Approaches to Middle Stone Age landscape archaeology in tropical Africa

into a barrier, a refuge, and a corridor between southern and eastern African populations. Alluvial fans presently blanket the areas adjacent to major river systems, which were key areas of prehistoric human habitation. These sets of variables have created conditions that are both challenging and advantageous to conduct archaeological research. Lateritic soil development has resulted in poor organic preservation and facilitated insect bioturbation, which has demanded an integrated micro-macro scale approach to building a reliable geochronology. An integrated ﬁ eld and analytical methodology has also been employed to identify the nature and degree of post-depositional movement in alluvial deposits, which preserve a wide range of spatial integrity levels in buried stone artifact assemblages between 47 and 30 ka in Karonga, northern Malawi. This paper describes the methodological advances taken toward understanding open-air Middle Stone Age archaeology in sub-tropical Africa, and explores the inferential potential for understanding Pleistocene human ecology in the important southern Rift Valley region.


Introduction
The Middle Stone Age (MSA) of Africa shows some of the earliest clear archaeological evidence for many behaviors that are either unique to or especially well developed in modern humans (d 'Errico and Stringer, 2011;McBrearty and Brooks, 2000). As a typotechnological unit, the MSA is defined by the significant presence of prepared core technology, nascent forms of symbolic expression and the absence of large bifacial tools such as handaxes, spanning from ca. 285 thousand years ago (ka) to ca. 20 ka (Conard, 2015;Sahle et al., 2014;Tryon and Faith, 2013). Much of the evidence for the MSA comes from more temperate parts of Africa, both in the southern (Brown et al., 2012;Henshilwood et al., 2011;Lombard, 2011;Porraz et al., 2013) and northern parts of the continent (d' Errico et al., 2009;Linst€ adter et al., 2012). This record is also heavily biased toward the long archives of human behavior that are found in cave or rock shelter sites.
There is a need to advance methods to understand the depositional and taphonomic context for open-air MSA sites in tropical and subtropical ecosystems. New geoarchaeological approaches are tailored for these situations and have equipped researchers with tools to investigate the environmental backdrop of cultural diversifications by MSA people that were critical in defining the trajectory of Pleistocene human evolution. Herein, we present a multi-tiered approach to the interpretation of open-air MSA archaeological sites in alluvial settings in Karonga, northern Malawi, which draws from well-vetted techniques but combines them in novel ways. High-resolution spatial analysis using 3dimensional (3D) Geographical Information Systems (GIS) and refitting from piece-plotted archaeological excavations reveal whether archaeological deposits have been disturbed by secondary processes. Micromorphological data provide information about modes of deposition, pedogenesis and bioturbation, contextualizing their potential to alter the Optically Stimulated Luminescence (OSL) dosing environment used to estimate the ages of the sites. Integration of data from phytoliths allows for cross-referencing of macro-and micro-scale sedimentary environment observations, and developing terrestrial vegetation histories. Finally, by comparing data from archaeological contexts with a nearby highresolution lacustrine paleoenvironmental record from Lake Malawi and adjacent regions, a model can be developed that explains human behavioral patterns in response to climatic changes throughout the Late Pleistocene. We describe these approaches to MSA sites in the tropics with specific reference to Chaminade II (CHA-II), which is typical of many archaeological sites in the Karonga District in its depositional context and artifactual composition.

Background
The southern Rift Valley of Africa is a critical, yet little-explored geographic transition zone for understanding the evolution and diversification of MSA populations across Africa. Near the modern town of Karonga, Malawi ( Fig. 1), pioneering research by Desmond Clark and Vance Haynes in the 1960s Clark et al., 1970Clark et al., , 1966, followed by more intensive geoarchaeological investigations by Zefe Kaufulu (1983Kaufulu ( , 1990, demonstrated the high potential of the region's archaeological record to reveal information about the MSA in tropical and subtropical ecosystems. However, the lack of absolute ages made it difficult to tie the human behavioral record to the significant paleoenvironmental changes the area experienced over the course of the Late Pleistocene Lane et al., 2013;Scholz et al., 2011Scholz et al., , 2007. More recently, work by the Malawi Earlier-Middle Stone Age Project (MEMSAP) has revised interpretations of sites first studied by Clark and colleagues Clark et al., 1970Clark et al., , 1966 describing the spatial-temporal distribution and attributes of artifactual assemblages and artifact raw material sources near Karonga (Thompson et al., , 2012(Thompson et al., , 2013Wright et al., 2014;Zipkin et al., 2015).
In northern Malawi, downwarping and fault-trough sedimentation during the Middle to Late Pleistocene (Betzler and Ring, 1995;Ebinger et al., 1993) in combination with variable climatic conditions (Crossley, 1984;Stone et al., 2011) activated alluvial fans and streams, which created riparian environments attractive to MSA people (Thompson et al., , 2012Wright et al., 2014). The remains of MSA activities in the Karonga region are embedded in remnants of an alluvial fan system known as the Chitimwe Beds Clark et al., 1970;Kaufulu, 1990;Wright et al., 2014). Alluvial fans can rapidly bury occupation surfaces and rework older deposits. Because many of the sediments have been subsequently modified through pedogenesis and bioturbation, artifacts can change their stratigraphic position and are often found in association with paleosols that formed at a later date. The materials and methods employed by MEMSAP were tailored to address these specific taphonomic conditions of the Karonga sites, which fall within an equatorial savanna with a dry winter (Aw) of the K€ oppen-Geiger classification system (Kottek et al., 2006).

Landscape-scale geomorphology and site-scale sedimentology
The focus of the geomorphologic, sedimentologic and pedologic investigation concentrated on the identification of source-to-sink processes and areas of sustained landform stability in which human activity would have been plausibly preserved. Archaeological surveys and test pitting were conducted across an alluvial fan west of the modern-day town of Karonga, Malawi, which resulted in the identification of CHA-II (Fig. 1). Recording of sedimentary lithofacies involved description of sorting, bedding features, particle sizes, inclusions, rounding properties, hardness, plasticity and unit boundaries. Pedologic recording accounted for relative degrees of weathering by recording soil color, structure and the presence of authigenic and translocated minerals in the solum. Geomorphologic reconstruction of landform evolution was made by combining sedimentologic and pedologic analyses from numerous test units in combination with topographic maps. Sampling for OSL dating, micromorphology and phytolith analyses was performed to constrain environmental conditions within a dated context.

Excavation and artifact analysis
The CHA-II site (9.955 S, 33.892 E) was excavated in 2011 and 2012. A large eroded surface to the west of CHA-II contained 100,000s of artifacts in a series of drainage gullies that incise the Chitimwe Beds. CHA-II was mechanically excavated as a 4 Â 50 m trench to a depth of 2 m, and then excavated by hand within a central 2 Â 32 m area. The mechanically excavated sediments were placed in ten piles of 40 m 3 each, sieved, and found to confirm an average artifact density of 1 per m 3 . The center trench was excavated in natural stratigraphic units and 1 Â 1 m squares to a maximum total depth of 4.6 m, producing in total more than 10,000 stone artifacts. At the close of excavations, a geologic test pit was excavated at the southern end of the trench to a total depth of 6.6 m before the entire trench was backfilled. All artifacts, sedimentary units and samples were piece-plotted with a Nikon C-series 5" total station, and all artifacts with a discernable long axis were recorded with one point at each end.
Lithic artifacts were analyzed using complementary quantitative and qualitative methods on all plotted and sieved artifacts (sensu Mackay, 2008). When feasible, artifacts were also classified according to weathering stage (sensu Thompson et al., 2012), raw material, technological component (complete or fragmentary flakes and cores, flaking shatter, hammerstone, etc.), cortex cover and reduction method. Refits were sought to determine the extent of depositional reworking of the artifact assemblage, both horizontally and vertically. ArcGIS 10.2 was used to analyze the spatial relationships of artifacts of different sizes, raw materials, and rounding classes. The Optimized Hot Spot Analysis tool was used to identify clusters of artifacts within depositional units. Artifact orientation data were analyzed using Stereonet 9 to produce rose diagrams and establish if artifacts had preferred or random orientations in both the vertical and horizontal planes (McPherron, 2005).

Micromorphology
In 2012, seven block samples were collected from CHA-II for micromorphological analysis. The samples were oven-dried at 60 C and impregnated with resin under vacuum. The resin was prepared with seven volume units of unpromoted polyester resin (Viscovoss N 55S), three volume units of styrene (styrene for synthesis) and 5e6 ml/l hardener (methyl ethyl ketone peroxide, MEKP). After hardening, the blocks were sliced with a rock saw into slabs from which 6 Â 9 cm sized thin-sections of 30-mm were produced. Micromorphological descriptions were conducted following analysis using a petrographic microscope with plane polarized light (PPL), crossed polarized light (XPL), blue light fluorescence and incident light at magnifications of 20Â to 500Â following Courty et al. (1989) and Stoops (2003).
A total of 90 microscopic Fourier transform infrared spectroscopic (m-FTIR) measurements were conducted directly on the thin sections using a Cary 610 FTIR microscope attached to a Cary 660 bench (Agilent Technologies). Spectra were collected in two modes: (1) 64 co-added scans at 4 cm À1 resolution in transmission mode with backgrounds collected on epoxy resin over glass under typical ambient conditions and (2) 32 co-added scans at 4 cm À1 resolution using a germanium crystal attenuated total reflectance (Ge-ATR) objective with backgrounds on air. The transmission measurements allowed us to characterize the absorption peaks in the region of 4000e3500 cm À1 (Beauvais and Bertaux, 2002;Robin et al., 2013), while the Ge-ATR measurements provided peaks from 4000 to 570 cm À1 . Clay-sized materials were identified to kaolin and smectite group minerals by comparing the peaks to reference spectra (Downs, 2006;Salisbury and Vergo, 1991; Van de Marel and Beutelspacher, 1976) with priority given to peaks in the OHstretching region.

OSL geochronology
Five sediment samples were collected from the profile of CHA-II for OSL dating. Quartz grains with diameters of 180e250 mm were prepared using wet sieving, acid treatments (10% HCl, 10% H 2 O 2 and 40% HF) and density separation and analyzed using a TL/OSL-DA-20 Risø reader. A dose recovery test following the Single Aliquot Regeneration (SAR) protocol Wintle, 2000, 2003) for the lowermost sample LM12-10 determined that the measured and regenerated dose ratios were within 10% for all the preheat temperatures applied (Fig. 2). Equivalent dose (D e ) values were then determined in the plateau region in which the measured D e was independent of the preheat temperatures (Fig. 3). Dose rates (D r ) of the samples were estimated using a Canberra BEGe 5030 low-level high-resolution gamma spectrometer. OSL signals were measured based on aliquots composed of several tens of quartz grains (Small Aliquots, SA), and the final ages were derived using the Central Age Model (Galbraith and Roberts, 2012;Galbraith et al., 1999).
Two samples, LM-3161 and LM-3162, were interpreted as potentially spurious due to high over-dispersion values generated in the SA OSL analysis (Arnold and Roberts, 2009) and were therefore subjected to single-grain (SG) analysis. For SG dating, OSL signals in 1000 individual quartz grains from each sample were measured using a 10 mW Nd:YO 4 solid-state, diode-pumped green laser (532 nm). The optical stimulation was carried out at a readout temperature of 125 C for 2 s, and the photon detection was through a 7-mm Hoya U-340 filter. The single grain D e values were obtained using the SAR protocol (Murray and Wintle, 2000), and the feldspar grains were identified using IR depletion ratio (Duller, 2003). The OSL ages of the samples, together with dosimetry data and over-dispersion values, are given in Tables 1 and 2

Local paleoenvironmental reconstruction
Phytolith samples were taken as small sediment samples from the top to the base of the southeastern profile and then continuing down the same profile on the western side of the deep geological sondage. Sixteen sediment samples were processed for the purpose of creating a vegetation landscape reconstruction. Extraction protocols were followed that were successfully employed on neighboring MSA sites (Mercader et al., 2013) and modern topsoils (Mercader et al., 2011) from the Mozambican side of the lake.
To understand the relations between specific morphotypes and taxonomic identification, the inferential baseline was grounded on the statistical analysis of local phytoliths from plants and soils (Mercader et al., 2010(Mercader et al., , 2009) and supporting references (Albert et al., 2009;Barboni and Bremond, 2009;Bremond et al., 2008;Fern andez Honaine et al., 2009;Novello et al., 2012;Piperno and Pearsall, 1998;Runge, 1999;Twiss et al., 1969). Preservation was adequate for morphometric analysis and type identification, although marked dissolution was a concern in some samples. A total of 6089 discrete phytolith shapes were classified and grouped into 59 morphotypes (Fig. 5).

Lithology and site formation processes
MSA artifacts at CHA-II were entrained in alluvial fan deposits, but sedimentation of the site changed through time, reflecting different ecological conditions commonly found in tropical Africa. Site formation processes were interpreted on the basis of microscale (Supplementary Online Material 1) and macro-scale analysis of sediments and pedofeatures (Fig. 6).
Units 1 and 2 are interpreted as alluvial deposits based on the presence of horizontally laminated bedding features and wellrounded medium to coarse sandy channel sediments (Fig. 7a) in which an over-thickened Bt horizon developed during aggradation (Fig. 7bec). These deposits are capped by either a lag channel or deflated or winnowed alluvial fan surface (Unit 3). Units 1 and 2 are culturally sterile, and Unit 3 contains only sparse and heavilyrounded artifacts within the same size grade as the surrounding clasts (Fig. 8). OSL ages dating to approximately 65 ka (LM-3161)   (2017) 64e77 were assayed from Unit 1, but authigenic mineral formation associated with groundwater fluctuations and gleying of the sediments have likely compromised the D r estimate as demonstrated by the results of the SG OSL analysis. SA OSL and SG OSL analyses from Unit 2 (LM-3162) do not provide reliable estimates of age of deposition of this unit based on the over-dispersion values obtained.
Within Unit 4, very poorly-sorted, coarse grained alluvial fan deposits have undergone multiple phases of Bt horizon formation, which also destroyed bedding structures that may have been present (Fig. 7dee). Artifacts (mainly unweathered and some refitting) are concentrated in the upper 10 cm of the unit, which was deposited ca. 47 ka (LM12-10; Table 1), although authigenic ironmanganese (FeeMn) nodule formation would have altered the dosing environment and affected the D r over the burial life of the sample. Additionally, multiple fluctuations in moisture initiated the shrinking and swelling of smectites present in the soils, resulting in pedoturbation and the formation of vertic features (Fig. 7feg). Vertic processes led to the physical deformation of illuvial clay coatings, but Bt horizon formation continued, as evidenced by a later phase of coatings (Fig. 7h). At one point, Bt horizon formation ceased when the groundwater table rose significantly, which gleyed much of the lower profile and contributed to the formation of FeeMn nodules within the inactive Bt horizon. After the groundwater level fell, the uppermost of the nodules and large clasts (including artifacts) settled downwards forming a secondary deposit of nodules in the profile, which accommodated bulk volume changes from dewatering and winnowing of fine particulates.
Following the lowering of the FeeMn nodule rich horizon, deposition of finer-grained, distal alluvial fan sediments recommenced prior to 30 ka (LM12-11; Table 1) and successive formation of Bt horizons continued with clay coatings forming around lower nodules (Fig 7i) and cementation from iron leaching and accumulation within the profile. Artifact counts are highest in the bottom 30 cm of Unit 5 and are virtually absent in sediments that accumulate thereafter. Following landform stability after 17 ka (LM12-12; Table 1), lateritic soil formation pedogenically weathered quartz grains, feldspars and smectite clays top-down in Unit 5, enriching the unit in kaolinite. Additionally, termite bioturbation, associated with the formation of the laterite, occurred downwards to the top of the iron-nodule concentration (Fig. 7j).

Lithic analysis
The lithic artifacts from CHA-II form a typo-technologically generalized MSA assemblage, but provide useful data for understanding site taphonomy and land use strategies over time. Analysis of a sample of 640 piece-plotted artifacts from an area of 8 m 2 shows both dorsal and ventral sides of the artifacts were frequently coated in FeeMn precipitates, with only five artifacts demonstrating differential weathering on opposing sides. More detailed examination suggests that there may be two or more lenses of artifacts within this vertical concentration (Fig. 8, lower left). Vertical refits and conjoins are non-existent between designated sedimentary units. Within Unit 4 (Lower Concentration), five sets of conjoining/refitting artifacts were identified, along with eight similar sets in Unit 5 (Upper Concentration), found in discrete clusters (Fig. 9). Artifacts from CHA-II are typically more square than elongated, so that within this sample only 125 had a maximum length:breadth ratio of !1.5, and thus a discernable long axis. Of these, neither the upper nor the lower lens shows a preferred orientation that would suggest winnowing (McPherron, 2005). Additionally, all artifacts are lying at angles <22 (most far flatter), indicating that they have not been post-depositionally reoriented by bioturbation (Bernatchez, 2010;Lenoble and Bertran, 2004). In total, <1% of artifacts are retouched, and over 95% of cores retain some cortex coverage, with an average 44% cortex coverage on all cores. a Depths of the samples are the vertical distance from the modern ground surface. b Average water content of the samples during burial. c Data from high-resolution low level gamma spectrometer were converted to infinite matrix dose rates using conversion factors given in Olley et al. (1996). d Cosmic ray dose rates were calculated using the equations provided by Prescott and Hutton (1994). e Over-dispersion. f Central Age Model (CAM) ± 1s error (also calculated for D e ). * Cosmic ray dose rates were calculated using the equations provided by Prescott and Hutton (1994) and provided in Table 1. a Depths of the samples are the vertical distance from the modern ground surface. b Data from high-resolution low level gamma spectrometer were converted to infinite matrix dose rates using conversion factors given in Olley et al. (1996). c MAM ¼ Minimum Age Model (Galbraith et al., 1999). d CAM ¼ Central Age Model (Galbraith et al., 1999). e Over-dispersion. f Minimum Age Model, ± 1s error (4 parameters sensu Galbraith et al., 1999). g Central Age Model, ± 1s error (Galbraith et al., 1999). De ¼ 37 ± 22 Gy, CAM age ¼ 63 ± 18 ka, Minimum age ¼ 21 ± 12 ka; (b) Small aliquot (SA) analysis of LM-3161, CAM D e ¼ 118 ± 9 Gy, OD ¼ 37%, Min D e ¼ 96 ± 13 Gy, CAM age ¼ 66 ± 7 ka, Minimum age ¼ 54 ± 8 ka; (c) LM-3162 (SG), CAM D e ¼ 55 ± 6 Gy, OD ¼ 51%, Minimum D e ¼ 27 ± 4 Gy, CAM age ¼ 27 ± 3 ka, Minimum age ¼ 13 ± 2 ka; (d) LM-3162 (SA), CAM D e ¼ 174 ± 19 Gy, OD ¼ 51%, Minimum D e ¼ 75 ± 11 Gy, CAM age ¼ 85 ± 10 ka, Minimum age ¼ 37 ± 6 ka; (e) LM12-10 (SA), CAM D e ¼ 95 ± 8 Gy, CAM age ¼ 47 ± 4 ka; (f) LM12-11 (SA), D e ¼ 72 ± 3 Gy, CAM age ¼ 30 ± 1 ka; (g) LM12-12 (SA), D e ¼ 42 ± 1 Gy, CAM age ¼ 17 ± 1 ka.

Phytolith analysis
Phytolith data provide an additional environmental context for the depositional and artifactual data (Supplementary Online Material 2). Phytoliths from woody (arboreal) tissue average 86.2% (range: 61.5e100%) and therefore they constitute the dominant type analyzed from CHA-II. Overall, this is indicative of a generally moist arboreal domain; however, there are cyclical peaks and drops in the abundance of arboreal phytoliths that can be classed into four vegetation phases illustrating the alternation of three wet periods with two drier ones (Fig. 6). There was no evidence for mixing of phytolith taxa between the following phases: 'Phase 1' is from Unit 1, and is likely much older than 47 ka and inferred to be ca. 65 ka based on SA OSL and SG OSL analyses with low confidence in the accuracy of this age estimate. The lower part of these sediments contains heavy arboreal representation, which transforms into fragmented woodland (arboreal phytoliths: 61.5%) in the upper part of Unit 1, where seasonal (panicoid) and wet-adapted (bambusoid) communities still abound (Poaceae phytoliths combined: >30%). This zone is interpreted as reflecting a wet woodland community. 'Phase 2' also predates 47 ka, and it yields the highest levels of arboreal phytoliths from the entire series, with grasses being minimal or absent. Phytoliths from this zone are interpreted as reflecting a forest environment, and correspond to geologic Units 2 and 3. 'Phase 3' records significant arboreal reduction starting at the stratigraphic level dated to 47 ka through 30 ka, and covering the upper solum of geologic Unit 4, and lower 30 cm of Unit 5. Phytoliths from the dry-adapted chloridoid clade peak at 58.6% for all Poaceae types. This zone is interpreted as signalling the presence of predominantly dry, grassy woodland. 'Phase 4' postdates 30 ka and extends to <17 ka. Taxa analyzed from this zone indicate a return to arboreal dominance for the remainder of the stratigraphic column, albeit with small fluctuations in seasonal-grass types. This phase is inferred as reflecting dense woodland, which is distinct from the presentday Zambezi woodland vegetation classification of the region (White, 1983).

Discussion
The integrated paleoecological and archaeological reconstructions of CHA-II were designed to maximize the information potential of an open-air MSA site in tropical Africa. Preservation of organic remains is typically poor in tropical regions, and especially so in open-air settings, and CHA-II was no exception. However, by analyzing inorganic fossil remains of plants, sediments and artifacts, MSA environments and potential drivers of human behavior can be inferred on both a local and regional level.

Depositional and age models of CHA-II
In tropical environments, where rainfall tends to be heavy and seasonal, alluvial fans are typically activated during a high precipitation event following a drought severe enough to devegetate large swaths of an upland region (e.g., Kumar Singh et al., 2001;Nott et al., 2001;Thomas, 2004;Waters et al., 2010). Tectonic activity can also catalyze alluvial fan formation, but in the absence of   . Note that artifacts in the upper portion of Unit 5 were not piece-plotted, but that sieving showed they averaged <1/m 3 . A close-up of squares A16-B20 is shown at lower left, with only artifacts in the main concentration pictured for clarity. These have been divided into two possible lenses, with a separate rose diagram at lower right given for each. The points on the rose diagrams represent the plunge of the artifacts, while the roses themselves represent their orientations. significant upwarping or solifluction alluvial fan formation normally occurs in tandem with variable precipitation (e.g., Bull, 1977;Harvey, 2002;Kumar et al., 2007;Ritter et al., 1995;Wang et al., 2001). Erosional and depositional processes across alluvial fans are not uniform and sheetwash zones, channels and debris flows are distributed irregularly across landforms (see papers in Harvey et al., 2005 for a comprehensive review).
Based on the available evidence, sedimentation of CHA-II occurred primarily as pulses of distal alluvial fan deposits, which were sometimes reworked and deposited by intermittently occurring fluvial systems (Fig. 11). Sediment packages were poorly sorted, well mixed and do not appear to vary appreciably over the 32 m exposed length of the site. Clast sizes generally range between fine and coarse sand with pockets of sediments with higher gravel fraction.
The potential for large-scale bioturbation at CHA-II is high, but the evidence suggests that such activity was localized. Groundwater fluctuations and lateritic processes facilitated the formation of the FeeMn nodule-rich lens that overprints bedding features from the upper portion of the sondage. Subsidence from water table depletion has been observed at CHA-II based on the uniform, undulating sediment packages observable in the profile (Fig. 8), but large accommodation zones for fine clasts to move downward do not appear to have formed. Conjoins and refits from Units 4 and 5 (Fig. 9) and clusters of chipped stone debris indicate that these are primary or minimally compressed archaeological deposits reflecting discrete activity areas. Furthermore, the micromorphology indicates that sediment mixing was restricted to distinct sedimentary zones and large-scale lateral movement of sediments decimeters from their primary depositional environment was not significant.
In addition to being a tool for dating sediment deposition, OSL is commonly used as a tool to understand bioturbation and pedoturbation within archaeological contexts (e.g., Arnold and Roberts, 2009;Gliganic et al., 2012;Gu erin et al., 2015). Because the sedimentary matrix reworking at CHA-II was localized, SA OSL was employed as the method of first resort. Recent studies demonstrate that if over-dispersion is within a threshold range of <40%, it is possible to obtain accurate age estimates from SA OSL analysis as long as the skewness values within a population of aliquots are low indicating a normal or log normal distribution (Alexanderson and Murray, 2007;Armitage and King, 2013;Arnold and Roberts, 2009;Forman, 2015;Rowan et al., 2012). Radial plots of samples from CHA-II demonstrate a central tendency of the aliquots with the exceptions of LM-3161 and LM-3162 (Fig. 4), which show scattered D e values indicating some sediment reworking, pedoturbation and non-solar resetting of sediments prior to deposition. Based on the combined evidence, one age is rejected (LM-3161) and two ages are considered highly uncertain (LM-3162 and LM12-10), however post-depositional effects did not significantly impact the accuracy of LM12-11 and LM12-12.
The combined sedimentary, micromophological, artifactual, phytolith, and OSL data infer a depositional model that is inconsistent with Crossley (1986) interpretation of site taphonomies from the Chitimwe Beds. Based on granulometric analyses of the sand sheets and termite mounds in the Chaminade area, Crossley (1986) concluded that the vertical (upward) transfer of selected (medium to coarse) sand grains by Macrotermes falciger termites led to the accumulation of the sand sheets and concentration of larger particles and lithic artifacts in the lower reaches of the solum. Instead, we argue that Bt horizon overthickening and the formation of vertic features associated with soil formation and groundwater variability degraded the integrity of bedding features in the profile commonly found in alluvial deposits, making the deposits appear similar to biomantles to the naked eye. Fig. 9. Results of refitting program and GIS cluster analysis. Following Sisk and Shea (2008), "refits" are artifacts that were separated through the knapping process/conchoidal fracturing, while "conjoins" are fragments of formerly whole flakes that fit back together. In both the Upper and Lower Concentrations, refitting and conjoining artifact sets occur predominantly within cluster of artifacts identified using the Optimized Hot Spot Analysis functionality in ArcGIS 10.2. This tool identifies spatially concentrated incidences (of artifacts), and identifies the likelihood that the spatial relationships within them reject the null hypothesis of a random distribution, at increasing confidence levels.

Landscape and behavioral models of CHA-II
Archaeological evidence from CHA-II records recurrent occupation associated with MSA foragers throughout the late MIS4 and entire MIS3 periods in the Karonga region. Diatom and geochemical proxy reconstructions of the Lake Malawi ecology indicate that site sedimentation occurs when the local environment fluctuates between mildly xeric and pluvial conditions (Finney et al., 1996;Gasse et al., 2002;Stone et al., 2011). The diatom assemblage from Lake Malawi following the eruption of Mount Toba (ca. 75 ka) and prior to 50 ka shows several prominent and rapid ecological changes (Lane et al., 2013;Stone et al., 2011), which broadly accord with the phytolith assemblage from Units 1 and 2 of CHA-II. Similarly, phytolith taxa from CHA-II connote the presence of dry, grassy woodland from this time, which are echoed in the reconstructed paleoenvironments of the Mozambican highlands for the interval 50 to >42ka (Mercader et al., 2013: 330). Phytoliths entrained in Units 1e3 reflect the predominance of wet woodland to forest environments, indicative of a moist precipitation regime during periods of site sedimentation. Proportions of woody vegetation drop markedly in Unit 4, which was filled by alluvial fan deposits, and then increase again through Unit 5. Occurrences of the MSA in the area are inferred prior to 47 ka by the presence of heavily-rolled artifacts in deposits associated with Unit 3 and the bottom of Unit 4.
Human occupation at the site began in earnest after 47 ka and appears to have occurred during a period of profound landscape variability. Based on magnetic susceptibility of an assayed lake core at Lake Masoko (Fig. 10d), located 70 km to the north of CHA-II, the region experienced longer dry seasons and potentially stronger seasonality in precipitation compared to present conditions, but was dominated by Zambezian tree, open savannah vegetation (Garcin et al., 2006). Similarly, the environment at CHA-II became more open, which is supported by on-site grassy phytoliths and offsite diatom data from Lake Malawi reflecting higher alkalinity (Stone et al., 2011). After 30 ka, the phytolith spectra at CHA-II reflect the presence of a dense woodland community, which is in agreement with pollen and biogenic silica data derived from the Lake Malawi sediment core (Beuning et al., 2011;Johnson et al., 2011;Stone et al., 2011) and Lake Masoko (Garcin et al., 2006).
The concentration of artifacts within Unit 5 is the product of repeated occupation of the same land surface as soil aggraded over an extended period of time concurrent with slow sedimentation of the landform prior to 30 ka as the climate transitioned to wetter conditions (Fig. 11). Toolstone types and limited reduction of nodules indicate that toolstone availability was not a limiting factor. Rather, MSA foraging strategies incorporated regular visits to riparian zones for lithic raw material, which was processed and then transported outside the riparian areas for usage. Similar patterns of landscape utilization within diagenetically-altered alluvial fan sedimentary environments have been documented in the Kapthurin Formation (Tryon, 2010) and Mukogodo Hills (Pearl and Dickson, 2004) of Kenya and Jebel Gharbi in north-west Libya (Barich et al., 2006) possibly representing a generalized site taphonomy and behavioral strategy of MSA foragers operating in dynamic sedimentary and climatic environments. Artifacts entrained above the lower 30 cm of Unit 5 are relatively rare and likely bioturbated within the sediments either through termite activity or entrainment in the alluvial fan. Some artifacts from the upper portions of Unit 5, including some small pieces, can be characterized as Later Stone Age (LSA), whereas none from within the zone of artifact concentration can be assigned firmly to any typo-technological complex other than the MSA. This adherence to cultural stratigraphy provides additional support for our interpretation of the site's formation (Fig. 11).
The micro-scale data are corroborated by lithic orientation data at CHA-II, which show most artifacts as lying horizontally, and by the lack of mixing of small LSA elements into the sediments dated >30 ka. This demonstrates that MSA behavior at both site and landscape scales can still be discerned in tropical settings that have been subsequently affected by tectonic, authigenic, pedogenic and climatic changes since the times of original artifact deposition.

Conclusion
Archaeological research of MSA occupation of CHA-II located on an alluvial fan in northern Malawi integrated paleoclimatologic, sedimentologic and spatial datasets to better understand human habitation patterns within a tropical to subtropical ecosystem. The primary human occupation of the landform occurred between 47 and 30 ka, concentrated on slowly aggrading alluvial fan sediments within a grassy woodland zone that was drier than during preceding and following phases of sedimentation. Although this is the driest period reflected in diatom assemblages after 70 ka (Stone et al., 2011), the magnitude of ecological variability in Lake Malawi appears to have been far less than in surrounding basins at  (Scholz et al., 2011), red dotted line is a reconstructed lake level based on the diatom assemblage from the Lake Malawi Drill Cores (Stone et al., 2011), dashed lines after 50 ka represent age model uncertainty from discovery of Toba Ash as reported in Lane et al. (2013); (c) Biogenic silica mass accumulation rate (mg SiO 2 /cm y 2 ) from Lake Malawi ; (d) Magnetic susceptibility (Â10 À5 ) from Lake Masoko (Garcin et al., 2006); (e) Gray line shows Dppt from modern conditions (mm) from Lake Tanganyika, red line shows time-averaged linear trendline (Bonnefille and Chali e, 2000); (f) d 13 C of monocarboxylic fatty acids (C 28 ) vs. VPDB from Lake Tanganyika (Tierney et al., 2008). Numbers on right margin correspond to proxy points identified in Fig. 1. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) the same time (Bonnefille and Chali e, 2000;Tierney et al., 2008). The evidence for repeated visits to the site and lack of conservation of toolstone suggests that this was a relatively attractive environment for MSA foragers. CHA-II demonstrates similar patterns in landform utilization to other MSA sites in the Karonga region (Thompson et al., 2012;Wright et al., 2014), on the Mozambican side of the lake (Bicho et al., (in press);Mercader et al., 2011Mercader et al., , 2013 and elsewhere in Africa as cited above.
The taphonomic situation at CHA-II and other sites did not allow for the preservation of evidence for other human activities such as faunal resource exploitation or hearth construction. Much archaeology conducted in tropical ecosystems faces similar challenges due to generally poor preservation conditions in biomes with high, seasonal rainfall and rapid soil formation. Bioturbation from insects tends to be particularly severe due to the lack of cold conditions to seasonally limit tunnel construction activities. Laterite formation and highly acidic soils also degrade the integrity of organic deposits and buried A-horizons, as does redoximorphic weathering of sediments associated with fluctuating ground water levels. More specifically to the MSA, the lack of diversity in the material culture, particularly from open-air settings, limits behavioral interpretations (Sharon et al., 2014).
In spite of these limitations, archaeological studies of open-air MSA sites from tropical regions are critical for developing robust models explaining human behavioral evolution and diversity (Shea, 2011). Therefore, we advocate a multiproxy, multidisciplinary research strategy that can sufficiently tackle the challenges of working in tropical ecosystems and yield information needed to contribute to such discussions. This study shows that detailed information about MSA landscape use, lithic exploitation, and local environments can be reconstructed using multiple lines of evidence tailored to these challenging preservation conditions. Furthermore, regional paleoecological datasets augment site-scale archaeology and sedimentology, providing a broader context for deposition, soil formation and aspects of human behavior.

Acknowledgments
We thank our collaborators at the Malawi Ministry of Sports and Culture for their assistance and permission in facilitating this research. Specific thanks is given to former Deputy Funding for the micromorphological analysis and mFTIR was provided by the Deutsche Forschungsgemeinschaft (MI 1748/3-1, MI 1748/1-1, and ME 4406/1-1). AC thanks the Lake Malawi Drilling Project, NSF-Earth System History Program (NSF-EAR-0602404), DOSECC Inc. and LacCore for support. Composing this manuscript took place at the University of York, which generously hosted DW while on sabbatical from Seoul National University. Thanks to the two anonymous reviewers and organizers of this special edition of Journal of Archaeological Science, Mike Morley and Paul Goldberg who greatly improved the quality of this manuscript.