Holocene variations in Lake Titicaca water level and their implications for sociopolitical developments in the central Andes

Settings and Transfer Function Development

Lake Titicaca (16°S, 69°W) is a high-altitude lake on the northern Altiplano (average = 3,810 masl) of Bolivia and Peru (23). The lake consists of two subbasins (Fig. 1), Lago Mayor (7,131 km², 125 m average depth) and Lago Menor (1,428 km², 9 m average depth), connected by the Strait of Tiquina (39.5 m depth, SI Appendix, Fig. S1). When the lake level exceeds 3,804 masl, the lake discharges southward via the Río Desaguadero into the southern Altiplano to Lago Poopo.

Lake-level fluctuations are driven by the relative balance between precipitation and riverine inflow, and evaporation (which is closely linked to temperature and solar radiation) and riverine outflow. Variations in lake level can be used to infer fluctuations in this moisture balance. A substantial portion of the moisture variation is controlled by the changing intensity of the South American Summer Monsoon (SASM). During the instrumental period (1915 to 2022), evaporation has accounted for ~90% of water output (23), the remainder being attributed to the highly variable flow of the only outlet, the Rio Desaguadero. Annual lake-level rise (driven by the SASM seasonality) and fall averages ~0.69 m. Variability of lake-level rise is greater than variability of lake-level fall, and interannual fluctuations in lake level are correlated with precipitation variability (3). Over the last century, the total range of lake-level variation was ~6 m (24).

Both sediment cores and underwater archaeological test pits (SI Appendix, Figs. S2 and S3) enabled the identification of intervals of past lake-level lowering, including exposure surfaces (i.e., peat beds and gleyed soil), erosion surfaces or paleoshorelines (i.e., scour marks, abrupt transitions to dense sandy sediments, and highly fragmented shell materials), and non-depositional environments (i.e., soil horizons and hiatuses in sedimentation characterized by abrupt changes in radiocarbon ages). In addition, changes in the source of organic matter, as recorded by its carbon stable isotope signature, can be used to reconstruct historical displacements of the littoral margins (5). Along the gentle slopes of Lago Menor [average slope = 0.04 ± 0.01% (25, 26)], littoral areas <2 m deep are mainly colonized by emergent sedges (Schoenoplectus totora) and to a lesser extent by Hydrocharitaceae (Elodea sp.), Lemnaceae, and Haloragaceae (Myriophyllum sp.). Benthic macrophytes (e.g., Characeae and Potamogetonaceae) are found between 2.5 and 15 m depth, with maximum development of Characeae between 4.5 and 7.5 m (27). Stable isotope values of sediment organic carbon (δ¹³C_org) in the euphotic zone are more negative at depths less than 2.5 m, with values typical of emergent C3 sedges (av. = −25.6 ± 1.1‰), and higher at water depths between 2.5 and 15 m, with values characteristic of Characeae δ¹³C_org (av. = −10.2 ± 0.8‰) (SI Appendix, Fig. S4). In the aphotic areas (>15 m deep), the sediment δ¹³C_org values are dominated by those of pelagic algae (av. δ¹³C_org = −23 ± 2‰). The imprint of this ecological distribution with contrasting δ¹³C_org values along the slopes is inherited by the underlying sediments (5, 27–29) (SI Appendix, Table S1 and Fig. S4). Eroded soil organic material (av. δ¹³C_org = −24.5 ± 0.9‰) can also affect the δ¹³C_org values of littoral sediments through mixing processes. In deeper areas, atmospheric deposition (i.e., soil dust) during dry periods may reduce carbon isotope values (30, 31).

To reconstruct a continuous history of Middle to Late Holocene water level for Lago Menor, 13 short sediment cores were taken at water depths ranging from 3 to 43 m below the modern lake level (3,810 masl) to ensure good coverage of the amplitude of past lake-level variation (Fig. 1 and SI Appendix, Table S2A). This methodology enables different levels of resolution to be achieved as sedimentation rates and the number of sedimentary discontinuities decrease with depth. In addition, 18 underwater archaeological test pits (SI Appendix, Table S2B) were made between 1 and 15 m deep at Ok’E Supu in Lago Mayor, just north of the Strait of Tiquina (8), and 3 were dug in Lago Menor at K'anaskia and Ojelaya (SI Appendix, Fig. S3E), to refine the spatial resolution of paleoshoreline indicators and provide constraints on connectivity between the northern and southern basins of the lake. These sites were dated by determining the radiocarbon age of charcoal in selected stratigraphic layers. For each core, a transfer function (detailed below) that links the change in sediment δ¹³C_org to the change in lake level was computed based on a modern maximum interannual amplitude of lake variation. Paleoshoreline data were used to constrain the lowest lake level for each time period covered by the sediment archives (SI Appendix, Table S3 A and B).

Based on the instrumental record of lake level over the last century (24), a depth-δ¹³C_org relationship was developed and calibrated from two cores that span the last century (Fig. 2), with chronologies established from short-lived radionuclides (SI Appendix, Fig. S1 A and B). The model of lake-level variation for the last century (Fig. 2A) was based on two end-member depths, i.e., the lowest (3,807.7 masl) and highest (3,810.8 masl) average lake levels recorded in 1944 and 1986, respectively, and their associated δ¹³C_org values corrected for the Suess effect (32–34). The lake level was modeled using a transfer function (TFC_org) following Eq. 1:

Within the sediment cores and archaeological test pits (8), dated paleosoils and paleoshorelines were identified (SI Appendix, Table S3C), which enables us to constrain the elevation of multiple high and low stands (Fig. 2B). The lithologies of the cores were correlated to develop a composite stratigraphy within depth transects (SI Appendix, Fig. S2).

Within each lithological unit, the lowest (highest) δ¹³C_org value was associated with the highest (lowest) lake-level elevation end-member (i.e., EMt and EMb, with b as the bottom and t as the top), and the lake level of each unit was modeled following Eq. 2:

where Δδ¹³C_org is the difference between the δ¹³C_org of EMb and that of the depth considered in the profile. For intervals without high-stand end-members (EM), we used the maximum value of lake-level variation from the 100-y instrumental record (24).

Models were developed for each core (Fig. 2B) and subsequently combined into a single curve (Fig. 3) by averaging over 25-y intervals for the period 1913 to 1750 CE, 50-y intervals for the period 1700 CE to 900 BCE, 100-y intervals from 1000 to 2400 BCE, and 500-y intervals from 4000 to 6000 BCE, with the decrease in resolution of the dataset associated with increasing age (Fig. 2). To constrain the timing and amplitude of lake-level variations, we used several complementary methods, including paleobiotic, geochemical, and sedimentological analyses, as well as information on the paleohydrology and paleoprecipitation of the central Altiplano.

The fraction of freshwater algal biomarkers (n-alkane C_25-31) provides information about the water depth and salinity of the lake (increasing values with rising lake level and decreasing salinity) (30). Chlorophyll was used as a proxy for algal productivity (36), which was found to reflect primarily the productivity for Characeae in Lake Titicaca, caused by better preservation in encrustations of calcium carbonate (CaCO₃) (30). Hence, high concentrations of benthic macrophyte biomarkers (chlorophyll) and CaCO₃ in shallow areas reflect higher production of Characeae. Elsewhere in the lake, higher CaCO₃ values reflect lower lake level and the resultant higher chemical precipitation of CaCO₃ with increasing salinity (30, 37). Finally, because soils are enriched in Ti compared to lake sediments (38), Ti was used as an inorganic tracer of detrital inputs to the lake. Soil erosion from the catchment is mainly recorded at water depths less than ~2.5 m, because the dense totora sedges on the lake’s shore intercept more than 90% of terrestrial eroded particles (39). In deeper areas, the detrital contribution is likely dominated by aeolian inputs, deposited mainly during dry periods (40).

Results and Discussion

Our reconstruction provides continuous quantitative constraints on Titicaca water-level variation during the last 4000 y at multidecadal scale. The pattern of change in the reconstructed lake level (Fig. 2) tracks the measured water levels for the last century (R² = 0.70 and 0.73, P < 0.01), with maximum errors for modeled lake level of ± 2 m. The modeled error can be attributed to the lower resolution of the sediment records compared with the monitored lake level, coupled with cumulative errors associated with the age model.

This reevaluation of Lake Titicaca water-level variation provides higher temporal resolution than did previous reconstructions (7, 21) and is constrained with sedimentological data that enable the timing and amplitude of variation to be refined, particularly for the Middle and Late Holocene.

In the Middle Holocene, from 6000 to 4000 BCE, a rapid and large-amplitude decline in lake level of ~15 m characterized the southern basin of Lake Titicaca. δ¹³C_org in deep areas increased from <−22 to approximately −15‰, and n-alkanes C_25–31 dropped to the lowest values in the record (Fig. 3), indicating that Lake Titicaca transitioned from a deep, fresh, and overflowing lake to a shallow lake with its sediments colonized by Characeae. Subsequently, between 4000 and 2400 BCE, no sedimentation was evident in the cores collected in the deepest area (Chua trough), suggesting that the southern basin was almost desiccated during that interval. This observation is consistent with the conclusions of prior studies from Lago Menor (6, 21) and the Middle Holocene low stand reported for the Lago Mayor between 4000 and 3000 BCE (3, 6, 35), when lake level dropped ~85 m below the modern level (3, 9). We have reevaluated the depth of the current sill between the two basins and placed it at 39.5 m below modern lake level (SI Appendix, Fig. S1). Hence, during the 4000 to 2500 BCE period Lago Menor was no longer fed by Lago Mayor. This period of long-term falling lake level coincided with low austral summer insolation values and with the Northern Hemisphere (NH) Middle Holocene climate optimum. Both factors (low insolation, warm NH temperature) are known to be associated with reduced intensity of the SASM and lower amounts of precipitation (41).

Increased moisture and lake-level rise are evident in Lago Menor beginning ~2400 BCE, with the flooding of all sites located 25 m or more below modern lake level (i.e., post-hiatus lake sedimentation recovery in cores E2 and E3). Chlorophylls and CaCO₃ concentrations are high, and values of δ¹³C_org are above −15‰ (Fig. 3), consistent with the presence of a shallow Lago Menor colonized by Characeae, a condition likely restricted to the deeper eastern and western troughs. These shallow conditions lasted until at least 2000 BCE, after which the lake rose rapidly during the following 500 y. Consistency between the eastern (E1, E2, and E3) and western (W1) lake-depth records suggests that both the eastern and western basins of Lago Menor were connected by a channel that passed north of Suriqui Island (SI Appendix, Fig. S1).

Inter-basinal flow via this connection probably occurred even when Lago Menor was below its overflow level (3,804 masl). Subsequently, all areas 10 m or more below the modern level were flooded starting around 1500 BCE (cores W2, W3), and those below 7 m were flooded around 1300 BCE (cores W4, W5), as corroborated by successive abrupt drops in Ti at these sites, which characterizes the transition between soil and overlying lake sediment (Fig. 3 and SI Appendix, Fig. S2). At the same time, δ¹³C_org, CaCO₃, and chlorophyll concentrations increased, indicating that sediments were flooded by more than 3 m of water, enabling Characeae to colonize the area.

During the Late Holocene, stepwise increases in average lake level produced three phases, each with lake-level oscillations around a different baseline elevation: i) from 1300 to 700 BCE, when the lake fluctuated above and below ~3,798 masl; ii) from 700 BCE to 500 CE, with fluctuations around ~3801 masl; and iii) from 500 to 1450 CE, with fluctuations around ~3,804 masl. During that period of sequential lake-level rise, CaCO₃ decreased gradually, and algal biomarkers (n-alkanes C_25–31) increased, both in response to the progressive decrease in lake salinity associated with rising lake level (20). The reconstruction suggests that Lago Menor reached the modern overflow level after 500 CE (flooding of OJ and KA sites). In response, benthic macrophytes disappeared from the deep areas (drop in δ¹³C_org, CaCO₃, and chlorophyll) but likely colonized the newly flooded shallow ones. During the entire Late Holocene, decadal to secular lake-level declines coincided with increases in both Characeae biomarkers (chlorophylls and CaCO₃) and Ti values. For the latter proxy, one cannot exclude anthropogenic influence, because major development of agropastoralism on the lake shores occurred during the emergence of the Tiwanaku culture (42), which may have enhanced clastic inputs to the lake.

In the final period of Late Holocene lake-level increase, between 1450 and 1750 CE, lake level rose steadily to reach its modern level (>3,810 masl). The reconstructed period of lake-level rise at the beginning of the Fifteenth century is consistent with reports by Spanish chroniclers for the Inca and Colonial periods of great inundation of nearshore areas (43, 44) and corroborated by increased underwater offerings by the Inca in two ritual sites near the Island of the Sun, which were partially submersed at that time (11, 45). Following a lake level maximum at ca. 1750 CE, two substantive lake declines are reconstructed at ~1850 and 1940 CE. Both are consistent with reports of a course reversal of the Rio Désaguadero into the lake at 1835, 1845, and 1865 CE (46, 47) and the most intense recorded drought in the last century at 1943–1944 CE (21, 24).

Our reevaluation of the Middle to Late Holocene lake-level variation in Titicaca is in relatively good agreement with previous reconstructions of lake elevation from sedimentology, diatoms, and geochemistry (Fig. 4 B–D), notably for the timing of major lake rises (~1600 BCE, 700 BCE, 200 BCE, 500 CE, 800 CE, and 1500 CE) and declines (~1000 BCE, 400 BCE, 100 CE, and 1300 CE) (3, 7, 20). Our record differs in that it does not show abrupt changes in lake level of 10 m or more, or lowstands that persisted for centuries, such as those apparent in the stratigraphic record (7) around 300 BCE, 150 CE, and 1300 CE (Fig. 4B). These persistent intervals of no apparent sediment accumulation likely resulted from erosion of portions of the stratigraphic record during lowstand intervals.

The higher spatial and temporal resolution of our record relative to earlier studies yields smoother oscillations (of ~5 m) and a more gradual pattern of lake-level rise over time, which is consistent both with the observed pattern of interannual oscillations in the instrumentation period of the last century and with the evolution of regional precipitation reconstructed from the Quelccaya ice cap (48), Lake Umayo (49), Huagapo Cave (50), Huascaran ice core (51), and Lake Junin (52) (SI Appendix, Fig. S5).

Changes in lake stage inferred from the Middle to Late Holocene portions of our sediment record coincided with many periods of cultural change, which suggests that the inferred changes in the availability of water and arable land had consequences for people who resided around the lake. During (pre)history, the largest lake transgressions all coincided with major sociopolitical changes in the region. They include the shift from mobile hunting-gathering to a more settled agropastoral and fishing lifestyle (18) at the transition from the Late Archaic to the Early Formative period, which occurred during the 2500 to 1700 BCE abrupt lake-level rise. Persistent regional moisture during the Formative period (1500 BCE to 500 CE), which resulted in moderately high lake levels, likely promoted the expansion of agriculture, pastoralism, and the appearance of ranked societies (18, 53). In particular, the first chiefdoms and ceremonial centers, such as Tiwanaku, Chiripa, Lukurmata, and Pajchiri (Fig. 4E), were established around the lake during the Middle Formative period (MF: 800 to 200 BCE). At the time, all these sites were situated above the high-stand lake shoreline and therefore were protected from inundation during lake transgressions, such as those recorded at ~700 BCE and 200 BCE. Rain-fed terrace agriculture also began as early as the MF on the steep shores of Lake Titicaca (18, 54, 55), which provided additional land to meet agricultural needs related to population growth and to compensate for the loss of land covered by the lake during the MF (Fig. 4E).

Early development of complex societies in the Titicaca Basin occurred primarily during the Late Formative (LF: 200 BCE to 500 CE) and involved the control of domestic labor by emergent elites, intensification of agricultural systems, development of raised field agriculture (56), expansion of interregional trade, creation of elite ideologies, and competition with other elites (18, 57). By 400 CE, Tiwanaku stood alone, and other sites, such as Lukurmata, came under its influence by ~600 CE (58). This major change in the political landscape coincided with the largest recorded lake-level rise ~500 CE (Fig. 4A), when the lake reached a level close to that of the present. It is likely that a large number of agricultural and potentially residential areas around ceremonial centers located in the lake basin (e.g., Lukurmata) were impacted by this transgression, with the exception of the Tiwanaku capital (Tiahuanaco), which is located farther inland. Such a transgression could also have shifted the trade routes of the llama caravan to the Yungas along the western shores of Lake Titicaca through Tiahuanaco (59). The coincident major increase in population (60–62), together with the restructuring of Tiahuanaco and the development of the agropastoral landscape, including raised field systems (13, 18, 42, 63), suggests that this event influenced a migration of human populations, from previously occupied flooded regions, to cities located on higher ground. This densification may have contributed to the rise of the Tiwanaku state. The second half of the Tiwanaku state period was drier overall, marked by a gradual decline in lake level from 850 to 1000 CE, followed by a lake level rise around 1150 CE. This latter event coincides with the terminal portions of the sedimentary hiatuses identified in previous lake-level reconstructions (7, 15) and corroborates the hypothesis that portions of those stratigraphic sequences were lost via erosion and thus the duration of dry periods was shorter than those prior reconstructions suggested. Nonetheless, although reduced precipitation during that period likely affected Tiwanaku's residents, it does not imply a drought-induced collapse of the Tiwanaku culture (15, 64), given the evidence that prolonged aridity can be endured without major cultural changes (10, 65, 66).

In general, the trend of increased wetness during the Late Holocene promoted agricultural and social development, although at times, major lake rises may have influenced population migration, because of flooding of nearshore areas. This was certainly the trigger for the emergence of the Tiwanaku culture at the expense of other early sociopolitical formations, whose arable lands were flooded during the 500 CE lake transgression. Other geoarchaeological evidence buried in the sediment of Lake Titicaca will allow us to complete and deepen the chronology and context of this pivotal period in the development of Andean societies.

Materials and Methods

Supplementary Material

Data, Materials, and Software Availability

All complementary information from this study and other studies are included in the article and/or SI Appendix. Lake Titicaca water-level reconstruction data are archived at the Mendeley Digital commons Data repository, doi: 10.17632/9fkdt5j35v.1 (80).

Supporting Information