Aarhus University Seal / Aarhus Universitets segl

Amazonian Biogeography

Amazonian Biogeography   

Integrating satellite and field data at a continental scale    

We test hypotheses about species distribution patterns and evolutionary processes across Amazonia by taking advantage of a novel combination of extensive field datasets and a new basin-wide Landsat TM/ETM mosaic. Our field data consist of 1700+ quantitative inventories already assembled; together they are among the largest floristic datasets for Amazonia, and by far the largest with standardized taxonomy.

Our new Landsat TM/ETM+ mosaic (Fig. 1) has solved technical problems (such as high cloud incidence and an artefactual east–west brightness gradient) that have hampered previous broad-scale analyses. These advances make it possible to test the hypothesis that:

  • Amazonian biogeography is more strongly shaped by geologically induced habitat differences than by dispersal barriers formed by rivers.

Past and present river barriers have until now been thought to be important in delimiting species distributions for both animal and plants and in acting as the main driver of allopatric speciation in Amazonia (Smith et al. 2014). However, recent research by us and others has suggested that compositional turnover at the landscape scale is more strongly related to habitat variation, especially differences in soils. In addition, such variation has been identifiable using Landsat data. Now we will assess to what degree the Landsat mosaic allows reconstructing edaphic gradients and patterns of compositional differentiation at the basin-wide scale. We believe that our research could help solve a long-standing mystery: How has Amazonia accumulated such high species diversity, even though obvious dispersal barriers that could promote allopatric speciation are almost lacking? If we are able to show that geological habitat differences are the main drivers of diversity, future studies of Amazonian biogeography will have to give edaphic variation a more central role. Our analyses will provide a basis for identifying priority areas that may harbor unique but as yet undocumented biodiversity. This will help in protecting Amazonia’s biota, which is highly threatened by deforestation and climatic change.

Fig. 1 Landsat TM/ETM+ mosaic
Fig. 1 Landsat TM/ETM+ mosaic

Field data on biota and soils — Most field data for this project had been collected already in 1700+ field inventories (Fig. 1) for ferns, palms, and melastomes using standardized taxonomy and including soil data. These data were collected by our teams at the universities of Aarhus (AU) and Turku (UTU) in collaboration with colleagues from several Amazonian countries. New field inventories are being made in two hitherto unexplored areas: 50 inventories in the Rio Negro basin in 2020 and 50 in Cayenne in 2021. Both areas draw attention in the Landsat image mosaic because of their unusual reflectance patterns, which suggests they may harbor undiscovered biodiversity. Field sampling follows our standard protocol of 5 ´ 500 m transects (Balslev et al. 2010; Tuomisto et al. 2003a), including three soil samples per transect.

Remotely sensed data — Spectral patterns that might indicate biotic differentiation are identified visually and with object-based image analysis in our Landsat image. Predictions about the degree of floristic and edaphic distinctness among patches, as well as abruptness and permeability of the identified boundaries is made on the basis of the spectral data and tested with field data. Rigorous testing of the predictive power of the Landsat mosaic in different contexts is one of our main aims. For instance, initial analyses combining Landsat and field data will assess how well the spectral values from the mosaic can predict soil properties and species occurrence, turnover, and richness patterns of the focal plant groups across Amazonia. We already know that species differ in their optima and ranges along edaphic gradients (Cámara-Leret et al. 2017; Tuomisto 2006; Zuquim et al. 2014), but so far lack of reliable environmental data layers has made it difficult to apply this knowledge for making predictive maps. Following the initial analyses, we identify the most important biogeographical boundaries and ecological subdivisions in Amazonia. We use the maximum entropy approach to produce species distribution models for each focal species from the different plant groups (Elith et al. 2011). Various environmental data layers, including the Landsat mosaic, are used to predict suitable areas for each target plant species. Then we assess whether the currently existing network of conservation areas (including indigenous reserves) covers the identified biotope diversity within Amazonia, and whether any biotically unique areas seem to be threatened by imminent deforestation.

Background - Amazonia is famous for its spectacular biodiversity and for being the world’s largest continuous rainforest area. Its survival is of great concern: deforestation and climate change may drive species to extinction (Colwell et al. 2008; Olivares et al. 2015). Failure to recognize environmental constraints on species occurrences could dramatically impact long-term preservation of the region’s unique biota and its sustainable use. Currently, knowledge on biodiversity in Amazonia is fragmented, which seriously hampers attempts to predict how it might react to foreseeable threats.

The traditional image of Amazonia as a vast uniform forest is shattering; floristic composition is now thought to be heterogeneous and strongly linked to geologically driven variation in soils (Tuomisto et al. 1995). For example, many species that dominate one soil type are entirely absent in others (Cámara-Leret et al. 2017; Kristiansen et al. 2012; ter Steege et al. 2013; Tuomisto et al. 2003a,c, 2016; Zuquim et al. 2014). Unfortunately, predicting species distributions and abundances is difficult because available soil maps are inaccurate and have low spatial resolution (Moulatlet et al. 2017). Nevertheless, it appears that landscape structure and soils can create dispersal barriers associated with differentiation (Tuomisto et al. 2016).

Access to most of Amazonia is difficult and field data are scanty, making remote sensing crucial for understanding biotic differentiation. We have used Landsat images and the Shuttle Radar Topography Mission (SRTM) elevation data to interpret the biotic significance of terrain characteristics at the landscape scale (Cordeiro et al. 2016; Tuomisto et al. 2003a,b). Recently, we reported a 1000-km boundary line, visible in Landsat images, that separate two edaphically and floristically distinct forest areas (Tuomisto et al. 2016). That boundary crosses several major rivers, contrary to the idea that rivers form the main dispersal barriers in Amazonia.

Ironically, standard methods have failed to produce spectrally consistent Landsat mosaics. Cloud cover has hampered comparison of adjacent areas and forest type differences have been swamped by the effects of the bidirectional reflectance distribution function (BRDF), which causes an artifactual east–west gradient in reflectance (Muro et al. 2016; Tuomisto et al. 2003a). We have recently compiled a Landsat TM/ETM+ image mosaic based on a full decade (2000–2009) of Landsat data (Van doninck & Tuomisto 2018), that provides a spectacular view of Amazonia’s landscape variability at 30-m resolution (Fig. 1). This major breakthrough makes it possible to see Amazonia with new eyes, making ecological inferences possible, for the first time across the entire Amazonia at a spatial resolution that hitherto has only been achievable at the landscape level.

Modeling and comparing species distributions — Our main objective is to develop a solid approach to species distribution modelling (SDM) in Amazonia and to compare SDM results among remotely related plant groups. We carry out SDM analyses for three major plant groups (ferns, palms, and melastomes) using our own species occurrence data from 1700+ sites supplemented with data from the Global Biodiversity Information Facility (GBIF, www.gbif.org). As explanatory data, we use variables derived from our Landsat mosaic and other remote sensing products such as SRTM, for hydrology, and cloudiness, and climatic data (CHELSA, www.chelsa-climate.org). There are significant differences in lifeform, size and dispersal mode, etc. among our focal plant groups. Palms are terrestrial, never epiphytic, understory and canopy plants, and bird and mammal dispersed; ferns are terrestrial and epiphytic, small and wind dispersed; melastomes are terrestrial shrubs and herbs dispersed by birds. We use SDM to test if these trait differences are reflected in the distribution and biogeography of each of the groups.

Testing biogeographical hypotheses — Specific biogeographical hypotheses that we test include the following. 1. Areas with similar reflectance values in the Landsat mosaic represent ecologically analogous forests even when geographically distant. 2. Spectral boundaries in the Landsat mosaic correspond to edaphic and floristic boundaries in the forest. 3. Palm-, fern-, and melastome-community compositions and biogeography respond similarly to the Landsat reflectance values. We test these hypotheses mainly using our own field datasets (Fig. 1), but whenever possible we amend these using occurrence record data available online (GBIF; SpeciesLink, www.splink.cria.org.br, BIEN (http://bien.nceas.ucsb.edu/bien/).

Assessing the ecological representativeness of conservation area networks — Due to high deforestation rates, protected areas are critical for conserving intact Amazonian forest (Rosa et al. 2013). We will assess how well the existing conservation area network captures the ecological and biogeographical heterogeneity of Amazonia. These analyses help to prioritize areas that should be targeted for conservation actions (Alvez-Valles et al. 2018). Other land use planning can benefit from the results as well, since soil characteristics have a big effect on primary productivity.