Summer rainfall zone precipitation variability

The key climatic events in the longest and highest-resolution proxies (Figure 3.2) will hereafter be discussed on a minimum of decadal timescales over the last 1200 years.

First, however, results from the correlation analysis will be assessed to further establish the proxies that are most likely to represent palaeoprecipitation.

3.2.1 Establishing proxy palaeoprecipitation

To understand the general correspondence between records and their potential precipitation inferences, three sets of correlations were performed. The first, shown in Table 3.2, reveals the correlations of the inter-annual records over the common period of 1036-1890. As expected, the T7 stalagmite δ¹⁸O, δ¹³C and grey-scale records demonstrate relatively strong agreement, and alone give little indication of the differential precipitation-temperature signals of each variable. It is potentially significant, however, that the T7 δ¹⁸O shows an increase in isotopic values in the latter half of the twentieth century (Figure 3.2), which is consistent with the nature of global temperature change since 1950. This direction of change is also observed in the T8 δ¹⁸O record, but not in the other T7/T8 variables, and may reaffirm previous studies that link stalagmite δ¹⁸O mainly to temperature but caution its use as a precipitation proxy.

TABLE 3.2 Inter-annual correlation values between proxy records over the common period of 1036-1890. Shaded values are significant at p <0.01. Note that the T7 grey scale record has an inverted scale, thus negative correlations represent co-variance. The Karkloof record is correlated from 1310-1890 and the Mashonaland record is correlated from 1796-1996.

T7 δ¹⁸O T7 δ¹³C T7 grey-scale Baobab Karkloof Mashonaland T7 δ¹⁸O -

T7 δ¹³C 0.45 -

T7 grey-scale -0.43 -0.35 -

Baobab 0.06 -0.02 -0.01 -

Karkloof 0.00 -0.05 0.10 -0.39 -

Mashonaland 0.07 0.05 -0.16 -0.05 0.19 -

No significant correlations are present between the T7 records and the Baobab record. 30-year running correlations (Figure 3.3) demonstrate that precipitation in certain periods does significantly co-vary between these records, but this is limited to multi-decadal periods such as in the early 1600s, and never for more than a century.

This may indicate that the Limpopo Baobab record is generally more illustrative of local precipitation variability, or could relate to either dating disparities or the influence of other climatic (or non-climatic) variables or processes on the stalagmite isotopes. To enable statistical comparison with the Karkloof record, the other four records were reduced to 7-year running means and correlated over the period 1320-1890. Where results are significant, they show inverse relationships between the

Baobab record (r = -0.39, p <0.01) and inferred precipitation in the grey-scale record (r

= 0.1, p <0.01, note the inverted scale) (Table 3.2). These inverse correlations, particularly between the Baobab and Karkloof record, are noteworthy, as one might expect a positive relationship between SRZ tree-ring records. It is therefore possible that as the Karkloof record has no supporting data from the same location, and because of the age of the dataset (Hall 1976), dating errors strongly influence the series (Tyson 1986). However, significant positive correlations are present between the Mashonaland tree-ring and the Karkloof record over their common 200-year correlation period, as well as between the Mashonaland tree-rings and the T7 grey scale record (Table 3.2).

FIGURE 3.3. Running (30-year) correlations between the Limpopo Baobab record and the T7 stalagmite proxies. Dashed lines indicate significance thresholds.

The second set of correlations, given in Table 3.3, assess lower-frequency coherence between the 30-year running means of these inter-annual records. Slightly stronger relationships are observed between the T7 records, yet no statistically significant correlations are present with the Limpopo Baobab record, which may again suggest that local variability is more important in this record.  

TABLE 3.3 Correlation of 30-year running means of inter-annual proxy records over the common period of 1036-1890. Shaded values are significant at p <0.01. Note that the T7 grey scale record has an inverted scale, thus negative correlations represent co-variance.

T7 δ¹⁸O T7 δ¹³C T7 grey-scale Baobab

T7 δ¹⁸O - 0.47 -0.49 0.05

T7 δ¹³C 0.47 - -0.45 0.02

T7 grey -0.49 -0.45 - -0.02

Baobab 0.05 0.02 -0.02 -

The final correlations test the relationships between the non-annually resolved proxies and the degraded annually-resolved records, as well as the association between the different non-annually resolved proxies from the same sites (Table 3.4). Most proxies from the same sites show significant and relatively strong correlations, the only exception being the T7 and T8 stalagmite records, where no significant correlations are found between T7 δ¹⁸O and T8 δ¹³C, T7 δ¹³C and T8 δ¹³C, and T7 grey-scale and T8 δ¹³C, the latter of which is marginally insignificant. The varying strength and significance of correlations of proxies from the same site reinforces the complexities of understanding precipitation signals in isotope records from speleothems, and restricts conclusions on the relative importance of each record for consideration of precipitation. Nevertheless, that the T7 grey-scale record demonstrates coherence with a number of other records further suggests that it is representative of regional precipitation. While each of the other T7/T8 records may be of use in establishing local and regional precipitation signals, caution will be applied due to the ambiguity of their signals. This is particularly relevant for the δ¹⁸O records, which have been suggested to be more representative of temperature in the literature (Stager et al. 2013; Sundqvist et al. 2013).

Correlations between the two Sibaya diatom records and the two Dante Cave speleothem isotope records are strong (r = -0.45 and r = 0.64, p <0.01 respectively).

TABLE 3.4 Correlations of non-annually resolved records with degraded inter-annual records over the common mean period of 1036-1890. Shaded values are significant at p <0.01. Note that the T7 grey scale and Sibaya conductivity records have inverted scales, thus negative correlations represent co-variance. The Mashonaland record is correlated from 1796-1996.

Table 3.4 also reveals two significant correlations between sites, including between the T7 grey-scale and Lake Sibaya conductivity record (r = 0.36, p <0.01), and the Limpopo Baobab and Dante Cave stalagmite δ¹⁸O record (r = 0.42, p <0.01) (Figure 3.4). This positive correlation of proxies from distant sites may signify that they are representative of wider precipitation variability across the SRZ. Nevertheless, the apparent high co-variance between the degraded Baobab record with Dante Cave δ¹⁸O may be a result of the uneven distribution of dating in the Dante Cave stalagmite, which has 50-year gaps in some places, and the resultant masking of important lower frequency variability in the Baobab record. This is particularly evident in the 1500s, where high inter-annual variability is observed in the Baobab record, yet only three dates are available in the Dante Cave stalagmites. This is reflected in the limited visual

T8 δ¹⁸O T8 δ¹³C Sibaya Intra-site T8 records: 0.30 Sibaya records: -0.45 Dante records: 0.64

T7 δ¹⁸O 0.4 0.19 0.09 -0.04 0.06 -0.11

T7 δ¹³C 0.09 0.01 -0.02 0.07 0.05 0.09

T7 grey-scale -0.26 -0.17 0.36 -0.13 0.05 0.09

Baobab 0.13 0.13 0.08 -0.18 0.42 0.1

Mashonaland -0.02 -0.01 -0.32 0.34 -0.26 -0.34

correspondence in panel (b) in Figure 3.4. While the dating of the Lake Sibaya diatom variables is also uneven, this is to a far lesser extent than the Dante Cave speleothem record, and hence its co-variance with the T7 grey-scale record is more reliable and visually clear (Panel (a), Figure 3.4). Limited statistical correspondence could also relate to the relative importance of local precipitation variability and the various rain-producing processes that operate in different parts of the region such as the ITCZ and the strength of the Indian Ocean winds. For instance, the Mashonaland tree-rings may lack significant correlations with other records due to the primary influence of the ITCZ in this area’s precipitation. This is reinforced in the positive but insignificant correlations with the ITCZ-sensitive Lake Sibaya record (r = 0.34).

FIGURE 3.4. Significant inter-site correlations between inter-annual and non-annually resolved proxy records.

Considering the literature and this initial analysis, the records to be used with higher confidence as palaeoprecipitation indicators are the T7 grey-scale series, the Baobab tree-rings, the Lake Sibaya conductivity and % planktonic records, and the Mashonaland tree-rings. Stalagmite δ¹⁸O is usually associated with temperature, and while stalagmite δ¹³C is conceptually linked to precipitation, poor inter-site correspondence suggests that a number of climatic or non-climatic variables influence both of these proxies. Caution will thus be applied when relating these proxies to rainfall. Similarly, caution will be used when considering the Karkloof record, as uncertainty arises from the reliance on a single specimen. With these concerns kept in mind, the next section analyses precipitation variability since AD800.

3.2.2 Precipitation variability since ^AD 800

Visual examination of multi-decadal scale variability between southern African proxy records over the last 1200 years reveals periods of general correspondence (Figure 3.5).

The direction of overall change in inferred precipitation conditions is negative, and is particularly strong in the Limpopo Baobab’s and both of the Cold Air Cave stalagmite δ¹³C records. This is in contrast to the Cold Air Cave δ¹⁸O records, which show either an increase in inferred warm-wet conditions or no overall change. At the beginning of the analysis period, relatively warmer and wetter conditions (relative to the 1036-1890 mean) appear to predominate in several records from c. AD 800-1200, possibly suggestive of a region-wide MCA, although considerable variability is found within this timeframe. The ninth century, the first in the period under consideration, is shown to be relatively wet in most records, particularly in the T7 δ¹³C and grey scale records, where 87 and 71 years exceed +1 standard deviations from 1036-1890 mean respectively. Between c. ^AD 800-830, each of the available Cold Air Cave records show warmer and wetter conditions relative to the 1036-1890 mean, with wetter conditions also shown at Lake Nhauhache. At c. AD 830, an abrupt shift to relatively cool-dry conditions took place in the T7 isotope and grey scale values, and the Dante Cave isotopes, prevailing until c. AD 850. An abrupt increase in δ¹⁸O and δ¹³C in the T7 and Dante Cave speleothem records was thereafter sustained until c. AD 900. The T7 grey scale and T8 δ¹³C timeseries shift to wetter conditions slightly earlier (c. AD 880).

Decreased diatom conductivity at Lake Sibaya further indicates that the SRZ was wetter than average, and Lakes Nhauhache and Nhaucati corroborate this. This period of relative wetness, reflected in each of the proxies, persisted until around ^AD 970.

From c. AD 970-1020, the most depleted δ¹³C values occur in the T7 stalagmite, while T7 δ¹⁸O values were only marginally below average. Although T7 grey intensity and Dante Cave isotope values in this 50-year period appear to suggest wetter conditions, the T8 δ¹⁸O and δ¹³C variables reflect the cool-dry conditions suggested in the T7 isotope values. A period of above average conductivity, suggestive of drier conditions, is also shown in the Lake Sibaya conductivity record. No coherent signal is present between records in the decadal period between c. AD 1020-1030. In the two decades after, normal conditions relative to the 1036-1890 mean predominate in the T7 isotope and T8 δ¹³C records, although T7 grey intensity trends towards drier conditions in the same timeframe. Markedly drier conditions at around AD 1050 are inferred from the Lake Sibaya conductivity record, the Dante Cave speleothem isotopes, and lake levels at Nhauhache and Nhaucati, suggesting generally drier regional conditions.

Similarly, the latter part of this period marks the driest conditions in the entire Baobab record, which commences in AD 1036. These dry conditions extended to c. AD 1070 in the Baobab record, and c. AD 1110 in the Lake Sibaya conductivity and T7 grey scale records. In the T7 isotopes and T8 δ¹³C chronologies, values remained above average until c. AD 1090, while the Dante Cave speleothem isotopes began to trend towards

wetter conditions around c. ^AD 1050, indicating a degree of regional incoherence over the extent of this possible dry spell (Figure 3.5).

Regional correspondence in inferred precipitation is relatively poor at the beginning of the twelfth century, with a number of ambiguities between records. From c. AD 1100-1140, dry conditions are observed in each of the Cold Air Cave speleothem datasets. Over the same timeframe, markedly wetter conditions are present in the Baobab record, while the Lake Sibaya conductivity record abruptly shifts from very dry to very wet, with a similar but less abrupt trend at Lake Nhaucati from c. ^AD 1100.

The sediment core from Lake Nhauhache, on the other hand, indicates variable but trending towards drier conditions. The Dante Cave speleothem isotopic values peak at c. AD 1100, and although conditions trend towards cool-dry, they mostly remain above the 1036-1890 mean until c. AD 1200. The period from c. AD 1140-1180 shows conditions in the T7 isotope records trending towards wetter conditions. Similarly, the T7 grey scale record suggests wetter conditions, as does the T8 δ¹³C record, while the T8 δ¹⁸O values are trending towards inferred warmer conditions. The Baobab record continues to show wetter conditions in this period, as does the Lake Nhaucati record. The diatom conductivity record from Lake Sibaya infers the second-highest stand in wet conditions over the 1200-year period at c. ^AD 1150, before an abrupt decline to very dry conditions at c. AD 1180. These later dry conditions are also inferred from the Lake Nhauhache core, and from c. AD 1170-1190 in the T7 isotopes. An overall variable agreement in palaeoprecipitation is observed in the twelfth century. In the Baobab record, for instance, 25 years exceed +1 standard deviation from the 1036-1890, while no years exceed these thresholds in the T7 variables.

A higher degree of correspondence in wet and dry conditions is present between nearly all records throughout the thirteenth century. Drier conditions persist into the first two decades of the century in most records excluding the Limpopo Baobab’s and T8 δ¹³C, which show a relatively low variability and sustained wet conditions throughout the century. At c. AD 1220, the T7 grey scale, T8 isotope and Lake Sibaya records show an abrupt increase in inferred precipitation, with this occurring two decades earlier in the T7 isotopes. Wetter conditions were sustained in the Lake Sibaya diatom until around AD 1300, but in the T8 δ¹³C record, a decline set in at around ^AD 1280. In the T7 records, this began at c. ^AD 1270 and persisted until at least c. ^AD 1300. The isotopic records from Dante Cave in Namibia show less coherence with the other SRZ records in this century. Here, an abrupt decline in isotopic values onset at about AD 1200, and was punctuated only by short periods of warm-wet conditions until a recovery in c. AD 1320. The wetter nature of the 1200s is reflected in that 34 and 31 years exceed +1 standard deviations of the 1036-1890 in the T7 δ¹³C and Baobab records respectively.

FIGURE 3.5 Main coherence of wet and dry periods, blue shading indicates correspondence of wet periods, and red shading dry periods. Mean lines represent 1036-1890 means.

Reasonable regional coherence in dry conditions is observed in the early decades of the fourteenth century, yet it is unclear from simple observation of Figure 3.5 and the previous palaeoclimate literature whether this constitutes the onset of a sustained reduction in precipitation associated with the regional LIA. At the beginning of the fourteenth century, conditions trended towards drier at Sibaya and in the Limpopo Baobab record, with very dry conditions observed at Sibaya from around AD

1310-1330. Although the Baobab values are still above the 1036-1890 mean at this time, they show their lowest values since c. AD 1080. Both the T7 and T8 δ¹³C records infer drier conditions until c. AD 1330, as does the T7 grey scale series. The δ¹⁸O records for both Cold Air Cave stalagmites, however, show increased values suggestive of warmer conditions. Drier conditions are also observed at Lake Nhauhache, though Lake Nhaucati continued to show wetter conditions until around AD 1400 before drying out completely (Ekblom and Stabell 2008). The Dante Cave speleothem isotopes suggest relatively warm-wet conditions until c. AD 1400, while isotopic data of faunal remains from the Shashe-Limpopo basin (Smith 2005) show that rainfall remained steady until at least c. AD 1415. In most other records, relatively drier conditions prevailed until at least c. ^AD 1380. In the last two decades of this century, though, the records are characterised by a high degree of ambiguity, with an equal split between proxies inferring warm-wet and cool-dry conditions.

Ambiguity between records continues into the fifteenth century. No dominant signal is shown until c. AD 1430, where generally dry conditions are observed through to c. AD 1450. Particularly abrupt shifts towards cool-dry conditions are shown in the T7 isotopes and towards dry conditions the Lake Sibaya conductivity record, while a less severe but equally abrupt drop is observed in T8 δ¹³C. Generally drier conditions are present in the T7 grey scale record, and narrow ring-widths predominate in the Karkloof record. The only proxies with relatively warm-wet conditions here are the T8 δ¹⁸O and the Baobab. At around ^AD 1450, a recovery towards markedly wetter conditions began in numerous records. The strongest and most abrupt of these are witnessed in the T7 isotope and grey scale records, T8 δ¹³C, the Dante Cave isotopes, and conductivity at Lake Sibaya. These conditions appear to last until c. ^AD 1480 in these records, and thereafter undergo a decline through to the end of the fifteenth century. The T7 isotopes undergo a decline from the inferred very warm-wet conditions to relatively warm-wet, and this is also the case in the T7 grey scale record.

T8 δ¹⁸O experienced a decline to below average conditions, followed by a recovery to the highest values in the series at c. AD 1500. T8 δ¹³C trends towards wetter conditions in this 20-year timeframe, as does the Sibaya conductivity record, albeit more abruptly.

By contrast, the Baobab record trends towards drier conditions, with the lowest rainfall observed since the eleventh century at c. ^AD 1480. As will be illustrated later, a high degree of variability is present in the Karkloof record in these two decades, with an abrupt shift from very dry to very wet conditions.

At c. ^AD 1500, conditions were wet or trending towards wet in most records, the exception being the Karkloof tree-ring, where an abrupt decline in ring-width is followed by an abrupt return to normal conditions. The Baobab record also shows an increase in precipitation, but this is only slight and reflects a return to normal conditions relative to the mean. At around AD 1520, each of the Cold Air Cave speleothem variables, Dante Cave speleothem δ¹³C, and the Karkloof tree-ring undergo an abrupt decline in inferred precipitation and/or temperature, with the T7 grey scale series showing the greatest proportional abrupt shift to dry conditions out of any record over the 1200-year period. This decline is also reflected in the Sibaya diatom at c.

AD 1530. Average conditions relative to the 1036-1890 mean, however, remain dominant in the Baobab. Recovery towards normal or slightly wetter than mean conditions did not begin until c. AD 1570 in most records excluding the Baobab, where an abrupt increase to the wettest conditions of the entire record lasted until c. AD 1560, after which conditions remained wet, but trended towards drier until c. AD 1600. These wetter conditions followed after c. AD 1570 in most other records, excluding the T7 δ¹³C and grey-scale records, where inferred-precipitation remained below average. In this century, the Baobab and T7 δ¹³C records show 22 and 28 years respectively exceeding +1 standard deviation from the common mean, while the T7 δ¹³C record also shows 22 years exceeding -1 standard deviation from the common mean.

Dry conditions prevailed from c. ^AD 1600-1640 in nearly all records, the only exceptions being slightly earlier recoveries in the T8 isotopes, and a period of increased ring growth in the Karkloof tree at c. ^AD 1620. This period also shows the highest grey intensity value of the entire record, inferring the lowest precipitation. A slight recovery in precipitation occurs in all records from c. AD 1640-1670, excluding the T7 grey scale record, conditions then trend towards very cool-dry in all records apart from T7 δ¹⁸O.

This period, lasting until around AD 1720, is the coolest and driest period in the T8 speleothem and Sibaya conductivity records, the second driest in the Limpopo Baobab record, and the joint driest in the T7 grey scale record. This appears to unambiguously indicate that this period was the driest part of the LIA in the region. It is not until c. ^AD 1710, though, that this abrupt decline starts in the T7 isotope records, where conditions

This period, lasting until around AD 1720, is the coolest and driest period in the T8 speleothem and Sibaya conductivity records, the second driest in the Limpopo Baobab record, and the joint driest in the T7 grey scale record. This appears to unambiguously indicate that this period was the driest part of the LIA in the region. It is not until c. ^AD 1710, though, that this abrupt decline starts in the T7 isotope records, where conditions