Between the tropical Andes and the Amazon Characterizing phytolith assemblages over environmental gradients

(1)

Word count: 10006

Between the tropical Andes and the Amazon

Characterizing phytolith assemblages over environmental gradients

Source: globesecret.com

Master Thesis Earth Sciences 2017-2018 (30 EC) Track: Environmental Management

Serge Mooijman Student number: 10448489 Supervisor: Dr. William Gosling Co-assessor: Dr. Carina Hoorn

Faculty of Science

Institute for Biodiversity & Ecosystem Dynamics University of Amsterdam

(2)

Abstract

Application of phytolith fossils in paleovegetation reconstructions has become increasingly popular since the last change of century. Paleovegetation reconstructions can provide information about response and resilience of vegetation through time on

environmental changes, including climate change and human impact.

The Andes mountain range and the Amazon contain high plant biodiversity. Andean-Amazonian gradients of climate, land use and human disturbance provide an opportunity to relate phytolith assemblages to taxonomic turnover of vegetation.

In this research, phytoliths from soil from Ecuador’s Tropical Andean-western Amazonian forests were counted and related to plant taxa. In order to find controls of forest structure and functioning, taxonomic variance was related to a measured set of

environmental gradient of climate, land use and human disturbance using Canonical Correspondence Analysis.

Stronger human disturbance affected taxonomic composition most, but taxa remained relatively resilient to moderate human disturbance, potentially informing

environmental management strategies. Temperature related gradients also strongly affected taxonomic composition. Impact of precipitation related gradients was unexpectedly low, possibly as a result of limited reliability of the precipitation data.

Palm taxa thrived at warmer low elevations. Grass taxa were more abundant at cooler higher elevations. No increase of arboreal taxa was found above 1200 meter,

although expected as a result of increased moisture availability due to persistent and permanentmontane cloud formation.Near local roads more agricultural phytoliths were found, reflecting their use by local farmers.

In this research, phytolith analysis reflected existing knowledge, emphasizing potential as a paleo proxy.

Keywords

Tropical Andean-Amazonian corridor, vegetation, phytolith assemblage, environmental gradient, taxonomic variance

(5)

1 1. Introduction

Knowledge about vegetation response to environmental changes through time can increase our understanding of current patterns in biodiversity and resilience (Dickau et al., 2013; Girardin et al., 2014; Urrego et al., 2011). Additionally, paleovegetation response to past climate changes can facilitate estimation of responses of vegetation to current and future climate change (Blois et al., 2013; Braconnot et al., 2012; Moritz & Agudo, 2013). While scientific knowledge about the impacts of climate change on plant biodiversity is still limited, paleovegetation reconstructions can help to develop effective strategies for

biodiversity preservation and nature management (Bellard et al., 2012; Fordham et al., 2013; Henne et al., 2013). Vice versa, the presence and abundance of recognizable remains of plant taxa with known tolerance ranges for temperature and precipitation, combined with dating techniques like the 14C method, allow reconstruction of temperature and precipitation regimes through time.

Environmental gradients between the Andes and Amazon like elevation-temperature and precipitation relate to high biodiversity (Malhi et al., 2010). The Tropical Andes are estimated to harbor one-sixth of all plant species on less than 1% of global surface area, including ~20.000 endemic plant species, more than anywhere else in the world and 6.7 % of all plant species (Myers et al., 2000). This species richness offers opportunities for research on environmental controls of forest structure and functioning (Asner et al., 2014; Girardin et al., 2014; Malhi et al., 2010; Terborgh, 1977). Paleoresearch over these gradients can provide insights into the effects of climate change through time on a variety of

ecosystems over a relatively short distance (Bush et al., 2007; Urrego et al., 2011). For example, Amazonian-Andean elevational gradients offer opportunities to study how lowland plant species cope with climate change by upward migration (Girardin et al., 2014). Another example is the impact of changes in precipitation regimes on the water retention in shallow soils, like the leptosols that are found frequently in the Andean-Amazonian corridor. Water retention in shallow soils is an important factor for plant survival (Quesada et al., 2011). Additionally, the role of Andean tropical montane forest ecosystems in carbon storage and their response to anthropogenic climate change is poorly understood, yet of global relevance (Anderson et al., 2011; Chimner & Carlberg, 2008; Huasco et al., 2014). Another factor that is still insufficiently understood is to what extend the vegetation composition in the Andean-Amazonian corridor has been affected by ancient human activities (Barlow et al., 2012; McMichael et al., 2017). The ability to quantify human impact on paleovegetation subsequently allows to the evaluation of the resilience of that type of vegetation to such anthropogenic disturbances. Such knowledge could provide guidelines for current forest preservation (Barlow et al., 2012; McMichael et al., 2017).

In paleoecology and paleoclimatology, several proxies, often each with different assets and constraints, can be applied over various timescales. Pollen, a common

paleoecological proxy, exhibits several limitations that can complicate broad applicability. For example, pollen signals reflect regional vegetation better than local signals, due to Aeolian dispersion of certain pollen grains (e.g. grass (Poaceae) and Pinus pollen) (Barboni et al., 2010). Additionally, pollen usually only preserve well in lake sediments and peat soils, thus limiting geographical applicability (Barboni et al., 2010; Faegri & Iversen, 1950; Havinga, 1967). Another limitation of pollen based vegetation reconstructions is the inability to distinguish morphologically similar grass pollen subfamilies (Barboni et al., 2007; Dickau et al., 2013). Changes in the ratio of C3 and C4 grasses are used to reconstruct temperature

(6)

2

changes, as C4 grasses thrive at higher temperatures than C3 grasses (Bremond et al., 2008b; Dickau et al., 2013; Twiss, 1987). Therefore, a widely applicable paleovegetation reconstruction methodology that lacks the limitations of pollen grains is desirable.

The period since the last change of the century is characterized by a rapid increase of developments in and applications of phytoliths in paleoresearch. This is substantiated by an increased number of articles about phytoliths and a diversification of interrelated research topics (Hart, 2016). The latter is illustrated in figure 1.

Figure 1: diversity and interrelatedness in modern phytolith research (Hart, 2016)

Phytoliths are microscopic silica (SiO2.nH2O) bodies produced in vascular plants. Phytoliths can remain preserved in a variety of soils types for thousands, sometimes even millions of years and reflect local vegetation (Morcote-Rios et al., 2016; Pearsal, 2015; Piperno, 1988; Piperno, 2006; Strömberg, 2004). Phytoliths can be applied for plant identification in

angiosperms, mostly at the family level, but sometimes, and increasingly often, at finer taxonomic resolution (Bowdery, 2015; Morcote-Rios et al., 2016).The diagnostic capacities of phytoliths in combination with their abundance and resistance to stress over time provide a valuable line of evidence in the fields of paleobotany and paleoecology (Bremond et al., 2008b; Dickau et al., 2013; Morcote Rios et al., 2016; Pető, 2013; Piperno et al, 2006). In some cases, phytolith analysis can distinguish taxa within vegetation families where other proxies fail to do so. For example, in phytolith analysis grass subfamilies can effectively be distinguished, allowing temperature reconstruction by calculation of C3:C4 ratios (An et al, 2015; Bremond et al., 2008a; Twiss, 1987). Additionally, identification of phytoliths related to agriculture is of archaeological interest (Ball et al., 2016; Barlow et al., 2012; Dickau et al., 2013; McMichael et al., 2012; Pearsal & Piperno, 1993; Piperno, 1984; Piperno et al., 1988;

(7)

3

Piperno, 1990; Piperno et al., 1990; Piperno el al., 2000; Piperno et al., 2015). For example, certain palm tree (Arecaceae) subfamilies with a known history of domestic use can be distinguished from other subfamilies through phytolith analysis (Morcote-Rios et al., 2016). Additionally, the recognition of ancient human activity can provide information about the geographical spread of human impact on the distribution of taxa with agricultural utility in the Andean-Amazonian corridor (Dickau et al., 2013; Levis et al., 2017; McMichael et al., 2012; Morcote-Rios et al., 2016; Piperno et al., 2015). However, phytolith datasets relating

taxonomic plant turnover to environmental gradients between the tropical Andes and the Amazon remain scares, in spite of potentially broadly relevant information such studies could offer (Asner et al., 2014; Bush et al., 2007; Girardin et al., 2014; Malhi et al., 2010; Urrego et al., 2011).

1.1 Aim and relevance

This study aims to relate taxonomic turnover in modern phytolith assemblages to measured environmental gradients of climate, human disturbance and land use between the tropical Andes and the Amazon. A phytolith dataset analyzing taxonomic turnover along a set of measured environmental gradients of climate, human disturbance and land use in the Tropical Andean-Amazonian corridorcan provide reference for paleoresearch that potentially informs climate change science, biodiversity conservation and archaeology.

1.2 Research Questions

 How does taxonomic composition vary in phytolith assemblages over measured environmental gradients of climate, human disturbance and land use in the tropical Andes Mountains down to the Amazon floodplain?

 What is de impact of human disturbance on phytolith assemblages in the study area?

1.3 Hypothesis

Phytolith assemblages differ most between sample areas ordered along gradients of elevation-temperature and precipitation. At lower elevations, taxa that thrive at higher temperatures, like palm taxa, are expected to increase in abundance. At mid elevations (1200 - 1500 m), arboreal taxa are expected to increase in abundance, due to increased moisture availability as a result of montane cloud formation at ground-canopy level (Bruijnzeel, 2001). At higher elevations and in dryer areas grass taxa are expected to increase in abundance. Human disturbance is reflected in an increase of the presence of phytoliths related to agriculture.

2. Literature background: progress and limitations in

phytolith analysis

Although an increasing number of studies relate phytolith morphology to plant types on different taxonomic levels, much can still be done to improve identification levels

(Bowdery, 2015; Morcote-Rios et al., 2016). Some phytolith shapes are unique for certain plants at a taxonomic level of family, genus or even species, leaving no doubt about taxonomic inference. Other plant taxa produce a range of different phytolith morphotypes (multiplicity), or phytoliths that are morphologically similar to those of other plant taxa

(8)

4

(redundancy) (Piperno, 1988; Strömberg, 2004). In general, multiplicity and redundancy complicate taxonomic identification of phytoliths (Strömberg, 2004). However, in some cases certain characteristics can be exploited for taxonomic inference. Some Amazonian palm trees for example, produce the same set of phytoliths shapes, but in ratio’s that are typical for a certain family, genus or species. By calculating the ratio’s between co-occurring palm phytoliths and comparing them to the ratio’s that are diagnostic for a specific taxonomic level of Amazonian palm trees, allocation can sometimes be achieved (Morcote-Rios et al., 2016).

Moreover, continued progress in linking phytolith morphology to more plants and at more precise taxonomic levels is expected to increase the future diagnostic power of phytoliths (Bowdery, 2015; Morcote-Rios et al., 2016). Although different palm species can produce similar shaped phytoliths, in certain cases it appears that species can be

discriminated based on subtle differences in phytolith morphology. Under a scanning electron microscope, factors like surface ornamentation in combination with degree of symmetry, length and number of the phytolith projections appear to be taxon specific and therefore useful for diagnostic purposes, even at species level in some cases (Bowdery, 2015; Morcote-Rios et al., 2016; Piperno, 2006). An additional factor of importance in taxonomic plant identification is knowledge of the local flora, emphasizing the value of regional reference collections (Morcote-Rios et al., 2016).

Similar to other palaeoenvironmental proxies, phytolith analysis has limitations. Non linear relationships between phytolith abundance and taxonomic abundance can be caused by variation in quantitative phytolith production between species, genera or plant families (Piperno, 1988). In calibration studies, where the result of modern phytolith samples can be compared to the actual vegetation on the sampling locations, correction factors could be calculated. Another factor that can complicate phytolith analysis is caused by the fact that taxonomical inference at species level is often not (yet) possible, and therefore a number of different species from a plant genus or family can be attributed to certain phytoliths. As a result, response to a specific environmental gradient can be bimodal or even multimodal (Correa-Metrio, 2011; personal communication, McMichael, C.H. November 2016; Urrego et al., 2011). Moreover, differential preservation, solubility, transport, inheritance and dispersal by wind and water can potentially impede accuracy when applying contemporary analogues to interpret palaeoecological and archaeological records (Dickau et al., 2013; Piperno, 1988; Strömberg, 2004). Some phytoliths are more susceptible for dissolution and/ or

fragmentation than others, which may introduce bias in the data of older sediments. For example, conical phytoliths typical of Cyperaceae tend to be poorly preserved in soils (Morcote Ríos et al., 2015). Soils with a Ph ≥ 9 can dissolve phytoliths (Piperno, 1985). Moreover, herbivory can cause mixing and transport of phytoliths (Strömberg, 2004). Another factor that can cause bias, inheritance, is caused by the residential persistence of phytoliths within soils (Fredlund & Tieszen, 1994; Strömberg, 2004). While vegetation may change over time due to changing environmental conditions, the assemblages in the soil can include phytoliths from past vegetations that no longer accurately reflect vegetation at a later point in time. Fredlund & Tieszen (1994) identified five processes that can cause dispersal by erosion, transportation and deposition of phytoliths in soil assemblages: (1) fluvial/ colluvial processes, (2) herbivory, (3) Aeolian processes, (4) fire-Aeolian processes and (5) gravity or decay-in-place. In dry open areas with frequent fires, like grasslands, lateral phytolith dispersal as aerosols by wind can occur (Piperno, 1988; Strömberg, 2004; Twiss et al., 1969). However, in closed habitats like forests, where Aeolian dispersal is limited, local deposition dominates and phytoliths can be expected to reflect local vegetation (Piperno, 1988; Strömberg, 2004).

(9)

5 3. Study area

The study area is located in Ecuador, South America (Fig. 2).

Figure 2: location of Ecuador within the South American continent. Source: worldatlas.com

3.1 Location and transect

Samples were taken in Ecuador’s Andean-Amazonian corridor (Figs. 3 and 4), with elevations ranging from 304 to 2266 meter above sea level (masl), an elevational gradient of 1962 m. The most southern sample location, near Lake Ayauchi, is situated at a latitude of S3.05 and a longitude of W78.03 decimal degrees (Fig. 4). The most northern location, Tena 2, is situated at latitude of S0.59 and a longitude of W77.88 decimal degrees (Fig. 4).

Sample locations are all within a geographical distance of 270 kilometers mostly in north-south direction. East-west spread between sample locations is maximally 69 kilometers. A full list of sample locations, their coordinates and their elevations can be found in appendix A. Values of environmental gradients on the sample locations can be found in Tab. 2 (p11).

(10)

6

Figure 3: topographic map of Ecuador. Samples were taken within the blue rectangle. Adapted from Varela (2018)

(11)

7

Figure 4: soil sample locations in Ecuador‟s Andean-Amazonian corridor (Google Earth image). The yellow line represents the border with Peru. Region borders are represented by white lines

3.2 Climate

The Köppen climate classification in the study area is tropical rainforest for all sampling locations. Being located in the direct vicinity of the equator, daily hours of sunlight (~12 hours) and temperatures vary little over the seasons in Ecuador. However, diurnal temperatures can fluctuate considerably, ranging from 10°C in the lowlands to over 20°C in the mid – and higher Andes (Jørgensen et al., 1999). Mean annual temperatures (MAT) vary with elevation from 15.8°C (Tena 2) to 25.1°C (Lake Ayauchi) (Fig. 5). The lapse rate in the study area is 5°C km-1 (González-Carranza et al., 2012; Jørgensen et al., 1999).

(12)

8

Figure 5: Mean Annual Temperature (MAT) from 1961 to1990 at the sample locations (IPCC Data Distribution Centre, 2007)

Precipitation can vary considerably in Ecuador between locations, seasons and years. However, although the annual north-south oscillations of the Intertropical Convergence Zone generally strongly affect seasonal precipitation patterns at regional levels in Ecuador,

seasonal variation is less pronounced in the Andean-Amazonian corridor, were moist air masses are blown in from the Amazon basin by prevailing easterly trade winds (Espinoza et al., 2015; Jørgensen et al., 1999; Killeen & Solorzano, 2008). These air masses, forced upwards by the eastern Andes, are subject to adiabatic cooling, resulting in superfluous precipitation all year round (Grubb & Whitmore, 1966). Geographical differences in precipitation patterns can possibly be explained by topography rather than elevation (Girardin et al., 2014). Annual precipitation between the sample location ranges from 1597 mm (Tena 2) to 4321 mm (Sumaco) (Tab. 1). A factor increasing moisture availability above ~1200 masl in the study area is continuous or persistent montane cloud formation at ground-canopy level (Bruijnzeel, 2001; Bush et al., 2007; Foster, 2001).

According to the mean annual precipitation (MAP) map (Fig. 6) all sample locations are in areas that receive at least 2000-3000 mm of MAP. However, the MAP data from climate-data.org (appendix B), retrieved from local weather stations, gives 1597 mm of average annual rainfall in Puyo, suggesting limited reliability of the MAP data.

(13)

9

Figure 6: Mean Annual Precipitation (MAP) at the sample locations (Google Earth layer from Bookhagen , in review)

3.3 Geology and soils

Intensive tectonic uplift of the Andes in Ecuador started 25 Ma ago (Coltorti & Ollier, 2000; Hoorn et al., 1995, Jørgensen et al., 1999). Additionally, elevations increased by strong volcanic activity during the mid to late Tertiary (25–2.5 Ma ago). In the central and northern Ecuadorian Andes, volcanic activity continued during the Quaternary, depositing thick ash layers throughout the Andes of northern and central Ecuador, including a 50 kilometers wide stroke of the Amazonian plains eastwards from the foothills of the Andes (Jørgensen et al., 1999). The Eastern Andes of Ecuador are mainly formed of Precambrian metamorphic rock composed of crystalline schist’s, with volcanic intrusions of granite (batholiths). Close to sample areas Sumaco and Tena (1 and 2), the Cordillera Galeras is mainly composed of Cretaceous Napo limestone. Near sample area Macas the Cordillera de Cutucú contains a mix of Napo limestone and older Jurassic sedimentary rocks. In between the Cordillera de Cutucú and the Cordillera del Cóndor at the southern side, the lowland sample locations of Lake Ayauchi and Kumpak were taken. The Cordillera del Cóndor is constituted of sandstones, shale’s and sandstones from the Tertiary and Mesozoic eras (Jørgensen et al., 1999). Where drainage in sample areas does not originate from such local Cordilleras, sediments are most likely to originate from the west, where the eastern Andes arise.

Soil formation in the Andean-Amazonian corridor is strongly influenced by the prevailing drainage direction of the precipitation. As moist air masses from the Amazons hit the eastern side of the Andes, the resulting precipitation flows back eastwards to the lower Amazonian plains, transporting Andean sediments that are relatively young compared to the

(14)

10

original soils of Amazonia (Hoorn et al., 2010). These Andean sediments still contain a relatively high amount of nutrients, notably Ca, P and Mg. As a result, deposition of Andean sediments is assumed to be one of the principal drivers of high biodiversity in western Amazonia (Hoorn et al., 2010). Ferralsols, old weathered nutrient poor soils that dominate the greater Amazons, are absent in the western Amazons (Quesada et al., 2011).

Generally, soil formation in the Amazons probably started between one and two million years ago (Quesada et al., 2011).Floodplain soils of western Amazonia are formed in the

Pleistocene and Late Holocene, and not much older than 5000 years. The western Amazons include large areas where leptosols on hill slopes dominate. In such shallow soils weathering of parent material forms an important contribution to soil fertility (Quesada et al., 2011). Cambisols also occur frequently in the study area. Cambisols are characterized by starting horizon differentiation and are found most frequent on hill slopes in western Amazonia and the foothills of the eastern Andes. When subsoil weathering exceeds erosion, cambisols develop into more mature soils. The fertility of cambisols varies with the quality of the weathering substrate (Quesada et al., 2011).

4. Materials & methods

4.1. Environmental gradients

In attempt to find the principal drivers behind the variance in the taxonomic composition between the different sampling locations, 11 environmental gradients were assessed (Tabs. 1 and 2). Climatic information was derived from climate-data.org (see also appendix B). Average temperatures on sites were calculated by adapting the value from the nearest weather station with a comparable elevation using a lapse rate of 5°C/ km. Standard deviations of the average temperature and precipitation within a year were calculated from the average values of 12 months. The (average) intra annual temperature range was calculated distracting the minimum temperature from the maximum temperature within an average year. Distances of sample sites to local and national roads were assessed using the distance ruler tool in Google Earth (Pro version 7.3.2.5487). Estimation of forested, open and agricultural area was executed by drawing a circle with a diameter of 5 kilometers around the sampling locations using the circle option of the ruler function in Google Earth, and subsequently scrutinizing the land use within the circle.

(15)

11

Table 1: set of environmental gradients selected to explain taxonomic plant variance

Table 2: sample location, elevational ranges, number of samples, mean annual temperatures and mean annual precipitation

(16)

12 4.2 Sampling strategy

At each site 1 to 16 samples were taken at different locations (Tab. 2). At each sampling coordinate (appendix A), 10 pinches of soil surface samples were collected within a 20 m radius after the leaf litter (if present) was brushed aside (personal communication McMichael, C.H., October 2016). This strategy smoothens differences in phytolith

assemblages between similar plots as reported by Morcote-Rios et al. (2016), Piperno (1988) and Piperno (2006).

4.3 Laboratory methods

4.3.1 Phytolith preparation

Since the total volume of phytoliths often consist of a relatively small portion of the soil sample, the phytoliths should be separated from the soil and concentrated in order to allow observation under a microscope (Piperno, 2006). For phytolith extraction and slide

preparation, 8 successive steps were taken that deviate at some points from the method as described in Piperno (2006), but nevertheless yielded good results (personal communication Philip, A., October 2016).

1) Approximately 1.5 cc of soil was heated in 33% hydrogen peroxide (H2O2) in order to get rid of organic material through oxidation.

2) Residual material was treated with a saturated permanganate (MnO42−) solution to

further oxidize organic material.

3) Residual material was sieved (212 µm) in order to remove the largest soil particles, which normally contain only few phyloliths, and roots and other redundant materials (Piperno, 2006).

4) Rinsing with demineralized water to wash away chemicals. This procedure is repeated until the demineralized water remains clear.

5) The residual material was cooked in 10% pyrophosphate (P2O74-

)

to remove clay

remnants.

(17)

13

7) Heavy liquid flotation separates the phytoliths from other particles using differences in specific weight of phytoliths (1.5 – 2.3 kg/l) over quartz (2.65 kg/l) (Piperno, 2006). In order to achieve this, the residual material of step 6 was centrifuged for 10 minutes at 1500 rpm in a bromoform (CHBr3) solution with a specific weight of 2.3 kg/l. The

density of the heavy liquid solution should be checked with a hydrometer or by weighting 1 ml of it to three decimal places on an analytical balance (Piperno, 2006). The phytolith fraction floating at the top of the tube should be removed with care and placed into a clean tube. The procedure was repeated for the sample remains several times in order to retrieve most of the phytoliths (Piperno et al., 2006). The removed phytolith fraction was decanted with ethanol in order to decrease the specific weight of the liquid. This allows the phytoliths to be separated from the liquid by centrifuging (personal communication, Philip, A., October 2016).

8) In the final step, the phytolith samples were prepared for microscopic analysis by putting a few drops of the extract (step 7) resolved in ethanol on slides (one for each sample). The phytoliths on the slides were subsequently embedded using Naphrax. Naphrax possesses a refractive index of 1.74, which is required for microscopic phytolith analysis (personal communication Philip, A., October 2016)

For a more elaborate description of phytolith preparation methodology, the different

possibilities and the reasons behind, it is recommended to read chapter 5 of Piperno (2006).

4.3.2 Phytolith identification

At least two hundred phytoliths were counted on each of the 80 samples. Taxonomic association of the observed phytoliths was achieved by microscope study, comparing to reference literature, photographing and counseling specialists (C. McMichael, D. Piperno).

4.3.2.1 Microscope and photography

Observation and photography of the phytolith bodies was carried out using a Zeiss Axiphot microscope (using a Zeiss Plan Neofluor 40/ 0.75 (object magnification = 400) and a Zeiss Plan Apochromat 63/ 1.40 Oil lens) and a Fuji Film X-M1 digital camera. Photos were made using the Zeiss Plan Apochromat 63/ 1.40 Oil lens and a total object magnification of 630. The Zeiss Axiphot microscope is based on Nomarski Differential Interference Contrast (DIC) transmitted light technique. Nomarski DIC microscopy uses a prism to divide a light beam before it passes specimen under the microscope, and more prisms to bring the image together again for observation. The advantage of this technique is that it renders more contrasts to the gradients of transparent specimen by creating a shadow effect and increases sharpness. The resulting image depicts very fine surface texture but provides limited information on shape (Sivaguru et al., 2012). In the case of phytolith identification Nomarski DIC microscope technique allows better discrimination of taxa that differ in surface details (personal communication, McMichael, C.H., December 2016).

4.3.2.2 Key references

Key references for morphology based phytolith identification during the microscope research came from Piperno (2006) for grass (Fig. 7) and tree taxa (Fig. 8.1-2), Morcote-Rios et al. (2016) for palm taxa (Fig. 9) and Morcote-Morcote-Rios et al. (2015) for additional information on grass taxa.

(18)

14

Figure 7: grass phytolith types and the sub family they belong to (left). Source: Piperno (2006)

Figure 8.1-2: reference images of the most common arboreal phytoliths encountered during the microscope research. Left: small rugose arboreal phytoliths. Right: large rugose arboreal phytoliths. Source: Piperno (2006).

(19)

15

Figure 9: reference images of palm phytoliths. A: globular echinate symmetrical (Ammandra

decasperma). B: globular echinate (Attalea butyraceae). C: globular echinate elongate (Attalea

butyraceae). D: globular echinate with short acute projections (Oenocarpus bataua). E: reniform echinate

(Oenocarpus bataua). F: globular echinate with long acute projections (Geonoma orbignyana). G: conical (side view) (Bactris sphaerocarpa). H: conical (top view) (Bactris sphaerocarpa). I: globular echinate symmetric (Geonoma camana). Source: Morcote-Rios et al., 2016

(20)

16 4.4 Quantitative analyses

4.4.1 Relative abundances

Relative abundances of each phytolith morphotype were calculated for each sample. The relative abundances of phytolith morphotypes with an average abundance >3% were visualized for each sample location using C2 software. Taxa indicating human disturbance (Chloroideae 1, maize, Heliconia and a formerly unknown arboreal phytolith we called “Moonface”) were included in the visualization even although their average abundance was lower than 3%.

Relative abundances were compared across sampling sites and the set of measured environmental gradients that was selected because they were anticipated to impact the taxonomic composition most.

4.4.2 Correlations

Pearson’s correlations were computed using PAST (v2) and MATLAB a2017 software, in order to assess respectively:

- cross correlations between environmental gradients. In case of high absolute cross correlation (>0.8-0.9) between environmental gradients, only one of these explanatory variables was included in further analysis in order to avoid multicollinearity (Graham, 2003; Legendre & Legendre, 1998).

- correlations between environmental gradients and taxa, to quantify how much the abundance of individual taxa changes over each environmental gradient. Correlations between the three principal different vegetation types (grass, arboreal taxa and palm taxa) were also computed, as well as correlations between phytolith morphotypes related to human disturbance (the sum of phytoliths from maize, Chloroideae 1, Heliconia and arboreal Moonface of each sample) and the distance to respectively the nearest local and national road. Only of significant correlations scatter plots with regression line and Pearson’s correlation value were included in the results.

4.4.3 Canonical Correspondence Analysis

In order to relate variance in taxonomic composition to potential drivers in community ecology, statistical ordination can be applied to ordinate taxonomic abundances and sample units along a set of environmental gradients (Gauch & Gauch, 1984; Palmer, 2004). During this study, total phytolith assemblages of all sampling sites were analyzed using an

ordination technique called canonical correspondence analysis (CCA). CCA computes and visualizes the degree of correspondence between measured environmental gradients, counted taxonomic compositionandsample sites(Palmer, 2004;Buttigieg & Ramette, 2014). CCA is a multivariate direct gradient analysis (DGA) technique. DGA computes what variance in taxonomic composition is explained by measured environmental gradients, thus directly assessing correspondence between taxonomic variance and potential drivers (McGarigal et al., 2000). This in contrast to indirect gradient analysis techniques that use redundancy in the sample - taxonomic frequency table to expose the total variance in taxonomic composition between the samples. Environmental -information, if available, is not involved in indirect gradient analysis,leaving it up to the analyst to try to figure out which gradients might be responsible for the variance in the taxonomic composition (Palmer, 2004; Buttigieg & Ramette, 2014; Gauch & Gauch, 1984).Toquantify to what proportion the chosen environmental gradients explain the total taxonomic variance, the explained variance can be divided by the total variance (Buttigieg & Ramette, 2014).

CCA is calculated by performing redundancy analysis on a matrix of samples (rows) and counted taxonomic abundance (columns), using the phytolith total per sample (row totals) as a weighting parameter. As a result, exclusively the taxonomic abundances that are maximally correlated to linear combinations of the included environmental gradients are

(21)

17

ordinated in a Euclidean space (Buttigieg & Ramette, 2014). This Euclidian space contains a number of axes, resulting from Eigenanalysis, equal to the number of included

environmental gradients. However, these axes do not directly represent the environmental gradients but linear combinations of them (Buttigieg & Ramette, 2014). These canonical axes are by definition orthogonal to each other, excluding shared variance, and

hierarchically ordered based on their correspondence with the taxonomic variance. The strength of this correspondence is expressed by their Eigenvalue, which represents a share of the total variance. The correlation of the environmental gradients with the canonical axes quantifies how much variance of each canonical axis corresponds with variance over each environmental gradient (Buttigieg & Ramette, 2014).

In CCA, significance is tested by a permutation procedure that repeatedly

randomizes the data are after which the test statistic is recalculated (often 999 times). Test results are ordered hierarchically. If the test results of the original data is higher than 95% of all results, the chance that this result happened by coincidence is smaller than 5%, allowing rejection of a null hypothesis that states that the real result happened by chance, and call the result significant (Palmer, 2004).

Being a technique based on multiple regression, CCA offers several advantages (Palmer, 2004). CCA can discover patterns in taxonomical abundance related to more than one environmental gradient. Single regression would not reveal this. Additionally,

assumptions are few for CCA. Environmental gradients over sampling sites should exhibit variance, and the chosen environmental gradients should relate causal and linear to the included taxa (Buttigieg & Ramette, 2014). There are no distributional assumptions to explanatory variables and CCA is robust to skewed taxonomic distributions and high noise levels (McGarigal et al., 2000, Palmer, 1993).

Some caveats should be taken in account (Palmer, 2004). Direct causation might not be assessed, as correspondences can be caused by factors not included in the analysis. The choice of explanatory variables and the quality of their data affects the interpretability of the CCA results. Moreover, CCA solutions can suffer from the “arch effect”, referring to

distortion in the computed results of the second and higher axes. This is usually more prominent when using a high number of explanatory variables. Therefore, it is advisable to remove strongly correlated environmental gradients from the analysis.

CCA results can be visualized in a triplot (Fig. 10). Triplots depict the two most important canonical axes, which is usually sufficient to capture a large part of the explained variance. Environmental gradients are visualized as vectors or arrows. The length of the arrow represents the impact of the corresponding environmental gradient on the taxonomic composition (McGarigal et al., 2000). The arrowhead (Fig. 10) indicates the direction of increase. The correspondence of these arrows with the resulting ordination determines the impact of the environmental gradients on the taxonomic composition (Buttigieg & Ramette, 2014). A small sample-to-taxa distance generally indicates these taxa are more abundant within that sample (Buttigieg & Ramette, 2014).

(22)

18

Figure 10: Canonical Correspondence Analysis triplot. The arrows depict quantitative explanatory variables (nutrient concentrations in this case). Filled circles depict objects, like for example sample sites. Hollow circles depict response variables (e.g. taxonomic abundances). Filled triangles depict the states of categorical explanatory variables, in this case clay, silt, or sand dominated sediment type. Source: Buttigieg & Ramette, 2014.

4.4.4 Data management and software

In this study correlations and scatter plots were computed in Matlab R2017a.

Correlation tables and CCA were computed using PAST (v2) software. CCA was

executed twice: once with downweighting in order to minimize bias as a result of

strong impact of rare taxa and outliers (Cao et al., 2015), and once with the original

data. Downweighting comprised of taking the square root of the original taxa data

(Toledo, 2010). As the downweighted CCA results differed only marginally from

those executed with the original data, they were placed in appendix C.

One phytolith morphotype of unknown origin was encountered. As it was

diagnosed as a human disturbance indicator (personal communication, Piperno, D. &

McMichael, C.H., December 2016), it was included in CCA.

5. Results

5.1 Phytolith morphotypes

A total of 29 phytolith morphotypes were identified in 80 samples.

The phytoliths included 6 arboreal, 15 grass and 8 palm morphotypes. One arboreal morphotype of unknown origin was encountered.

Arboreal phytholiths

One unidentified arboreal phytolith morphotype (Fig. 11) was encountered in low numbers (1-5) in 11 samples, and 51 times in one particular sample (Ayauchi 126, consisting of 20.2% of all morphotypes counted in that sample). This phytolith was identified by

specialists as an arboreal taxon indicating human disturbance (personal communication, Piperno, D. & McMichael, C.H., December 2016). For its resemblance with the face of the moon it was named “moonface”.

(23)

19

Figure 11: “Moonface” phytolith, an arboreal human disturbance indicator (personal communication Piperno, D. & McMichael, C.H., December 2016)

Small (Figs. 12.1-3) and large (Figs. 13.1-4) rugose phytholiths are irregular or lobbed morphotypes that can be distinguished by size (Piperno, 2006). Small rugoses are ~10 µm, while large rugoses range about 15-20 µm. Small rugoses can also be encountered with a smooth surface, and can sometimes be found attached to one another in pairs (Fig. 12.3).

Figure 12.1-3: Small rugose phytoliths

Figure 13.1-4: Large rugose phytoliths

Marantaceae phytoliths (Figs. 14.1-3) present a nodular surface with unevenly distributed prominences. Sizes range from 9 to 20 µm (Piperno, 2006).

Figure 14.1-3: Marantaceae phytoliths

Since Annonaceae phytoliths were encountered only four times in the samples, no image of sufficient quality could be generated. Therefore, an image from Neumann et al. (2009) is depicted below (Fig. 15).

(24)

20

Heliconia phytoliths are clearly recognizable by a deep trough in the middle of their mostly conical shape. Both leaves and rhizomes of Heliconia produce these morphotypes, however, those from the rhizomes are bulkier and larger (up to 55 µm) than those from the leaves (12 -25 µm) (Piperno, 2006). Since Heliconia was only encountered twice, no image of sufficient quality could be generated. Therefore, an image from Piperno (2006) is depicted (Fig. 16).

Figure 16: Heliconia phytolith (Piperno, 2006)

Grass phytoliths Bambusoideae

The following six phytolith morphotypes from the Bambusoideae family could be identified.

The tall saddle phytolith (Figs. 17.1-2 and Bambusoideae 1 in Fig. 7), often

measuring 15 µm or more, can be distinguished from other saddle shaped morphotypes (i.e. Chloroideae) by the longer seat of the saddle shape (Piperno, 2006; Kondo et al., 1994; Lu & Liu, 2003).

Figure 17.1-2: Tall saddle (Bambusoideae 1) phytoliths

The Bambusoideae 3 phytolith (Fig. 18). The pointed extremities make this morphotype clearly recognizable, although similarity in shape can be found with

Ehrhartiodeae (the middle variant in Fig. 7). However, the bulkier body allows discrimination of Bambusoideae 3 from Ehrhartiodeae.

Figure18: Bambusoideae 3 phytolith

The cross-shaped Bambusoideae morphotype (Fig. 19.1-3, Bambusoideae 4 in Fig. 7) can be distinguished from maize phytoliths by its more compact blocky body and smaller size (10-15 µm).

(25)

21

Figure 19.1-3: Cross-Bam (Bambusoideae 4) phytoliths

The rondel phytolith or Bambusoideae 5 (Fig. 20) is recognizable by its regular 3D shape. It can be distinguished from the long rondel morphotype (Pooideae 6 in Fig. 7) by its more compact body.

Figure 20: Rondel phytolith (Bambusoideae 5)

The two Spiked Irregular morphotype (Figs. 21.1-2), a bulky irregular phytolith of roughly 15µm with two distinctive spikes on one end, is produced by the bamboo grass species Chusquea grandiflora (Bambusoideae 6 in Fig. 7).

Figure 21.1-2: Two Spiked Irregular phytoliths/ Bambusoideae 6 (Chusquea grandiflora)

The collapsed saddle phytolith (Bambusiodeae 7, Figs. 22.1-6) is similar in size (~15µm) to Chusquea grandiflora, but less bulky due to a clear cut-out with a saddle shape. Additionally, this morphotype is less irregular and lacks the two spikes of Chusquea

grandiflora (Piperno, 2006).

(26)

22

Ehrhartoideae

Grass phytoliths of the Ehrhartoideae subfamily are also called scooped bilobates, due to their morphology (Piperno, 2006). Their most characteristically feature are two sharp little points on both extremities. There are three varieties (Fig. 7), of which the more slender (Fig. 23.1) and the most slender variety (Figs. 23.2-6) are depicted below.

Figure 23.1-6: Ehrhartoideae phytoliths

Pooideae

Of 7 phytolith morphotypes from the Pooideae subfamily (Fig. 7), only the long rondel (Pooideae 6) was encountered (Fig. 24).

Figure 24: Long rondel phytolith

Aristidoideae

Of both Aristidoideae phytolith morphotypes (Fig. 7), only the long slender shafted bilobate variety was identified (Fig. 25).

Figure 25: Bilobate Long Shaft (Aristidoideae 1) phytolith

Chloridoideae

Since the first variety of the Cloridoideae phytolith, the squat saddle morphotype (Fig. 7), was only encountered six times, no image of sufficient quality could be provided.

Therefore, an image from Piperno (2006) is depicted (Fig. 26.1). De second variety of the Cloridoideae phytolith (Figs. 26.2-3) is also called tall saddle bilobate for its large size and shape between bilobate and saddle-like (Piperno, 2006).

(27)

23

Figure 26.1: Squat saddle phytoliths Figure 26.2-3: Tall saddle/ bilobate (Chloridoideae 2) phytoliths (Cloridoideae 1). Piperno (2006)

Panicoideae

Three phytolith morphotypes of the Panicoideae subfamily (Fig. 7) were encountered. The first variety (Figs. 27.1-3), the bilobate morphotype, is a tall phytolith with rounded lobes and a clearly distinguishable shaft.

Figure 27.1-3: Bilobate (Panicoideae 1) phytoliths

The second Panicoideae variety (Fig. 28) is a cross-shaped phytolith that can be distinguished from maize phytoliths by its smaller size (10-12.5µm, Piperno, 2009), and from the cross shaped Bambusoideae (Bambusiodeae 7) by its more round and less compact body.

Figure 28: Cross-Pani (Panicoideae 2) phytolith

The third encountered Panicoideae variety (Fig. 7, Panicoideae 4) has a short rondel shape. Due to shape variability, especially under different angles, it was not easy to provide an image of a well recognizable short rondel. Fig. 29 was the best image taken.

Figure 29: Short rondel (Panicoideae 4) phytolith

Maize phytoliths (Panicoideae, Figs. 30.1-7) look similar to those of several wild grasses, especially Panicoideae 2 (Fig. 28), but are larger (12.7-15 µm, Piperno, 2009).

(28)

24

Figure 30.1-7: maize phytoliths

Palm tree phytoliths (Arecaceae)

The conical palm phytolith (Figs. 31.1-10) comes from the Bactris sphaerocarpa species. It can be recognized by its hat shape, wider at the base than the apex (Morcote-Rios et al., 2016).

Figure 31.1-6: Conical palm (Bactris sphaerocarpa) phytoliths (side view). Image 4 and 6 have been burnt

The verrucate outer circle of the phytolith of Bactris sphaerocarpa has a granulate surface structure, while the centre exhibits a variable number of granules (Figs. 31.7-10). The conical palm phytolith ranges in length between 1.6-20µm and in height from 1.4-9µm (Morcote-Rios et al., 2016).

Figure 31.7-10: Conical palm (Bactris sphaerocarpa) phytoliths (top view)

The globular echinate phytolith (Figs. 32.1-6) is produced by the palm species Attalea butyracea. It presents triangular but not very regular shaped nor divided spikes over

(29)

25

its surface. 10-12 peripheral projections can be counted under the microscope. Diameters range between 4-18µm (Morcote-Rios et al., 2016).

Figure 32.1-6: Globular echinate (Attalea butyracea) phytoliths

The globular echinate elongate phytolith (Figs. 33.1-2) is also produced by the palm species Attalea butyracea. It looks like a globular echinate phytolith that has taken an

elliptical shape after being squeezed (Bowdery, 2015; Morcote-Rios et al., 2016). Likewise, it presents 10-12 peripheral projectionsunder the microscope. Diameters of its longest axe range between 6-15µm (Morcote-Rios et al., 2016).

Figure 33.1-2: globular echinate elongate (Attalea butyracea) phytoliths

The globular echinate phytolith with long acute projections (Fig. 34) from the

Geonoma orbignyana palm species presents a limited number of asymmetrically distributed projections with a length of half the diameter of its body or more (Morcote-Rios et al., 2016). Its longest axis measures 6.8-8.8µm.

Figure 34: Globular echinate phytolith with long acute projections (Geonoma orbignyana)

The big globular echinate phytolith with short acute projections (Figs. 35.1-2) is produced by the palm species Oenocarpus bataua. It can be distinguishably large, with a diameter ranging between 8.3 and 33.3 µm. Its projections are dense, short and numerous compared to other palm tree phytoliths, giving it a sinuate periphery, while its dense

(30)

26

Figure 35.1-2: Globular echinate phytoliths with short acute projections (Oenocarpus bataua)

The globular echinate symmetrical phytolith (Figs. 36.1-2) is produced by the palm species Ammandra decasperma. It can be distinguished by triangular projections that are divided symmetrically over its surface. Usually, there are 8-10 peripheral projections visible under the microscope. Diameters range between 4-19 µm(Morcote-Rios et al., 2016).

Figure 36.1-2: globular echinate symmetrical phytoliths (Ammandra decasperma). Rare

The reniform echinate phytolith (Fig. 37), produced by the palm species Oenocarpus bataua, presents short acute projections at its periphery. Surface ornamentation is granulate. The longest axe ranges between 4.9 and 10.3 µm(Morcote-Rios et al., 2016).

Figure 37: Reniform echinate (Oenocarpus bataua) phytolith. Rare

The globular echinate symmetric phytolith, produced by the palm species Geonoma camana, resembles most a globular echinate elongate phytolith with inward curved ends (Morcote-Rios et al., 2016). Since it was only encountered 6 times in the dataset, no good image could be provided. Therefore, an image from Morcote-Rios et al. (2016) is depicted (Fig. 38).

Figure 38: Globular echinate symmetric (Geonoma camana) phytolith. Source: Morcote-Rios et al., 2016

5.2 Results of the statistical analysis

5.2.1 Relative abundance of the phytolith morphotypes

Relative abundances of all phytolith morphotypes with an average abundance >3% were visualized (Fig. 39). Phytoliths indicating human disturbance with an average

(31)

27

Figure 39: relative abundance per sample of all phytolith morphotypes with an average abundance >3%. Indicator taxa of human disturbance with an average abundance ≤ 3% are also displayed. The numbers on the Y axis relate to the identity number of the sample sites as displayed in the table of appendix A

5.2.2 Pearson‟s correlations between environmental gradients

A Pearson’s cross-correlation table of the environmental variables was computed (Tab. 3).

Table 3: Pearson‟s cross-correlations between the environmental variables

Table 3: Pearson‟s cross-correlations between the environmental variables. A description of the environmental variables can be found in paragraph 4.1

The average temperature (negatively) correlates very strong with elevation (r = -0.99034). Since elevation was assessed on the sampling location itself, it was likely to be more accurate than average temperatures calculated based on data from the nearest weather station and a lapse rate of -5 °C/km. Therefore, annual average temperature was

(32)

28

removed from further analysis. Additionally,very strong correlations were calculated between the percentages of forested area and open area (-0.9975), forested area and agricultural area (-0.90648) and the percentages of agricultural area and open area (0.90373). Open area excludes forest area, while ~90.4% of open area is agriculture. In order to simplify analysis, only open area was kept in further computation.

5.2.3 Scatter plots of normalized grass phytoliths versus environmental

gradients with significant Pearson‟s correlation

The slopes of the least-squares reference lines in the scatter plots are equal to the displayed correlation coefficients.

Figure 40: Scatter plot with Pearson‟s r between normalized grass phytolith abundance and Elevation (in m)

Figure 41: Scatter plot with Pearson‟s r between normalized grass phytolith abundance and Temperature range (in °C)

Figure 42: Scatter plot with Pearson‟s r between normalized grass phytolith abundance and the Standard deviation of the average temperature

Figure 43: Scatter plot with Pearson‟s r between normalized grass phytolith abundance and Distance to a national road (x 10 km)

(33)

29 5.2.4 Scatter plots of normalized palm phytoliths versus environmental

gradients with significant Pearson‟s correlation

5.2.5 Scatter plot of normalized arboreal phytoliths versus distance to a

national road with significant Pearson‟s correlation

Normalized arboreal phytoliths correlate significantly with distance to a national road. Figure 44: Scatter plot with Pearson‟s r between

normalized palm phytolith abundance and Elevation (in m)

Figure 45: Scatter plot with Pearson‟s r between normalized palm phytolith abundance and Temperature range (in °C)

Figure 46: Scatter plot with Pearson‟s r between normalized palm phytolith abundance and Distance to a national road (x 10 km)

(34)

30

5.2.6 Scatter plot of normalized human disturbance indicator phytoliths versus

distance to a local road with significant Pearson‟s correlation

5.2.7 Results of Canonical Correspondence Analysis

5.2.7.1 The canonical axes

The Eigenvalues of the canonical axes/ were ordered hierarchically (Tab. 4). The

Eigenvalues equal the share of total variance of the responsive taxa. The largest variance is represented by axes 1 and 2, adding up to 36.4 % of the total variance and 77.6% of the explained variance. Axes 5-8 did not yield significant results (p > 0.05). Therefore, the involved environmental gradients can explain 44.1 % (sum of the significant Eigenvalues * 100%) of the total variance and 95.3 % of the explained variance in the taxonomic

composition with a confidence level of 95%.

Table 4: Eigenvalues of the canonical axes

Axis Eigenvalue Explained variance% P 1 0,18859 40,77 0,001 2 0,17507 37,84 0,001 3 0,046859 10,13 0,001 4 0,030319 6,554 0,001 5 0,01194 2,581 0,193 6 0,0061066 1,32 0,387 7 0,003732 0,8067 0,307 8 1,30E-08 2,801E-06 0,076 Significant Totals 0,440838 95,294

Table 4: the canonical axes, their Eigenvalues, the explained variance they represent and significance of the result. P values of axes that did not yield significant results (p ≥ 0.05) are red, non significant Eigenvalues and explained variance blue

5.2.7.2 CCA plots

Environmental gradients

The vector length of the distance to a national road is 0.705, that of the annual temperature range 0.702 and that of elevation (-temperature) 0.687 (tab. 5 and fig. 48). The vector length of the standard deviation of the average temperature is 0.523. The vector length of the

intra-Figure 47: Scatter plot with Pearson‟s r between normalized palm phytolith abundance and Distance to a local road (in m)

(35)

31

annual variability of the precipitation is 0.368. The vector length of the distance to a local road is 0.348.The vector length of the standard deviation of the average annual precipitation is 0.363. The vector length of the open area is 0.27, that of annual precipitation 0.151.

Table 5: lengths of the environmental gradient vectors

Environmental gradient Vector length

Dst N road 0,704752 T range 0,702425 Elevation 0,686744 Std Avg T 0,523438 Std precip 0,368485 Dst Loc road 0,347513 Open area 0,274963 Precip 0,150789

There is strong correspondence between the direction of the vectors of precipitation, open area, elevation and to a lesser extend the average intra annual temperature range. The variability of the average temperature within a year opposes the direction of the earlier mentioned vectors, especially elevation (Fig. 48). By assessing similarity of direction, it becomes clear that several environmental gradients are correlated. This has been quantified in table 3, paragraph 5.2.2.

CCA plot of the environmental gradients

Figure 48: CCA plot of the environmental gradients

The location of sample spots within the CCA diagram (Fig. 49)

Perpendicular projections of most Tena High (= Tena 2) samples on the (imaginary

prolongations of) the environmental gradient vectors would end up at the far ends of T range and Elevation. Tena High samples are far from all other samples, except most Sumaco samples. Sumaco samples are not close to one another, although they are found mostly in the left half of the triplot. Most Kumpak samples are far away from other samples, but most close to Tena samples. Samples of Ayauchi, Tena, Palmitas and Macas Puyo are mostly near to each other and overlap.

(36)

32 CCA biplot of the environmental gradients and the sample sites

Figure 49: CCA biplot of the sampling sites and en environmental gradients

(37)

33 CCA triplot of the environmental gradients, sample sites and taxa (phytolith morphotypes)

(38)

34 6. Discussion

A factor that can complicate interpretation of CCA is the presence of taxa with low occurrence. Due to the calculation process, these taxa are found outside the cloud of site points in CCA plots and do not conform well to the general rules of interpretation of CCA (Legendre & Legendre, 2012). Whether such taxa should be transformed or removed for better CCA is a subject under debate. Some specialists argue that taxa with low occurrence can be removed from the data table without major change to the overall result, as they have only small influence on the first few Eigenvalues and axes of CCA (Gauch, 1982; Legendre & Legendre, 2012; McCune et al., 2002). Others state that removal of taxa in community ecology is undesirable as rare taxa can be important indicators of ecosystem stress and should therefore be avoided mostly (Cao et al., 1998; Faith & Norris, 1989; Palmer, 2004). A third point of view is that CCA performs best when taxa with low occurrence are

downweighted by transforming all abundance data (Cao et al., 2015). The CCA triplot result indeed depicts several taxa with low occurrence outside the cloud of site points. The position relatively far away from the centre in the CCA plots of reniform echinate (Oenocarpus Batau), “moonface”, Annonaceae, Heliconia, squat saddle and maize phytolits (Fig. 50) is likely a result of low occurrence rather than informative. Therefore, downweighting of taxa with low occurrence was applied by taking the square root of the taxonomic abundances (appendix D). However, this did not significantly change the CCA results, nor did it facilitate interpretation.

Based on interpretation of the CCA analysis (i.e. Fig. 48), it can be stated that the corresponding directions between the vectors of elevation and the average temperature range suggests generally stronger differences between maximum (day) temperatures and minimum (night temperatures) at higher elevation. This is confirmed by Jørgensen et al. (1999). The distance to a national road, the annual temperature range and elevation (= -temperature) affect the taxonomic plant composition in the Andean-Amazonian corridor of Ecuador most. The variability of the average temperature within the year has a lower impact. In descending order, the variability of the precipitation within an average year, the distance to a local road and the amount of open area further affect the taxonomic plant composition. Precipitation has the lowest impact on taxonomic plant composition of the measured environmental gradients. Strong impact of temperature related gradients was anticipated. However, the low impact of precipitation related gradients is unexpected. A possible explanation is limited reliability of the precipitation data.

The distance to a national road has most impact on taxonomic composition (gradient length = 0.705), while the impact of the distance to a local road is relatively limited (0.348). This suggests that although the taxonomic composition of vegetation in de

Andean-Amazonian corridor of Ecuador is affected most by more intense human disturbance, it is relatively resilient to moderate disturbance.

The impact on the taxonomic composition of open area, and with that also

agricultural area and forest was relatively low (0.277). Nevertheless, the correspondence between the vectors representing the elevation and the amount of open area suggest that there is more open area at higher elevations. Additionally, the upper left corner of the CCA plot (Fig. 50), at the far end of (imaginary prolongation lines of) the elevation and the average temperature range, is dominated by grass taxa. This is confirmed by a significant Pearson’s correlation of 0.56 between elevation and grass taxa (Fig. 40). The significant Pearson’s correlation of grass taxa with the temperature range is even slightly stronger (0.59, Fig. 41). Additionally, the abundance of grass taxa has a significant negative Pearson’s correlation (-0.38) with the seasonal (intra annual) temperature variability that, according to the CCA analysis, is stronger at lower elevations (elevation and Std Avg T oppose each other in Fig. 48). Moreover, the abundance of grass taxa has a significant negative Pearson’s correlation (-0.39) with the distance to a national road, likely to be caused by opportunities offered by clearance of closed vegetation by humans. Expectations

(39)

35

of higher abundance of arboreal phytoliths starting at elevations ~1200 masl upwards, due to increased moisture availability as a result of montane cloud formation, were not confirmed by CCA analysis, nor by a significant correlation between elevation and arboreal taxa or their scatter plot (not significant and therefore not included in the results). Arboreal taxa do not exhibit a significant correlation with any environmental variable except for the distance to a national road (Pearson’s correlation = 0.58), likely to be caused by tree cutting or burning. Palm taxa do have a significant negative correlation with elevation (-0.25) and the temperature range (-0.48), confirming expectations that palms prefer higher temperatures, especially in the absence of larger temperature fluctuations. This is sustained by various publications on the relation between environmental factors and palm species distribution (Eiserhardt et al., 2011; Reichelt at al., 2018; Salm et al., 2007; Svenning, 2001). Phytoliths indicating human disturbance do not correlate significantly with the distance to the nearest national road, but do correlate significantly and negatively (-0.24) with the distance to a local road. As local roads often function to transport farmers from and to their laboral activities, this is not unexpected.

Similarities and differences between environmental drivers and sample plots were assessed by interpretation of CCA biplot Fig. 49. The relations between taxa, samples and gradients were assessed from CCA triplot Fig. 50. The average temperature range and elevation-temperature have the strongest impact of the taxonomic composition of Tena High samples, which differs clearly from that of other samples except some Sumaco samples. In Tena High and such Sumaco samples small rugose and Aristoideae 1 are more abundant. Sumaco samples range more in composition than other samples, related most to

temperature- elevation and the average temperature range. In general, at higher elevations, where temperatures are lower and range more, Bambusiodeae 1, Bambusiodeae 7,

Chusquea grandiflora, Bambusiodeae 5, Pooideae 6, Marantaceae, short rondel,

Oenocarpus bataua, Bambusiodeae 3 and 4, Aristoideae 1 and Chloriodeae 2 phytoliths are more frequent. Whereas in Tena high samples taxonomic composition is driven by elevation-temperature and the average elevation-temperature range, taxonomic composition in Kumpak

samples is related to low human impact, with higher abundances of large rugose arboreal and Geonoma camana (Arececeae) phytoliths. Most low elevation Kumpak samples are far away from national roads and differ from other samples, but less from Tena samples. Most samples from Ayauchi, Tena, Palmitas, Macas and Puyo are similar in taxonomic

composition, with increased abundance of (Arecaceae) palm conical/ Bactris sphaerocarpa, globular echinate/ Attalea butyracea, globular echinate elongate Geonoma orbignyana and Panicoideae 1/ (bilobate phytoliths) grass. As elevations of these samples range between ~300 and ~1800 masl, these taxa must have a relatively wide tolerance range for

temperature.

The 8 environmental gradients kept in the final analysis explain explain 44.1 % of the total variance and 95.3 % of the explained variance in the taxonomic composition with a confidence level of 95%. Measuring and involving more environmental gradients might increase the amount of total variance that can be explained. Additional environmental gradients that might relate well to taxonomic plant composition are soil type, mother material (geology), nutrient content, fire frequency, grazing pressure and light availability.

An additional step in this research could be calculation of C3:C4 grass ratios. C3:C4 ratios can be applied as indicators of temperature changes, as C4 grasses thrive at higher temperatures. Comparing such results to elevation or average temperature data of the sampling locations could give an indication of the accuracy of the results of this research.

Another interesting step could be comparison of taxonomic abundance result with other vegetation data from the sample spots, like vegetation surveys, when available. Non linear relationships between phytolith abundance and taxonomic abundance can be caused by variation in quantitative phytolith production between species, genera or plant families (Piperno, 1988). Calibration with accurate vegetation data allows calculation of correction factors that restore linear relations between phytolith abundance and taxonomic abundance.

Between the tropical Andes and the Amazon Characterizing phytolith assemblages over environmental gradients