1 Introduction and Study Area

The East African Rift System (EARS) (Fig. 8.1) is perceived as a migration corridor linking eastern and southern Africa and providing potential ways “out of Africa” for early humans. In this context, we focus on the Northern Tanzanian section of the eastern branch of the EARS , the so-called Gregory Rift . As already stated by Leakey (1979) and numerous other authors, Tanzania is well known for prehistoric specimens of early hominids found, for example, in Olduvai Gorge, at Lake Masek or at Laetoli (Fig. 8.2). For the evolution of hominids and especially hominins, the EARS seems to be a crucial region. The geographic center of this study lies in the vicinity of Olduvai, yet within the half-graben of the Gregory Rift around the eastern margins of Lake Manyara and along the Makuyuni River. Lake Manyara (954 m a.s.l.) is located in an endorheic basin (Fig. 8.1) and is a shallow soda lake that periodically dries out completely (Deus et al. 2013). The primary structural unit of the Manyara basin is an asymmetrically shaped rift. To the west extends the 250–900 m high escarpment. A west-dipping monocline and the volcano Essimingor prevail in the east. The water supply originates from springs at the base of the rift escarpment and from several seasonal drainages (Schwartz et al. 2012).

Fig. 8.1
figure 1

The locations and extent of the study areas in the Lake Manyara basin in Northern Tanzania

Fig. 8.2
figure 2

Outlined sites of the Olduvai, Masek, Laetoli, and Manyara Beds, which comprise rich paleontological and archeological findings [with permission of Bachofer (2015)]

Paleo-lake evidences , like paleo-shorelines, sediments, and fossils, were found by Jaeger (1913) and Reck (1921) to the east of Lake Manyara already in the beginning of the twentieth century. Pleistocene sequences in the valley of the Makuyuni River were discovered early on by Louis and Mary Leakey and were later examined by Kent in 1935 (Kent 1941, 1942). Keller and colleagues collected Pleistocene faunal material and Middle Stone Age (MSA) lithics and published several stratigraphic sections (Keller et al. 1975). Recent studies investigated the geology, paleontology, and archeology of the Lake Manyara Beds, e.g., Kaiser et al. (1995, 2010), Kaiser (2000), Ring et al. (2005a), Frost et al. (2012), Schwartz et al. (2012), Mana et al. (2012), and Giemsch (2015).

Previous studies have shown that two fossil-bearing layers of different ages exist in the area, namely, the Lower and Upper Manyara Beds (UMB and LMB, respectively). Correlations with the sequence in Olduvai seem to indicate Lower and Middle Pleistocene ages for the Manyara Beds (Ring et al. 2005a; Frost et al. 2012; Schlüter 1987). Kaiser et al. (2010) show the corresponding stratigraphic succession. Upper terrestrial beds and lower lacustrine beds are separated by a clear boundary characterized by a color change from grayish (lacustrine sediments) to reddish (terrestrial deposits). The entire stratigraphy is interspersed with presumably reworked tuff deposits. Moreover, UMBs and LMBs bear fossil layers.

The large number of (fossil) bearing sites , including specimens of fossil vertebrates and artifacts, detected during various field campaigns of the authors from 2007 to 2014, was the motivation for a more extensive analysis of the spatial distribution of these sites, in relation to environmental characteristics and processes. Different methodological approaches, with a major emphasis on remote sensing , were utilized to contribute to the understanding of the paleo-landscape development and their relations in order to find assemblages.

We investigate the morphotectonic evolution of the study area, the complex lacustrine dynamics of the Lake Manyara, and its paleo-stages as well as the related lake terraces and shorelines at different levels (up to 80 m above today’s lake level). Additionally, we study the distribution of topsoils and outcropping substrates to decipher soil formation processes as well as erosional and depositional processes. Finally, the fossils and artifacts were characterized, and the probability distributions were determined using a stochastic approach . The proposed methods contribute to a better understanding of the paleoenvironmental interrelations within the study area.

2 Morphotectonics and Their Interpretation

In the EARS , landforms are caused by effects of the active continental extension zone within the graben systems as well as by volcanic, erosional, and depositional processes. Tectonic processes formed the recent drainage systems and landforms. The tectonics of this region were previously studied in terms of (1) kinematic and structural geology (Ring et al. 2005b), (2) faulting, (3) Neogene tectonics and volcanism (Dawson 1992, 2008; Le Gall et al. 2008; Albaric et al. 2009), (4) and structural geology (Albaric et al. 2010). However, the link between landscape morphology and tectonics has yet to be directly addressed.

Drainage network, stream longitudinal profiles, basin analysis, and lineament extraction can be used as methods for identifying tectonic activity and related features in rift areas (Flores-Prieto et al. 2015):

  1. 1.

    Relief and landscape patterns of the Lake Manyara basin reflect the evolution of a complex part of the rift systems that underwent combined effects of tectonic factors inherent to its geological settings and Quaternary geomorphological processes. In the Makuyuni catchment, lineaments were extracted from a Synthetic Aperture Radar (SAR) image (ENVISAT ASAR, 2011-08-02) (Fig. 8.3). They show an N–S orientation and reveal tectonic activity characterized by deviations and knickpoints of streambeds and the occurrence of erosion processes.

  2. 2.

    Concavity and steepness values of drainage network segments depend on basin morphology, underlying rock strengths, and hydraulic geometry (Snyder et al. 2000). These values are especially used for evaluating river system responses to different landscape forming and modifying processes, including patterns of tectonic uplift and deformation (Wobus et al. 2006). The general increase of concavity and steepness values within drainage longitudinal profiles in the Makuyuni catchment indicates an uplift associated with tectonic activity.

  3. 3.

    A distinct base level of a landscape is related to an erosional stage and can be considered a product of erosional tectonic events (Golts and Rosenthal 1993). The development and migration of knickpoints due to changes in the base level can be related to lineament structures.

  4. 4.

    The morphology of the stream longitudinal profiles (Shahzad and Gloaguen 2011), basin tilting (Hare and Gardner 1985) and basin hypsometry (Pérez-Peña et al. 2009) suggest the importance of tectonic activity in governing Quaternary geomorphological processes, such as river incision and soil erosion, and hence, the landscape evolution of this region.

Fig. 8.3
figure 3

Automatically and manually extracted lineaments

This analysis has allowed the differentiation of the Makuyuni catchment into two distinctive areas: (a) tectonic activity influences the hydrological controlled northern section, and (b) a more stable southern section, where the Precambrian lithology and less intense tectonics prove dominant.

3 Delineation of the Manyara Beds

The stratigraphy of the Manyara Beds includes two fossil-bearing beds, namely, the LMBs and the UMBs (Schwartz et al. 2012). The UMB is composed of terrestrial deposits of up to 13 m mudstones, siltstones, conglomerates, and breccias. The deposition took place between 0.633 Ma and 0.44 (0.27) Ma (Ring et al. 2005a; Schwartz et al. 2012). A distinct layer of tephra is situated at the transition of the UMB to the LMB, which was 40Ar/39Ar dated to 0.633 Ma (Ring et al. 2005a; Schwartz et al. 2012). Schwartz et al. (2012) conclude, from assumed sedimentation rates, the beginning of the deposition of the LMBs between 1.3 Ma and 0.98 Ma. The Manyara Beds are underlain Precambrian intermediate gneisses in the south of the study area and by a sequence of nephelinitic conglomerates, nephelinitic lavas, and rocks in the vicinity of Essimingor, which have a minimum age of 1.4 Ma (Bagdasaryan et al. 1973). The LMB represents lacustrine deposits , grayish in color, and consist mainly of tuff, marls, siltstones, mudstones, and diatomite. The LMBs indicate the maximum extent and highest level of Lake Manyara.

An ASTER multispectral satellite image (2006-08-23), and topographic information derived of a SRTM-X digital elevation model (DEM), was used as input datasets for the delineation of the Manyara Beds. The ASTER system is composed of a sensor for visible and near-infrared wavelengths (VNIR: 0.52–0.86 μm; 15 m ground resolution), a shortwave sensor (SWIR: 1.6–2.43 μm; 30 m ground resolution), and a thermal infrared sensor (TIR: 8.125–11.65 μm; 90 m ground resolution) (Fujisada 1995). For the bands of the VNIR and SWIR spectral region, 35 spectral indices were calculated (Bachofer et al. 2015). Spectral indices are ratios of distinct spectral bands emphasizing the presence or absence of vegetation and different mineral compositions. Twenty-five additional topographic indices were calculated from a SRTM-X DEM (Bachofer et al. 2015).

For the delineation of the specific stratigraphic layers of the Manyara Beds, we applied a support vector machine (SVM) classifier based on the library of support vector machines (LIBSVM) (Chang and Lin 2011). SVM is a supervised classification approach, which maximizes the margin for the input features (spectral and topographic indices), describing the intended target classes (UMB and LMB). For the optimization of margins between several input features, it is necessary to transfer the data into an n-dimensional feature space. Kernel functions are used to separate the independent input features by the target classes (Vapnik 1995, 1998). Four hundred ninety-eight ground reference points, which were collected in field campaigns, were used to train and validate the datasets. The UMB and LMB were identified with an overall accuracy (OA) of 80%, when merely applying the SVMs to delineate the LMB 91% OA were reached (Bachofer et al. 2015). The results exhibit the surface distribution of the LMB and UMB (Fig. 8.4). In the eastern part of the study area, lacustrine sediments were identified, which are similar to the LMB and bear also vertebrate fossils . Based on indications of previous research, due to the topographic position of these sediments, we propose the existence of a cascaded lake system parallel to an extended paleo-lake Manyara following the main faults toward the east.

Fig. 8.4
figure 4

Section with the classification result of UMB and LMB (based on ASTER and SRTM-X data)

4 The Paleo-shorelines of Lake Manyara

Paleo-shorelines and terraces result from different stages of quaternary lake-level high stands, which were already identified by various researchers in the early twentieth century (Uhlig 1909; Jaeger 1913; Leakey 1931). Certain shorelines were even mapped (Keller et al. 1975). The 14C and Th/U dating of stromatolites of a distinct paleo-shoreline level helped in identifying different humid periods at 22 ka, 27–23 ka, 35–32 ka, 90 ka, and 140 ka (Hillaire-Marcel et al. 1986; Casanova and Hillaire-Marcel 1992; Casanova 1986). The ages were supplemented by diatom analysis of two drilling cores in the Lake Manyara basin, which supported humid periods between 27.5–26 ka and 12.7–8 ka (Holdship 1976; Barker 1990; Goetz 1990) (Fig. 8.5). An in-depth spatial consideration of the shorelines had thus far been missing.

Fig. 8.5
figure 5

Evidence for high paleo-lake levels in the Lake Manyara basin and basin faulting [with permission of Bachofer (2015)]

Paleo-shorelines appear as small-scale landforms, which disappear in optical remote sensing images because of the uniform surface cover compared to their surroundings. A spectral analysis of the landforms proved to be difficult and led to unsatisfying results. Therefore, it was made use of the capabilities of SAR images. Most SAR systems are active remote sensing systems, which operate in the microwave electromagnetic spectrum. SAR systems transmit energy pulses and record the backscattered energy of objects (Lillesand et al. 2004). The backscattered signal is influenced mainly by the physical properties of a surface. In the case of paleo-shorelines and terraces, which occur as elongated steps and ridges, their geometry and the roughness of their coarse covering rocks led to an intense backscattered signal. The images from 2011 to 2013 were acquired by the TerraSAR-X satellite system (Bachofer et al. 2014). The scenes were radiometrically calibrated, and the paleo-shorelines were delineated using a Canny edge operator (Canny 1986). For the edge analysis, the elevation was extracted of a SRTM-X digital elevation model. Several prominent paleo-lake levels were identified and validated with reference data from field surveys (Fig. 8.6) (Bachofer et al. 2014). The maximum lake level identified resembles an identical elevation to the lowest possible outlet of the endorheic Manyara basin (Fig. 8.6A). The position of the outlet indicates an overspill into the neighboring Engaruka and Natron-Magadi basins (Fig. 8.7).

Fig. 8.6
figure 6

Identified levels of paleo-shorelines after (Bachofer et al. 2014). The colors indicate different elevations of the paleo-shorelines. “A” shows the highest shoreline with more than 80 m above the present lake level. Stromatolites were found on this level

Fig. 8.7
figure 7

The paleo-lake levels for the Manyara basin at 978 m and 1030 m

5 Lithosphere and Surface Soil Mapping

The distribution of soils and rocks yields valuable information for the interpretation of landscape evolution. To acquire information regarding the surface characteristics, especially in areas with scarce vegetation cover, remote sensing techniques prove highly suitable.

During fieldwork and through laboratory analysis of surface soil samples, nine soil and lithological target classes were identified. As input datasets for the classification served (1) a high resolution WorldView-2 scene, (2) the backscattered intensity information of TerraSAR-X and ENVISAT ASAR SAR images, (3) medium resolution ASTER spectral bands and corresponding spectral indices, as well as (4) SRTM-X-derived topographic indices. Because of the size of the input dataset, an image object segmentation was conducted to derive relative homogenous objects with the average value information of the underlying raster cells (Baatz and Schäpe 2000). Vegetated areas were excluded from further analysis by applying a normalized difference vegetation index (NDVI) threshold (Rouse et al. 1974). A nonlinear SVM approach was applied to classify the input dataset. Based on the spectral, physical, and topographical properties of the surface material, the nine target classes were identified with an overall accuracy of 71.9% (Fig. 8.8). By merging similar target classes such as “carbonate-rich sediments” and “calcaric topsoil,” the accuracy could be increased significantly.

Fig. 8.8
figure 8

Section of the pedo-lithological classification (based on WorldView-2, ASTER, TerraSAR-X, ENVISAT ASAR, and SRTM-X data)

By means of recursive feature elimination (RFE), the features with the highest impact on the classification are identified (Guyon and Elisseeff 2003). The six most important input features are presented in Table 8.1.

Table 8.1 The six features with the highest impact on the classification identified by RFE

The results show that topsoils with iron oxide properties are predominant when associated with mafic outcrops of the slopes of the Essimingor volcano. “Carbonate-rich sediments” represent chiefly the LMBs, while “calcaric topsoil” indicates soil development processes upon the LMBs or on soils influenced by secondary carbonates. The thin outcrops of tuffs were identified at the stratigraphic border between the LMBs and UMBs. “Silica-rich topsoil” results from denudation processes on soils developed on the Proterozoic intermediate quartzite and gneisses of the Masai Plateau in the south of the study area.

6 Archeological Settings and Site Prediction

The large number of sites with specimens of fossil vertebrates and artifacts detected during different field campaigns was the motivation for a more thorough analysis of the spatial distribution of these sites in relation to present-day environmental characteristics and processes. Therefore, we developed an integrative spatial modeling concept using GIS, remote sensing , and sophisticated statistical methodologies based on statistical mechanics, e.g., MaxEnt (Jaynes 1957; Phillips et al. 2006). This method is able to handle presence-only datasets such as the locations proper.

The predictor variables consist of 50 continuous raster datasets applied in the modeling. The topographic indices characterize erosion transport and deposition processes as well as climatic and geologic variations in the landscape. They therefore not only reflect the immediate vicinity of a specific spot but additionally refer to a wider territorial context. Furthermore, we exploited spectral satellite data from the ASTER platform from which we calculated derivatives (spectral indices and band ratios) as predictor variables. Though most indices are designed for specific materials, they may also be used for a relative distinction between various minerals, depending on the abundant materials in the study area.

The target (also called response or dependent) variables consist in sites with paleontological evidence such as fossils or artifacts. In total, 102 sites were identified with differential GPS and transformed into point vector objects. Subsequently, we distinguished between archeological findings (stone artifacts only) from Early Stone Age (ESA) (n A = 45) and from Middle as well as Late Stone Age (MSA/LSA) (n B = 14) (Giemsch 2015) and for paleontological locations (n C = 43; fossils only).

The MaxEnt model predicts potential areas in which further fossil and/or artifact sites (paleontological sites) may be located. Figure 8.9 shows the areas with high potential in red to yellow and areas with low potential in dark blue. The research method considers not only site-specific characteristics but implicitly also the related pedogenetic and morphogenetic processes. The areas with high potential are mainly related to the paleo-drainage network terraces and paleo-lake deposits.

Fig. 8.9
figure 9

Modeled probabilities and their variable importance in percentage for (a) Early Stone Age (ESA) artifacts, (b) Middle and Late Stone Age (MSA/LSA) artifacts, and (c) fossil sites

In order to evaluate the model’s predictive performance beyond classification matrices, we calculated the receiver operating characteristics (ROC) curve integral, also known as area under curve (AUC). Model performance is given by AUC values of 0.95 for artifact sites and 0.90 for fossil sites.

The relative variable importance charts illustrate the overall importance of the salinity index (Al-Khaier 2003). The index relies on the ASTER bands four and five. This parameter may describe terrain elements with high salinity related to the former lake shores where salt concentration was high due to high salt contents of the lake itself and high evaporation rates of arid to semiarid paleoclimatic phases. The same bands can also be used to delineate soils with high iron content (Bierwirth 2002; Cudahy 2012). Other important ASTER-derived parameters are the Burn Ratio (Hudak et al. 2004) as well as the Laterite index (Bierwirth 2002). The Burn Ratio is designed to highlight burned areas and however in our case correlates well with dark vertisols with sparse grass vegetation, which cover the lower river terraces and the LMB/UMB transition. The Laterite index shows high iron contents present in the terrestrial tephra deposits. The DEM gives hints about the absolute elevation, in this case correlating with the former lake levels. The aspect, which is derived from the DEM, indicates the geological structure with main fault systems oriented from northwest to southeast.

Generally, the analysis shows that the paleontological sites are correlated mainly with the LMB/UMB transition and the paleo-river network. The results indicate the dependence of early hominids on water resources and food resources such as game, which also concentrates in these ecological units. However, the major difference between the paleontological sites and the ESA/MSA/LSA sites is related to general higher elevations where fossils occur. Moreover, the fossils seem to be related to stable paleo-landscape features indicated by the quartz-rich rocks index (Rowan et al. 2005) and the Laterite index. In fact, the sites are mainly associated with cape rock material forming on lateritic paleo-soil horizons and outcrops of the felsic basement. The LS factor, a derivative of the DEM, points to a transport of the material by surface runoff (Moore et al. 1991).

7 Conclusions

The presented results from different digital geospatial methods draw a comprehensive picture of the landscape development in the Lake Manyara basin. Remote sensing data enabled area-wide analysis and interpretations. Attention should be paid on the individual contributions of remote sensing sensors. The application of SAR data and different optical sensors with high spatial and/or spectral resolution allowed a detailed analysis of present-day and paleo-landscape features. Stochastic - and machine-based learning algorithms provided insights into the driving factors concerning the spatial distribution of paleo-landscape features and find locations. In detail, we obtained the following results:

  1. 1.

    The analyses of morphotectonics of the Makuyuni catchment coupled with comprehensive research on topography, drainage networks, stream longitudinal profiles, and lineaments revealed a morphostructural control with an N–S trend for the uplifted Masai Block, as well as tectonic deformation in the NE of the study area.

  2. 2.

    The Manyara Beds were identified, and evidence of lacustrine sediments further east of the study area has been detected. Besides a probable direct connection to the paleo-lake Manyara, it is also possible that parallel endorheic lacustrine and/or palustrine systems existed further east, which were subsequently cut and drained by the Makuyuni River.

  3. 3.

    The Lake Manyara underwent different transgression and regression phases during the abovementioned humid periods. The distinct morphological features of the paleo-shorelines prove their extent on different elevations in the Lake Manyara basin. The most elevated shoreline proves an overspill into the Engaruka/Natron basins.

  4. 4.

    By analyzing a combination of remote sensing -derived surface characteristics and terrain features, the spatial distribution of topsoils and related soil types was derived. The results indicate a complex and heterogeneous evolution of the study area and of the Manyara basin as a whole. We identified strong relations between silica-rich topsoils and the Precambrian Masai Block, topsoils with iron oxide properties, and andosols on volcanic layers, as well as between carbonate-rich substrates and the Lower Manyara Beds.

  5. 5.

    We assessed the paleontological as well as archeological sites using stochastic models allowing also a prognosis of potential find locations. As shown, the highest potential is related to paleo-river network terraces and paleo-lake deposits. There are distinct differences between the fossil sites and the artifact sites. The latter seems to be closer to the paleo-shorelines and the paleo-river network, whereas the fossil sites seem to be related to cape rock formations in the higher parts of the fluvial terrace systems.

These above-summarized results have demonstrated the enormous potential of innovative remote sensing methods and statistical, machine-based learning algorithms to derive valuable information on paleo-landscape features on larger spatial scales. However, detailed fieldwork activity is required to properly calibrate and validate the proposed analyses.