As part of a broader study of landscape change and ecosystem function, we examined rangeland processes in the Lampasas Cut Plain of central Texas, USA. This savanna landscape is characterized by low buttes and mesas separated by broad, flat valleys. Local prevailing geology is Cretaceous limestone; soils are loamy and clayey, with occasional sandy loams, and are susceptible to sheet and gully erosion (Allison, 1991; Clower, 1980). The area is drained by the Lampasas River. Streamflow in the upper reaches of the river is runoff dominated, with localized contributions from spring flow (Prcin et al., 2013), and has been recorded at two primary stations (Fig. 1). Annual precipitation averages approximately 800 mm, decreasing to the north and west (Fig. 2). Winter mean temperature is around 7 ∘C and in summer 27 ∘C.

For the sediment study, we examined eight flood-control reservoirs and their watersheds within the Lampasas River basin. Reservoirs L1, L2, L3, L4, L9, and LX are located in Lampasas County and were constructed between 1958 and 1961. Before impoundment, the parallel watersheds of L1, L2, and L3, contributed to the downstream watershed of LX. Reservoirs M1 and M4, in Mills County, were completed in 1974. Basic attributes of the reservoirs and their watersheds are compiled in Table 1.

Current local land use is predominantly rangeland, and livestock numbers have fluctuated over the last several decades (Fig. 3a) while remaining among the highest in the region (Wilcox et al., 2012a). Cropland was widespread early in the 20th century (Fig. 3b) but had declined by nearly 80 % by 2012 (Berg et al., 2016). Amid this shifting land use, the area has been characterized by large fluctuations in the extent of woody plant cover, due to brush management and regrowth (Fig. 3c), and a dramatic increase in the density of farm ponds (Fig. 3d) over the last several decades (Berg et al., 2015a).

To investigate local hydrological trends, we analyzed historical precipitation and streamflow data for the Lampasas River basin. We created a composite record of annual precipitation using a Thiessen polygon approach, centering polygons on available National Weather Service (NWS) stations (Fig. 2). Daily streamflow data were derived from the two USGS (US Geological Survey) stream gage stations downstream from the study watersheds. The lower Youngsport station, with a drainage area of 3212 km2, operated between 1924 and 1980; the Kempner station, with a drainage area of 2119 km2 has remained active from 1963 to the present.

We performed an automated base-flow separation of streamflow data from each station (Arnold and Allen, 1999). This digital filter approach is objective and reproducible and partitions annual base flow and storm flow with high efficiency (Arnold et al., 1995) – enabling these components to be interpreted in light of changing landscape conditions.

Using the precipitation (1924–2010) and two streamflow data sets (1924–1980; 1963–2010), we applied a non-parametric Mann–Kendall trend test (Lettenmaier et al., 1994) to detect directional changes in precipitation, total streamflow, base flow, and storm flow. We performed two-tailed statistical tests for significance, with α = 0.10.

To shed light on sediment transport processes, we extracted cores from each of the eight reservoirs and analyzed sediments using cesium-137 (137Cs) and lead-210 (210Pb) tracers. 137Cs is present in the environment as a result of atomic weapons testing and accidental emissions. 210Pb occurs naturally. Both can be used to estimate sedimentation rates and interpret transport history in a variety of environments (Walling et al., 2003; Ritchie and McHenry, 1990; Appleby and Oldfield, 1978). Coring sites were selected by locating the thickest sediment deposits through exploratory hydroacoustic surveys (US Army Corps of Engineers, 1989, 2013; Dunbar et al., 2002). In each reservoir, we extracted sediment cores at identified sites near the dam structure, from locations corresponding to the pre-impoundment floodplain (Fig. 4). Taking cores from these areas reduces the likelihood of capturing mixed profiles, which skew analysis (Sanchez-Cabeza and Ruiz-Fernández, 2012). It also ensures the collection of fine sediments, to which the radioisotopes preferentially adsorb (Bennett et al., 2002). We extracted cores using a portable vibracoring system suspended from a floating platform. This method captures unconsolidated, saturated sediments with minimal disturbance and compaction (Lanesky et al., 1979). The cores were collected with an aluminum pipe lowered to the point of refusal, penetrating the pre-impoundment surface. Retrieved cores were sealed and transported upright to cold storage (∼ 5 ∘C).

We sectioned each core vertically in 3 cm intervals, drying each section for analysis according to IAEA (2003) protocols. A subsample of each core section was ground to homogenize its contents, sealed in a 50 mm × 9 mm Petri dish, and allowed to ingrow for at least 21 days so that 210Pb supported levels reached equilibrium. Counts for 210Pb and 137Cs were performed according to Hanna et al. (2014) using a Canberra low-energy germanium gamma spectrometer. Radioisotope activity was indicated by photopeaks at 46 keV (total 210Pb) and 661.6 keV (137Cs). Excess 210Pb was calculated by subtracting the supported activity of the 226Ra parent – obtained by averaging the 295, 351.9, and 609.3 keV peaks of the 214Pb and 214Bi daughter products – from total measured 210Pb activity at the 46 keV peak. Activity measurements were validated with IAEA-300 standard reference material.

To determine historical linear sedimentation rates, we used as a chronological marker the depth of peak 137Cs activity (corresponding to the 1963 peak in global atmospheric fallout) (Ritchie et al., 1973). We calculated average linear sedimentation rates for the post-1963 period by dividing this depth by the time elapsed between 1963 and the coring date for each reservoir; we calculated the pre-1963 rates by dividing the depth of sediment below the activity peak by the time elapsed between reservoir impoundment and 1963.

To complement 137Cs analysis, we used excess 210Pb activities to calculate the linear sedimentation rate for each core (Krishnaswamy et al., 1971; Bierman et al., 1998). We also searched for changing deposition rates within each core, as plots of the natural log of excess 210Pb versus depth indicate stable sedimentation rates over time when R2 approaches 1.0.

Finally, we obtained historical annual Palmer Modified Drought Index (PMDI) data for the region to identify potential climatic drivers of sedimentation during different periods. We plotted PMDI and annual peak flows (from USGS data) between 1924 and 2010, identifying episodes conducive to increased sediment production (in particular, a wet year or years following a period of intense drought).

Despite a great deal of interannual variability, there was no directional change in local precipitation in 1924–1980 (p = 0.90) or 1963–2010 (p = 0.22), which has remained near a long-term average of 800 mm (Fig. 5a). The same is true of total streamflow (1924–1980: p = 0.98; 1963–2010: p = 0.34), which has averaged between 60 and 70 mm (Fig. 5b). As a result, the rainfall–runoff ratio, the proportion of rainfall leaving a watershed as streamflow, also remained unchanged, at approximately 8 % (1924–1980: p = 0.90, 1963–2010: p = 0.45). Moreover, neither base flow nor storm flow exhibited a directional change over either period of record. However, base flow as a proportion of total streamflow did increase 1924–1980 (p = 0.02) despite minimal change in overall flow – almost doubling its contribution (Fig. 5c).

Sediment core profiles varied widely in depth between reservoirs – from less than 3 cm in LX to 162 cm in L1 (Fig. 6). Activity peaks of 137Cs supported the analysis of pre-1963 sedimentation rates for reservoirs L1, L2, L3, and L9. Overall, linear sedimentation rates were higher before 1963 (Table 2; Fig. 7). Except in the case of L3, sediment deposition has slowed since 1963 – by 54 % in L1, 76 % in L2, and 84 % in L9. In reservoir L3, it increased by 49 % after 1963. Reservoir L1 exhibited the highest sedimentation rate both before and after 1963. However, when normalized by catchment area, sedimentation rates varied much more widely. The rates in L9 were by far the highest – surpassing the next highest reservoir by nearly 1400 % for the pre-1963 period and by 423 % for the post-1963 period.

Cores from L4, LX, M1, and M4 did not display a 137Cs peak. For these cores, sedimentation was assumed to be post-1963 and was estimated by dividing sediment depth by time since impoundment. For cores L4 and M4, which did not capture the entire sediment profile, actual rates are likely higher than those calculated.

Cores from reservoirs LX and M1 showed vertical mixing that prohibited 210Pb analysis. However, remaining cores displayed high correlation between 210Pb activities and depth, indicating that linear sedimentation rates have remained quite stable over time (Table 2). 210Pb-based estimates generally resembled those based on 137Cs activities. In addition, rates calculated from 210Pb activities were similar to the post-1963 rates based on 137Cs activities (p = 0.84), suggesting good agreement between the two methods for the period since 1963.

Chronological data revealed periods of drought of varying intensity and occasional years of very high streamflow (Fig. 8). The historic 1950s drought was longer and more severe than any other over the last century; it was followed by periods of very high flow in 1957 and 1960. Comparable high flows in 1965 occurred in the middle of a multi-year drought, and the severe drought beginning in 2006 featured occasional elevated peak flows. In 1992, very high flows occurred during a prolonged wet period.

Given the varying trends in precipitation and streamflow observed in many regions (Lins and Slack, 1999; Andreadis and Lettenmaier, 2006), the dynamic hydrological stability in our study area is surprising. At the same time, such consistency sheds light on the effects of watershed changes on local water budgets. Studies at small spatial scales frequently indicate that landscape changes have important water resource impacts, with the specific response depending on the relative importance of evapotranspiration, recharge, and runoff (Foley et al., 2005; Kim and Jackson, 2012). Such changes affect local water budgets and influence water yields (Petersen and Stringham, 2008; Huxman et al., 2005; Farley et al., 2005). However, complicated feedbacks make effects at larger scales highly uncertain and often overwhelmed by climatic and physical characteristics (Peel, 2009; Wilcox et al., 2006; Kuhn et al., 2007). Our rainfall–runoff ratio of 8 % is essentially identical to early estimates of 7 % for the area (Tanner, 1937). The lack of a directional trend in streamflows suggests that this region, like many semiarid landscapes dominated by surface runoff, is largely hydrologically insensitive to shifting watershed characteristics (Wilcox, 2002). Perceived impacts due to changing rooting depths, longer growing seasons for evergreen woody plant species, and assumptions of very high shrub transpiration capacities are not borne out. Changes in land use and land cover – and even the impoundment of small reservoirs – have had negligible impacts on streamflow. These results confirm and add new insight to other research showing that woody plants in this region are shallow rooted and do not rely on deeper, perennial water sources (Heilman, 2009; Schwartz et al., 2013; Schwinning, 2008).

It is still not understood why base flow showed a proportional increase 1924–1980. In some landscapes, improving range conditions have led to increased infiltration (Wilcox and Huang, 2010). However, local livestock numbers have remained high, and karst features are limited – unlike other regions where base flow increases have been attributed to rangeland recovery. It is possible that infiltration from local impoundments has added to base flows. Despite minimal effects on total streamflow, even small dams can create localized groundwater recharge (Graf, 1999; Smith et al., 2002), and Lampasas River tributaries are characterized by a high degree of connectivity between surface water and local aquifers (Mills and Rawson, 1965).

Perennial flow in this part of the Lampasas River is maintained by isolated springs fed by an aquifer extending beyond the basin (Mills and Rawson, 1965). As a result, the effective catchment of the river is larger than it appears, and spring flow contributions complicate the interpretation of streamflows. At the same time, it is clear that the fundamental relationship between rainfall and streamflow has not changed over more than 85 years – suggesting that the Lampasas River is hydrologically resilient in the face of changing land use and land cover.

Because sediment deposition affects reservoir storage and flood detention, understanding sedimentation rates over time is critical to managing rangeland water resources. Though questions do remain regarding the opposing trend in reservoir L3, changes in rates make it clear that sedimentation was more rapid before 1963. The period since that time has been characterized by stable and lower yields. What explains the higher rates seen during the earlier period? Additional historical landscape data may offer a key interpretive lens.

Livestock can be powerful instruments of landscape change, both directly (trampling soils) and indirectly (disturbing protective vegetation). When grazing is prolonged or intense, sediment yield can be great (Trimble and Mendel, 1995). The high animal densities in this area around the time of reservoir impoundment doubtlessly contributed to erosion (Fig. 3a).

Crop production also can result in accelerated erosion by damaging soil structure and depleting organic matter (Quine et al., 1999). Cropland is a major source of sediment in many landscapes (Foster and Lees, 1999; Blake et al., 2012). In our study area, cropland acreage has declined dramatically since the 1930s (Fig. 3b). Further, nationwide improvements in soil conservation have reduced sediment yield from many agricultural lands (Knox, 2001).

While woody plant encroachment influences soil loss, removing undesirable shrubs and trees also elevates short-term sediment yields (Porto et al., 2009). Since the time of initial settlement, woody plant management has resulted in major land cover changes (Fig. 3c). Most early removal was done manually, and the first mechanical control methods were very destructive, leading to high erosion rates (Hamilton and Hanselka, 2004). In recent decades, however, brush removal has declined with shifting landowner priorities (Sorice et al., 2014).

Changes in precipitation frequency, duration, or intensity also affect sediment transport (Xie et al., 2002; Allen et al., 2011). Similarly, drought is an important driver of sediment dynamics in many rangelands. Extended dry periods can cause long-term shifts in plant cover, leading to sediment pulses when rains return (Allen and Breshears, 1998; Nearing et al., 2007). The Lampasas River experienced very high flows in 1957, 1960, 1965, and 1992, and some of these were associated in time with severe droughts (Fig. 8). Just before the impoundment of most of the reservoirs we examined, the region was in the grip of drought conditions unmatched since European settlement (Bradley and Malstaff, 2004). Our sediment records cover only the end of this drought but show pre-1963 deposition 220–630 % faster than subsequent rates. However, any direct effects of the 1957 drought-breaking floods would not be found in the sediments of the reservoirs, which were impounded beginning in 1958. Interestingly, we also did not find spikes in sedimentation associated with high flows or droughts later in the study period. The apparent low importance of drought and floods in sediment delivery in these watersheds is surprising.

Together, these factors have acted over multiple temporal and spatial scales to influence sediment yields in the study area. Yet because there is no clear link between contemporary land use, land cover, and sedimentation rates, it is possible that another process has reduced sediment yields.

To truly understand the local sediment processes at work, it is important to understand what our findings actually show. Sedimentation rates are poor indicators of in-field soil erosion and redistribution (Nearing et al., 2000; Ritchie et al., 2009); what they do reflect is more closely related to net watershed sediment yield. Sediment yield is buffered by internal storage. Especially at larger scales, watersheds can have a great deal of internal storage, so that very little eroded soil actually leaves the watershed, even in the presence of extreme erosion (Bennett et al., 2005; Porto et al., 2011).

In this study area, the increasing density of farm ponds (Fig. 3d) represents a key potential sink for watershed sediments. These ponds – usually < 0.3 ha when full – retain material that otherwise would be transported downstream, reducing sediment yields. Because of their smaller contributing watersheds, ponds have high trap efficiencies, magnifying their effects (Brainard and Fairchild, 2012). Indeed, impoundments may be the single greatest anthropogenic modifier of sediment transport; globally, most sedimentation now takes place in aquatic settings and will be retained therein for long periods (Renwick et al., 2005; Verstraeten et al., 2006).

In addition to this storage of eroded sediments in local ponds, a vast amount of sediment from past erosion likely remains on the landscape (Beach, 1994; Meade, 1982). The initial decades after European settlement in this area saw intensive cultivation and very high livestock densities (Jordan-Bychkov et al., 1984; Wilcox et al., 2012a). This destructive combination remained in place for nearly a century in the Lampasas Cut Plain. By the 1930s, many rangelands were already seriously degraded (Mitchell, 2000; Bentley, 1898; Box, 1967). While the methods we used do not allow us to determine whether reservoir sediments result from contemporary erosion or are a legacy of earlier land use, stabilizing sediment yields and observations of local gully erosion suggest that deposits from prior erosion continue to be a source of sediment (Bartley et al., 2007; Mukundan et al., 2011; Phillips, 2003).

The lack of sediments in LX appears to lend support to the importance of internal deposits. This reservoir's watershed is comparable in size to those of L2, L3, and M4, yet sedimentation rates were only 3–14 % of those in the other reservoirs. When L1, L2, and L3 were impounded, the effective catchment area of LX decreased by 86 %. Without the historical streamflows and sediment loads from those tributaries, deposits are no longer mobilized and transported downstream.

Given this complexity, we suggest that radioisotope tracers have great potential to elucidate the dynamics of rangeland systems, particularly as their use evolves from primarily research applications to use as a management and decision-support tool (Mukundan et al., 2012). Further strides can be made in understanding rangeland processes by (1) incorporating historical climate, land use, and land cover information to interpret sediment data (Venteris et al., 2004; Boardman, 2006) and (2) including sediment surveys of the farm ponds that are much smaller yet far more abundant than the reservoirs we examined (Downing et al., 2006).