Patch dieback of Colophospermum mopane in a dysfunctional semi-arid African savanna

Study area and study species

Venetia-Limpopo Nature Reserve (VLNR; 29°12'−29°23'E, 22°15'−22°30'S) is situated in the Northen Province, South Africa, approximately 30 km south of the Limpopo River, and is 34 500 ha in extent. This semi-arid region is characterized by wet, hot summers and dry, mild winters. Mean monthly minimum and maximum temperatures are, respectively, 20.3 and 32.0°C for summer (December) and 7.2 and 24.7°C for winter (June). The mean annual rainfall for Pontdrift, approximately 15 km north of the study site, is 366 mm (1967–1997). Rainfall, consisting mainly of thunder showers, falls primarily from October to March Annual rainfall over the past few decades has been erratic (36% coefficient of variation) with extended periods of above- and below-average rainfall (Fig. 1). There were severe droughts recently in 1982–1983, 1988–1989 and 1991–1992. Mean annual evaporation (2000–2200 mm; Midgley et al. 1994) is sixfold greater than the mean annual rainfall.

The study area is a basin of deep (>5 m) palaeo-alluvia derived mostly from Karoo sediments with more recent alluvia occurring adjacent to current seasonal rivers. Soil forms represented are Swartland, Valsrivier, Oakleaf and Dundee (Soil Classification Working Group 1991) with an average content of 20% clay and 19% silt (Botha 1994). Sheet and rill erosion is conspicuous because of the poor perennial grass cover (O'Connor 1999).

Livestock ranching at this site commenced during the 1940s. Patterns of soil and vegetation degradation (T. G. O'Connor, unpubl. data) suggest levels of stocking were heavy, but quantitative information is not available. Livestock were removed in 1991 when the area was consolidated as a nature reserve, after which the numbers of indigenous herbivores increased.

The study area is part of the Mopani Veld type (Acocks 1953). Colophospermum mopane is the dominant species of all the dryland vegetation types of VLNR (O'Connor 1992). Other common species include Boscia foetida, Salvadora angustifolia and Lycium austrinum (nomenclature follows Arnold and De Wet 1993).

Colophospermum mopane is a single- or multistemmed tree or shrub, usually <5 m in height, which occurs as in dryland situations in VLNR. The C. mopane woodlands of VLNR are dense, with >70% cover (O'Connor 1992). Characteristics of C. mopane include an affinity for heavier textured soils (Fraser et al. 1987), drought tolerance (Viljoen 1995; Johnson et al. 1996) and an ability to utilize soil water at a matric potential lower than that of herbaceous plants (Smit 1994). Severe intertree competition for water occurs in dense stands (Smit & Rethman 1998a). This species has a large root biomass and shallow root system (Smit & Rethman 1998b), and is physiologically adapted to xeric conditions (Prior 1991). Following topkill, an individual can coppice from the stem base. Hence, individual plant longevity is unknown. However, fire scars, attributed to the last recorded fire in 1948, indicate that most of the individual trees of VLNR are greater than 50 years old and thus were established prior to agricultural development (S. D. MacGregor, unpubl. data).

Data collection

A comparative approach was employed in 1998 in which 20 m × 20 m plots within 40 areas of discrete dieback (all <1 ha) were compared with adjacent paired plots, 20–150 m away (median distance 45 m), of ‘healthy’ vegetation on apparently similar topography and soils. The 40 dieback patches were distributed over the topography from crest to alluvial toeslope (Fig. 2). In each plot, we measured the size structure of plants, stem and tree mortality, and soil structure and chemistry. An inherent problem of a retrospective study of an event such as dieback is obtaining measures of both vegetation and potential causal agents of tree mortality that reflect circumstances as they were at the time of dieback. Differences between paired plots for some variables could arise as a response to mortality rather than indicating causes of mortality.

All stands appeared healthy following the 1982–1983 drought, but dieback appeared by 1991 following droughts in 1988–1989 and 1991–1992 and was obvious by 1996 (T. G. O'Connor, pers. obs.). The decay of standing dead trees (lack of bark shed, amount of fine branches lost) was consistent with mortality occurring in 1991–1992.

We recorded woody vegetation species, and measured height to the uppermost living and dead part of a tree, stem circumference (20 cm above ground level) and growth state (dead or alive) of each stem per tree. Dieback per plot is expressed as the ratio of dead to total basal area. Dieback for all ‘live’ plots collectively or all ‘dead’ plots collectively is expressed as the number of trees with dead stems or the number of dead stems for each class of stem number per tree.

Averages of surface water retention capacity (SWRC), erosion severity (ES), and ecological status (Est) were calculated from the mean of the four 10 m × 10 m quadrats in each 20 m × 20 m plot (Appendix 1). Surface water retention capacity and ES were adapted from Tongway and Hindley (1995) for the present study, using a 12- and seven-point ordinal scale, respectively (Appendix I). The Est describes whether a plot functions as a net source or sink of sediment and water, taking into account geomorphic processes and topographic position (Appendix I). Past and current ecological status were assessed using a seven-point ordinal scale, focusing on the flow of sediment and water across the plot prior to and since the advent of rill and gully erosion, respectively. Change in ecological status was the difference between past and current rankings. Aerial photography (1955–1987) revealed that gully erosion associated with the dieback sites preceded the death of trees but had occurred after 1955, that is, after the establishment of trees. An increase in SWRC occurs through the effect of litter (Kelly & Walker 1976) and from the assumed effect of sedimented rills storing water gained from upslope and transferring it to the adjacent profile, as opposed to water being shed from a compacted surface. These were included in the ranking procedure. The exposure of the surface of ‘dead’ plots since tree mortality may have resulted in exaggerated ranks for present Est, but this bias was consistent for change in Est.

Slope was estimated using an electronic level (Leica Wild 3000). Soil physical and chemical characteristics of both topsoil (0–200 mm) and subsoil (400–600 mm) were evaluated for each plot. Soil texture was analysed using the hydrometer method (Gee & Bauder 1986), pH (H₂O) was measured using a glass electrode pH meter (Crison micropH 2000), and the exchangeable cations (Na⁺, Ca⁺⁺, Mg⁺⁺ and K⁺) were determined by extraction with strontium (using 0.1 m strontium chloride) (Hughes & Girdlestone 1994) and expressed in meq/L^-1. The sodium adsorption ratio (SAR) was calculated using the equation from United States Salinity Laboratory Staff (1954) (Na/sqrt(Ca/2 + Mg/2)). The sodium adsorption ratio gives an indication of soil sodicity and thus the potential for physiologically induced drought. Total exchangeable cations (TEC) was calculated as the sum of the concentrations of Na, Ca, Mg and K.

Data analysis

Biotic and abiotic variables in ‘live’ and ‘dead’ plots were compared using paired t-tests. Tree height distributions were compared using a Komolgorov–Smirnov test. Differences in variables among soil types or topographic units were examined using Kruskal–Wallis or parametetric anova, depending on data properties, with ‘live’ and ‘dead’ plots analysed separately. A logistic regression was performed to test whether the probability of mortality on ‘dead’ plots depended on stem circumference, soil type or their interaction.

To assess whether dieback had changed the nature of self-thinning (log of mean plant weight and log of plant density), the relationship between individual plant size (volume) and stem density was examined. Because of the multistemmed growth habit of C. mopane, density was expressed as the number of stems per unit area. An index of plant weight was calculated from the volume of a sphere, using stem radius, because heights for individual stems of multistemmed trees were not available. We used the radius of trees that had died some years earlier because direct measures had shown radial growth of live trees was minimal over this period (O'Connor 1999). A weight correction was applied by multiplying by the wood density of C. mopane (1.25 g cm⁻³) (Venter & Venter 1996).

The relationship between plot dieback, expressed as the percent loss of basal area (square-root transformed as indicated by a Box–Cox analysis), and those independent variables found to be significantly different between ‘live’ and ‘dead’ plots, was examined with multiple linear regression. The standardized variables were mean tree height, change in ecological status, SWRC, ES, pH (subsoil), texture (topsoil), SAR (subsoil), TEC (subsoil), number of stems, basal area of C. mopane and basal area of Salvadora angustifolia. Because 66% of all fine (<5 mm) C. mopane roots occur in the top 400 mm of the soil profile (Smit & Rethman 1998b) in this study area, we considered cation concentrations relevant to dieback would be encountered in this layer. Thus subsoil values were included in the regression. Variables were entered one by one into the regression model according to their contribution to R² (adjusted) and reduction in s² at each step. Only main effects (P < 0.05) were included in the final regression as there was no evidence of interactions. One site located in riparian vegetation exerted undue leverage on the model and was excluded.

Patterns of dieback

Dieback of stem basal area ranged from 1 to 49% in live plots and from 55 to 100% in dead plots. Mean dieback was 88 ± 1.6% on ‘dead’ plots, but only 14 ± 1.9% on ‘live’ plots. The three woody species on the plots other than C. mopane made up only 5% of woody basal area.

There were 1–16 stems per tree across all plots. One third of the trees had single stems, 50% had two or three stems, and <2% had more than six stems. On ‘live’ plots, 27% of trees had dead stems, and 17% of stems were dead. On ‘dead’ plots, 89% of trees had dead stems and 86% of stems had died. Mostly entire trees died on ‘dead’ plots, whereas single-stem dieback prevailed on ‘live’ plots. Tree mortality (death of all stems) decreased with increasing stem number per tree (Table 1).

Abiotic variables influencing dieback

There was little variation in mean dieback on the upper, mid and foot slopes on live (12–14%) or dead (86–93%) plots. Dieback did not differ among topographic units on ‘live’ (P = 0.722), ‘dead’ (P = 0.449) or all plots (P = 0.986). Nor did dieback differ among soil types on ‘live’ (P = 0.438), ‘dead’ (P = 0.434) or all plots (P = 0.998) Means of variables from 'live' and 'dead plots' are shown in Table 2. Slope did not differ between ‘live’ and ‘dead’ plots.

‘Live’ and ‘dead’ plots did not differ in past Est, suggesting appropriate criteria were used for pairing of sites, but they differed markedly in current Est (Table 2). Plots were originally functioning, on average, as transfer zones with a slight tendency to act as sinks of water and sediment. The current average Est for ‘live’ plots was similarly positive, but ‘dead’ plots functioned as a transfer zone tending to a source. The change in Est reveals that both ‘live’ and ‘dead’ plots have, on average, become more dysfunctional, but the extent of change on the ‘dead’ plots was three times that of the ‘live’ plots (Table 2).

Erosion severity was greater and SWRC was lower on ‘dead’ than on ‘live’ plots (Table 2). Although negatively correlated (r = −0.62), the relationship between SWRC and dieback appears to be a threshold response with some very high incidence of dieback at SWRC < 2.5, but only low levels of dieback at SWRC > 2.5 (Fig. 3). The greater SWRC of ‘live’ plots correlated positively with current Est (r = 0.73). This underscores their functioning as sinks. The greater proportion of clay plus silt particles on ‘dead’ plots would have further reduced water availability during drought.

Differences in soil chemistry (pH, TEC) between ‘live’ and ‘dead’ plots (Table 2) were not large, possibly resulting from a change in nutrient cycling following mortality of C. mopane trees on ‘dead’ plots. The variables pH, TEC, and SAR were not correlated with textural variation. Individual cations (Ca, Mg and K) were not at significantly different concentrations in ‘dead’ and ‘live’ plots.

Role of competition

There were significant negative correlations between the log of mean stem ‘weight’ and the log of stem density on ‘live’ (r² = 75.9, P < 0.001) and ‘dead’ (r² = 68.6, P < 0.001) plots prior to dieback (Fig. 4a). The slopes did not differ (P > 0.5), suggesting consistent self-thinning. However, there was a negative correlation of percentage dieback with total basal area (r = −0.51) and stem density (r = −0.47) of C. mopane, and there was a greater stem density and total basal area on ‘live’ than on ‘dead’ plots (Table 2). This argues against a role of intraspecific competition in dieback. The self-thinning relationship was still evident on ‘live’ plots after dieback (r² = 72.2, P < 0.001) (Fig. 4b) and the slope of this relationship did not differ (P > 0.5) from that before mortality. In contrast, no self-thinning relationship was apparent on ‘dead’ plots after dieback (r² = 5.9, P= 0.08). Change in the log of mean stem ‘weight’ on ‘live’ plots (‘weight’ of [live plus dead stems] − [live stems]) was negligible (Fig. 4c) and remained constant (b₁ = 0.015), irrespective of stem density (P = 0.665). This negligible change probably results from low stem mortality. On ‘dead’ plots, however, there was a significant negative (b₁ = −0.868) linear relationship between change in the log of mean stem ‘weight’ and the log of stem density (r² = 11.0, P < 0.027), indicating a greater mean individual size after dieback at lower stem densities.

The combined stem basal area of the other three tree species did not differ between ‘live’ and ‘dead’ plots (Table 2). Thus interspecific competition is an unlikely explanation for mortality. Although S. angustifolia had significantly greater stem basal area on ‘dead’ plots (Table 2), it was absent on 38% of plots. However, there was a correlation (r = 0.3, P < 0.01) between its stem basal area and percent dieback of C. mopane, so interspecific competition with this species cannot be eliminated as a contributing factor.

Population structure and dieback

For C. mopane, mean stem circumference of live and dead trees was similar on 'live' and 'dead' plots (Table 2). Trees were significantly taller on ‘live’ than on ‘dead’ plots. However, height may have little ecological significance in explaining dieback (Fig. 5). The tree height distribution of dead trees on ‘dead’ plots was not different (P > 0.10) from the distribution of all trees on those plots prior to dieback. Within plots there was no tendency of trees of different heights to suffer different levels of mortality.

Influence of soil type

Vegetation structure and tree size depended on soil type. The distribution of tree height differed among soil types on both ‘live’ and ‘dead’ plots, whereas basal area differed only among ‘live’ plots. No differences were evident for stem density, mean tree height, and mean stem circumference.

Soil types differed in change in Est on ‘live’ (P = 0.006) and ‘dead’ (P < 0.01) plots, ES (P = 0.003) on ‘live’ plots, and SWRC on ‘live’ (P = 0.034) and ‘dead’ (P = 0.048) plots. These differences may be attributed to differences in texture on ‘live’ (P = 0.023) or ‘dead’ (P = 0.013) plots. Dundee and Oakleaf soil forms had a greater proportion of sand than Valsrivier and Swartland soils. Sandier soils had a lower ES, a greater SWRC and underwent less change in Est, suggesting that fine-textured soils are more susceptible to degradation and patch dieback. The probability of stem mortality on ‘dead’ plots was affected by soil type (P < 0.001), but not by stem basal area or their interaction (P > 0.05). Trees growing on Oakleaf soils were less inclined to lose stems than trees growing on the other soil types.

Regression model of dieback

The regression model of dieback was successful, explaining 70% of the variation (Table 3). Variation among plots in dieback was influenced mostly by SWRC and stem density, although ES, Est, vegetation height and pH had effects (Table 3). Less dieback occurred with a greater stem density, greater SWRC, least change in Est, lower pH, and on plots with smaller trees. The poor correlation between stem density and SWRC (r = 0.17) was improved (r = 0.53, P < 0.001) when ‘live’ plots containing rills were excluded from the analysis. This suggests an interaction between stem density and SWRC-influenced dieback. Although dieback was associated with decreased erosion, greater erosion severity was associated with the occurrence of sedimented rills, which were found only on ‘live’ plots (mean rill size: 25 cm deep by 115 cm wide). Sixty per cent of the 33 sites with rills had rills passing through the ‘live’ plots.

Preconditions and causes of patch dieback

Patch dieback is interpreted to have occurred where abiotic factors acted synergistically to reduce available soil water to a level incapable of sustaining the growth of C. mopane trees during the severe drought period of 1988–1992. Soil water availability on ‘dead’ plots was reduced by local and landscape features. For ‘dead’ compared with ‘live’ plots, the soil surface did not retain water as efficiently, areas receiving water and sediment changed to sources of these, and erosion was more severe. These variables all reflect a degraded soil surface and a subsequent shift in water availability beyond the threshold of stress of C. mopane (cf. Jimenez et al. 1985). Rills had consistently developed adjacent to ‘dead’ plots that would have redistributed run-on to rapid run-off, thereby precluding ponding and infiltration. Roads placed perpendicular to surface flow functioned as rills by intercepting and removing water to drainage lines. Patch dieback was therefore a symptom of a dysfunctional landscape (Tongway & Ludwig 1997) that became manifested during a drought period.

The evidence of self-thinning prior to dieback indicates C. mopane experienced intraspecific competition. This was also experimentally shown by tree-thinning trials conducted during the drought in this study area (Smit 1994). Competition, most likely for soil water (Smit 1994) was, however, judged not to be a cause of, but a precondition for patch dieback. Specifically, severe drought-induced dieback suggests intense competition during the drought on ‘dead’ plots that arose from their dysfunctional character.

The inverse relationship of dieback with stem density is not easily explained. Tree stems may have a similar effect to grass tussocks by intercepting rainfall, impeding surface water flow, and channelling water into the soil (Freudenberger et al. 1997). The high litter cover of these C. mopane woodlands (up to 42%; O'Connor 1999) is a critical control of infiltration and run-off (Kelly & Walker 1976) that may depend on woodland density. A lower woody density, therefore less litter cover (O'Connor 1999), would probably intercept less rainfall and expose more bare soil to compaction by rainfall (Finlayson & Statham 1980), thereby reducing infiltration and increasing run-off (Greene et al. 1994).

The hypothesis that patch dieback may be a consequence of the senescence of a cohort (Young & Lindsay 1988) was not supported, as the height distribution of dead trees on ‘dead’ plots covered the height range of live trees prior to dieback. Thus a savanna woodland of heterogenous rather than homogeneous structure was vulnerable to patch dieback. Synchronous dieback of different-sized trees suggests a species- rather than a size-specific threshold, possibly of soil water, was crossed. This emphasizes the susceptibility of monospecific compared with species-rich vegetation types to dieback.

The hypothesis that soil salinity may be implicated in dieback (Western & van Praet 1973) was refuted. Although soils were not sodic, the higher sodium concentrations on some ‘dead’ plots may have predisposed them to dieback through physiological ‘drought’ compounding natural drought. Dieback, however, occurred on sites across a range of Na concentrations.

Topo-edaphic variation in dieback

Although dieback did not differ among topography, slope or soil type, this study was restricted to dieback areas. However, C. mopane was also the dominant species on sandier soils derived from gneisses or aeolian sandstones (O'Connor 1992), on which no patch dieback was found. Sandy soils are not prone to the formation of rills and maintain infiltration, even when cover is poor (Barnes & Franklin 1970). At a broader landscape level therefore soil type, particularly texture and its relationship with SWRC, ES, and Est, was a key control on patch dieback.

Studies on rills have been concerned with water flow and have neglected to take into account their potential influence on vegetation. Rills are usually considered to promote a loss of water by converging flow of water and sediment from upslope along concentrated routes (Finlayson & Statham 1980). However, if rills become sedimented then the sediment can store water that can become available to adjacent trees.

Dieback: Manifestation of a dysfunctional landscape

Patch dieback of the prolific C. mopane signifies a dysfunctional system. The degradation of this system coincides with the 50-year period of commercial livestock ranching. Prior to ranching, the area formed part of a large, unfenced region occupied by wildlife. Aerial photography from the early time (1955) of agricultural occupation until 1987 provides diverse evidence of degradation. This includes the initiation and rapid expansion of areas of bare soil and gullying (T. G. O'Connor, unpubl. data), invasion of hydromorphic grassland by woody species (O'Connor 2001), a loss of perennial grasses and an increase in bare soil, especially on the fine-textured soils of the study area, which are vulnerable to sealing, run-off and erosion (O'Connor 1983, 1999), and a 20% increase in the cover of C. mopane woodland (O'Connor 1983). Most of these changes were precipitated by the severe, extended drought of the 1960s.

The occurrence of patch dieback of C. mopane between 1988 and 1992 is therefore interpreted as a consequence of the preconditioning of the system by an increase in C. mopane to levels of intense intertree competition (Smit 1994) and an increase in the drought proneness of fine soils through erosion, exacerbated in places by high Na concentrations. Water received during non-drought years was apparently sufficient for survival of these long-lived trees. Previous droughts (1968, 1973) did not result in patch dieback, apparently because the dieback areas were still functioning as sinks of water and sediment. Despite C. mopane's drought tolerance (Prior 1991; Smit 1994), its demand for water could not eventually be met on degrading patches, but could on patches functioning as run-on sinks.

A widespread model of rangeland degradation is an increase in the woody component with associated increases in soil erosion and water lost through run-off (Schlesinger et al. 1990). The present study indicates degradation can proceed another step in which previously increasing woody species suffer decline. On dieback patches then, productivity has been seriously impaired, carbon sequestration has become minimal, and soil and nutrient loss continue unabated. These are all indicators of an irreversible change in system functioning.