediment yield of catchments. An increase in the population density [5] has made the network of roads and paths denser and the area covered by the settlements larger in the study area. In the upper and middle sub-basins, the networks traverse steeper gradients of the terrain. Generating fine and/or coarse sediments— depending on the prevailing catchment conditions—the combined contribution from built-up village land is significantly over half of that in the study area, as observed in similar tropical watersheds elsewhere [18].

Reported intensive cut-and-burn agriculture [5] has left many natural environments degraded. The tropical storm rainfall conditions subject the degraded natural land to mass wasting, which ends up constituting a large proportion of the stream channel's bed load sediment. This sediment load poses greatest risk to the life expectancy of any reservoir constructed in the catchment as it usually occupies live storage space.

With many of catchment conservation strategies focusing on addressing agricultural induced erosion, the evaluation of the catchment land use above suggests efforts should be also concentrated otherwise for significant reduction in reservoir sedimentation. The little, though significant correlation (0.455 at the 0.01 significance level) between the increasing rainfall events and observed discharge levels causing the floods in the floodplain, shows that there is more to the increased flood occurrences in the catchment than just the increased rainfall. [4] reported continued land degradation and the sedimentation analysis above highlights substantial degradation.

Reduction of sedimentation is limited if the control measures are implemented in non-principal origins [18]. Again, with limited funds and technical help in the tropical catchment area, conservation efforts need to be directed where they will have the greatest impact. Figure 7 highlights the areas generating high levels of sediment, of which if controlled would have significant impact.

Though fraught with limitations even in the best of circumstances, this identification of sources of sediment serves to avoid guesswork in planning the management programs for sedimentation control in the watershed. It also develops awareness on whether the sources are amenable and whether the sedimentation control approaches can produce verifiable results. Figure 7 (a) shows the critical sediment sources within the built up village and arable land, quantitatively and for prioritization purposes classified into conservation requirements of high, medium and low (using quantile method). Similarly, Figure 7 (b) shows those from degraded natural land. Choice of the two categories was based on not only the fact that they are the dominant land use types in the area (see Figure 6), but more importantly, also that they have compromised relative roles in controlling sedimentation as they are disturbed land areas. Beyond the scope of this visualization, however, is an anticipated situation in which the locations of the engineering work of the dams results into generation of significant amounts of sediment. Two- to three-fold increases in estimated sedimentation from landslides triggered by roads cut into the slopes and borrow pits located upstream during construction has been reported [35]. Thus, poor planning and location of borrow pits and roads during construction would render conservation efforts fruitless.