Species Distribution Modeling of Commercially Harvested Medicinal Plants in Benin, Africa Can sustainability problems be predicted?

BACHELORS THESIS

BACHELORS THESIS Stephanie Croft July 2012 Supervisors

Dr. N. Raes Dr. Ir. E.E. van Loon Dr. T.R. Van Andel

SUMMARY

1

1. INTRODUCTION 1.1 MEDICINAL PLANT USE IN WEST AFRICA 1.2 FOUNDATIONS FOR RESOURCE MANAGEMENT 1.3 SPECIES DISTRIBUTION MODELING FOR CONSERVATION PLANNING 1.4 EXPECTED OUTCOMES

2 2 2 2 3

2. METHODS 2.1 THEORETICAL FRAMEWORK 2.2 SPECIES DISTRIBUTION MODELING IN MAXENT

4 4 4

2.3

2.2.1 DATA COLLECTION & PREPARATION Study Area Distribution Data Environmental Predictor Variables 2.2.2 MODEL DEVELOPMENT 2.2.3 MODEL EVALUATION GIS-BASED CHARACTERIZATION OF SUITABLE HABITAT 2.3.1 DATA PREPARATION Land Cover Characterisation Species Ecology 2.3.2 GIS-MODELS Suitable Habitat Extraction Cost Path Distance to Markets

5 5 5 5 6 6 7 7 7 7 8 8 8

3. RESULTS 3.1 SPECIES DISTRIBUTION MODELING 3.2 COST DISTANCE TO MARKET AREAS 3.3 PREDICTING SPECIES VULNERABILITY 3.4 SUMMARY OF RESULTS

10 10 13 14 15

4. DISCUSSION

16

5. CONCLUSION

18

REFERENCES

GLOSSARY APPENDIX

Summary The aim of this research was to evaluate whether species distribution modeling and a GIS-based analysis of habitat suitability can be used to predict species vulnerability of commercially extracted medicinal plants in Benin, Africa. Market surveys conducted in West Africa indicate that commercialisation of wild-harvested medicinal plants may have significant impact on the sustainability of plant populations and pose a risk to the longevity of the medicinal plant market. In order to determine the potential impacts of commercialisation it is essential to acquire knowledge of species distributions. Generating geographic distributions can be challenged by limited occurrence data, deforestation and land use change, and the evaluation of model accuracy. This research employs species distribution modeling and spatial analysis of habitat suitability as a tool for predicting species vulnerability. Species distribution modeling was performed with MaxEnt due to its effective performance in previous research. Information of species ecology was then used to extract suitable habitat from predicted presence maps generated in MaxEnt. Thereafter, a GIS model was developed to evaluate whether suitable habitat of wild-harvested species could be used to quantify vulnerability of habitat with regards to proximity of habitat to transport networks and key markets where medicinal plants are sold. Geographic distributions were generated in MaxEnt for 23 species, 16 had statistically significant results. Habitat suitability was visualised in maps for both West Africa and Benin. Six species, five of which were sold frequently at Ghanian medicinal markets, reveal significantly reduced habitat. Four of these species are not recognised on the IUCN Red List. Population viability analyses were deemed advisable for Okoubaka aubreveleii, Sphenocentrum jollyanum and Piper guineense. Four at-risk savanna species were reviewed. Although their geographical range and suitable habitat did not indicate vulnerability these species are harvested for timber and their abundance unknown. A model developed to identify at risks areas concluded the analysis, evaluating the prospects for characterising geographic distributions to predict vulnerability. The model lacked in precision and complexity by which conclusions can be drawn. This research concluded that species distribution modeling and spatial analysis in GIS can be effective predictive tools for identifying potential impacts of commercialisation of wild-harvested species, but drawing conclusions on species vulnerability is conditional upon field-based analyses to verify predictions.

Introduction This section covers several aspects relevant to the principal aims of this project. Firstly, the societal context of this research is introduced. Secondly, the central problem of sustainability of medicinal plant harvesting and the value of novel scientific methods in answering key knowledge gaps is stated. Thirdly, the theoretical framework and research methodology are briefly outlined. Lastly, how results can answer the research aim is summarised.

1.1 MEDICINAL PLANTS IN WEST AFRICA Traditional medicine and customary ritual has a longstanding role in communities throughout Western Africa. Research concerning medicinal plants has focused on the integral role they have in primary healthcare, as well as the increasing dependence on medicinal plant products as a trade commodity (Cunningham, 1993; van Andel et al., 2012; Schippmann, 2002). The medicinal plant market is a sector of considerable and increasing economic value, providing additional income to many persons who may be involved in plant collection, preparation, transportation, and the sale of plant products at markets (van Andel et al., 2012). Recent research in Ghana and Benin has reiterated concern that wild-harvested species may be threatened by unsustainable harvesting practices, and that overexploitation is of particular concern in the vicinity of populated areas and growing markets (van Andel et al., 2012). The potential impacts of commercialisation have been highlighted in Benin where agricultural expansion has disturbed and depleted natural landscapes and commercialisation has skewed profitability along the modern supply, increasing demand for and overexploitation of some plant species (Vodouhe, 2008).

1.2 SUSTAINABLE HARVESTING Diverse strategies for sustainable use of medicinal plants may be required to ensure the longevity of the medicinal plant market in Benin amidst increasing commercialisation and land use change. By means of example, in-situ conservation may be recommended for species and habitat in human-modified landscapes where harvesting requires monitoring. Conversely, ex-situ conservation through, for example, cultivation may be possible to relieve pressure on natural habitat and facilitate pro-poor

schemes for small-scale harvesters or entrepreneurs. Because this issue straddles conservation of biological diversity, access to healthcare and wellbeing, and the fostering of sustainable livelihoods it is an important field for extended and interdisciplinary research. Recent quantitative market surveys have identified suspected vulnerable and overharvested species (Table 1.1). This list has been amended with results of research conducted in Benin, owing to the extensive fieldwork of Tinde van Andel, Alexandra Towns and Diana Quiroz. These species were ascertained through evaluating market price, plant part harvested, plant provenance (primary forests, woody savannahs), prevalence of plant material at markets and interviews with market vendors on scarcity (van Andel, personal communication). Knowledge of localities of extraction, species abundance and impacts of harvesting practices is crucial to assessing population viability. This information is not readily shared by market vendors, but vital to identifying the potential impacts of extraction of medicinal flora. The plants on this list are used for medicinal purposes as well as ritual, hereafter collectively referred to as medicinal plants.

1.3 SPECIES DISTRIBUTION MODELING In the absence of extensive occurrence data, Species Sistribution Modeling (SDM) is a promising tool in the field of conservation planning and resource management to predict the ecological niche of a species (Franklin, 2009; Guisan et al., 2006; Miller et al., 2004; Phillips et al., 2006). SDM comprises a host of statistical and machine-learning methods to interpolate biological and environmental data with the view to extrapolate found relationships in space and time (Elith et al., 2009; Franklin, 2009). SDM has been researched extensively, the nature of previous studies ranging from the review of its theoretical basis and practical applicability, to comparative studies of modeling techniques, as well as model accuracy and evaluation (Franklin, 2009; Elith et al., 2009). MaxEnt has been used successfully to analyse species distribution and map geographic distributions for conservation planning (Gaikwad et al., 2011: Miller et al., 2006; Sanchez et al., 2010). Whilst SDM has been employed in a variety of fields, this project contributes a new facet of research in the context of sustainable harvesting of medicinal plants. Using existing herbaria records and environmental data a niche model can be trained for West

3 | Introduction

Africa, later focusing on species distributions in Benin. A GIS-based analysis can employ land satellite data (2006) to account for land use change and identify suitable habitat within predictive maps. Moreover, a GIS-based spatial analysis can estimate relative exposure of suitable habitat to medicinal markets.

1.4 EXPECTED OUTCOMES

1. For which predictive maps generated in MaxEnt is statistically significant predictive performance indicated?

This research anticipates that SDM and GIS-based spatial analyses can provide new information of medicinal plants geographic distributions in West Africa. Model accuracy will be evaluated with a null-model analysis as advocated by Raes & Steege (2007). Predictive maps will be used to identify predicted presences and suitable habitat in Benin and its surrounds. Results will be interpreted by comparing and contrasting results with the IUCN Red List and results of market surveys recently conducted in Ghana (van Andel et al., 2012). A basic GIS model will be developed that can identify suitable habitat in the proximity of key medicinal plant markets. It is this expected these three objectives can fulfill the research aim and evaluate if SDM and GIS-based

2. Is species vulnerability indicated by predictive maps in Benin regarding geographic suitable habitat ?

characterisation of suitable habitat can be used to predict vulnerability of commercially extracted wild-harvested medicinal plants.

The central aim of this research is to evaluate whether SDM and GIS-based characterisation of habitat suitability can be used to predict species vulnerability for commercially extracted medicinal species. The research questions are as follows:

3. Can a GIS-based spatial analysis be used to predict species vulnerability regarding proximity of suitable habitat to key markets?

Table 1.1 Candidate species list for species distribution modeling. Ammended list of plant products encountered during market surveys in Accra, Kumasi, Cape Coast, Tamale and Akoase (Ghana), July—August 2010. Source: van Andel, 2012; Genus and species

Habitus

Plant Part Sold

Use

Abrus precatorius L. Acridocarpus smeathmannii Bridelia ferruginea Benth. Clausena anisata (Willd.) Hook.f. ex Benth. Daniellia ogea (Harms) Holland Daniellia oliveri (Rolfe) Hutch. & Dalziel Dichrostachys cinerea (L.) Wight & Arn. Ehretia cymosa

liana liana tree tree tree tree shrub shrub

leaves X bark, root root, leaves bark, resin bark wood, root X

asthma X vaginal discharge, STDs abortion, infertility ritual incense, ritual headache X

Entada gigas L. Khaya senegalensis (Desv.) A.Juss.

liana tree

seeds bark, wood

ritual blood tonic, aphrodisiac, fever

Mondia whiteii (Hook.f.) Skeels Okoubaka aubrevillei Pellegr. & Normand Paullinia pinnata L Pericopsis elata (Harms) Meeuwen Piper guineense Schumach. & Thonn. Pteleopsis suberosa Engl. & Diels Pterocarpus erinaceus Sarcocephalus latifolius (Sm.) E.A.Bruce Secamone afzelii (Roem. & Schult.) K.Schum. Securidaca longipedunculata Fresen. Senna alata (L.) Roxb. Sphenocentrum jollyanum Pierre Strophanthus hispidus DC. Vitellaria paradoxa C.F.Gaertn. Zanthoxylum zanthoxyloides (Lam.) Zepern. & Timler 


liana tree liana tree liana tree Tree tree liana shrub shrub shrub liana tree tree

root bark, seed root, wood, leaves leaves, wood seeds, wood bark X root leaves root leaves root root seed, fat bark, root

aphrodisiac ritual, aphrodisiac menstrual pain ritual spice, asthma, convulsions clean uterus, STDs X infertility pregnancy nausea afrodisiac, phlegms puerperal fever aphrodisiac pregnancy fever, body pain, STDs skin boils strengthen after birth, enhance breast milk

Methods 2.1 THEORETICAL FRAMEWORK The central theory underlying SDM is the differentiation of a realised niche and fundamental niche (Figure 2.1). Whereas a fundamental niche encompasses the environmental requirements required for species survival, the realised niche may be a smaller range due to resource availability, biotic limitations, and artificial or natural barriers (Hutchinson, 1957, cited in Phillips et al., 2006). SDM supposes that species’ distribution data can be extrapolated in space and time based on the environmental data that relates a species occurrence to it’s habitat suitability (Franklin, 2009:3); constraints in environmental conditions governing species distribution can be inferred to make a prediction of geographic distribution, i.e. a prediction of it’s fundamental ecological niche within the time window of the data used. The potential limitations of SDM should be considered so results can be interpreted within the constraints with which they are made. Firstly, a key concern is that SDM extrapolates in both space based on data relating to conditions at a certain point in time. Occurrence data is often related to explanatory variables which result from an interpolation of data over a larger time period (e.g. interpolated climate surfaces). Additionally, a mismatch in spatial scale exists: whereas localised conditions may make a spot suitable for a given organism it is not possible to acquire the explanatory variables at that resolution (van Loon, personal communication). In both cases, the result is a static set of relationships that model a dynamic phenomenon— species dispersal and habitat selection. Secondly, a key consideration is the degree to which an occurrence can

be assumed to be in equilibrium (Guisan & Zimmermann, 2000). If a species is at the verge of it’s conditions required for survival the measured environmental data at that point does not constitute its ecological niche—in fact, it may be to the contrary. Thirdly, the driver of that this so-called state of equilibrium may be earlier biotic than abiotic factors. By means of example, a species that occurs in coastal areas or along river banks, such as Entada gigas, survives not in the least within certain environmental conditions despite their reproductive mechanisms that harness water for dispersal, but also because of them. Lastly, sampling bias is an important factor in to be critical about in interpreting SDM results. Occurrence locations are not sampled randomly in space, there may be specific mechanisms—be they human or scientist behaviours—why certain environments were sampled and others not. That being the case, a resulting data set may reflect sampling mechanisms on top of ecological relationships. The risk of having a biased sample decreases through including data from a larger time window and from different sources. Perhaps most pertinent, the decisions during the model process can shape ‘trade-offs’ incorporated in prediction outputs. One aspect of this is the model type that is used: a high degree of precision may neglect the generality needed to predict suitable habitat. It may not be clear, however, from the model technique and output in which domain its shortcomings lie (Guisan & Zimmermann, 2000). A second aspect, is the decisions concerning study area, species data, and particularly the environmental variables which are used as predictor variables. This will be elaborated upon in the following section.

2.2 SPECIES DISTRIBUTION MODELING In the domain of SDM there are numerous methods available for modeling. These vary according to the statistical technique or decision-making process that relates available data to extrapolate in space and time. In this research the MaxEnt method has been chosen because

Figure 1.1 Illustration of the relationship between fundamental and realised niche in geographical and environmental space. Source: http://biodiversityinformatics.amnh.org/

of it’s efficacy in generating probability prediction using presence-only data. It has been used for diverse ecological and conservation purposes and performed competitively against other modeling methods despite sometimes limited occurrence data (Elith et al., 2006; Philips et al., 2006). The maximum entropy principle applied in SDM states that for

5 | Methods

a study area wherein an unknown probability distribution determines the likelihood of a species being either absent or present, the best approximation is one that satisfies the known constraints present within the species-environment relationship whilst maximizing entropy within that probability distribution (Phillips et al., 2006). Importantly, MaxEnt relies only on environmental data at known occurrences to establish the constraints of the probability distribution.

2.2.1 DATA PREPARATION Study Area

The niche model was trained on a region covering West and Central Africa ranging from 15º North to -14º South and 34º West to -7º East (Figure 2.1). The model was projected for the same geographical extent allowing greatest accuracy, whilst the study area of interest, Benin, can be the focus of further analyses in the map outputs. It is assumed that the larger extent is broad enough to include the scope of environmental gradients for the species occurrences in tropical West Africa. Although for some species the study area can be reduced this was decided against to ensure the scope of environmental gradients is sufficient and the models can be run efficiently as a batch. It was noted that the study area includes variation in sampling intensity. Whilst it has been found that large extents of presence-only data may compromise performance where a spatial bias exists (Franklin 2009:71), the more pertinent aspect is the potential

bias that occurs within environmental gradients present in the study area and this is inadvertently tested through a nullmodel analysis.

Distribution Data

Candidate species were selected according to frequency of harvest and commercialization (van Andel, personal communication). The candidate species comprise trees, shrubs, lianas and herbs and do not include semi-cultivated plant species (Table 1.1). Weeds originally provided for ‘model testing’ were removed from the original candidate list. Occurrence data was collected from the Nationaal Herbarium Nederland (NHN) BRAHMS database and the Global Biodiversity Information Facility (GBIF) (http:// www.gbif.org/). The BRAHMS database incorporates data collected for the ECOSYN research project and importantly both plant data of Hawthorne & Jongkind (2006) and that of the Flore du Benin project. Synonyms present in the GBIF database were aggregated under the accepted names. Where possible, incomplete data records were georeferenced to the locality level using an online gazetteer program, BioGeoMancer (Guralnick et al., 2006).

Environmental Predictor Variables

Environmental predictor variables used as input in MaxEnt comprise bioclimatic and soil characteristic variables at 1km resolution (Table 2.1) with projected cordinate according of the World Geodetic System 1984. The bioclimatic variables are interpolated climate surfaces from

2.1b

2.1a

Legend

Study Area

Figure 2.1 a. Study area used for for niche model training ranging from 15º North to -14º South and 34º West to -7º b. Study area for predicting species vulnerability: Benin, Africa (Source: www.africa.upenn.edu, accessed from < http://mappery.com/maps/Benine-Map.gif>)

6 | Methods

Table 2.1 
 Environmental predictor variables. 
 Variable

Description

BioCLIM 01

Annual Mean Temperature

BioCLIM 03

Isothermality (Mean Diurnal Range/Temp Annual Range)*(100); Diurnal range=Max month T-Min Month T Minimum temperature of the coldest month

BioCLIM 06 BioCLIM 12 BioCLIM 19 Drainage Ref Dept T CaSO4

T CEC Soil

T ECE T ESP T Texture AWC Class

T Gravel

T Silt

T OC T PH20

Annual Precipitation Precipitation of coldest quarter Soil drainage classes are based on soil type, texture, soil phase and terrain slope. Reference depth of the soil unit. Topsoil calcium carbonat, common to dry places and indicates limited clay translocation. Can be couple with iron deficiency in high concentrations Topsoil cation exchange capacity indicating nutrient fixing capacity and correlated with clay content and organic matter compatible with plant growth. Topsoil electrical conductivity with regard to salt content of a soil. Plant growth can be inhibited by salt content of soils. Topsoil Sodicity indicates sodium percentage in the top soil, which is used to regenerate chemical compounds in photosynthesis. Topsoil Textural Class for the first 30cm Available water storage: The AWC classes have been estimated for all soil units of both FAO classifications accounting for topsoil textural class and depth/volume limiting soil phases. Volume percentage gravel in the top- and subsoil. Gravel stands for the percentage of materials in a soil that are larger than 2 mm. Percentage silt respectively in the in the top- and subsoi. Silt size is between 0.002 and 0.050 mm (USDA classification) and between 0.002 and 0.0625mm (ISO and FAO classification). Percentage organic carbon in the topsoil. pH measure in soil-water solution is a measure for the acidity and alkalinity of the soil.


 1950–2000, acquired from the WorldClim dataset (http:// www.worldclim.org; Hijmans et al., 2005). The Soil Terrain variables were collected from the Harmonized World Soil Database (HWSD). Notably, this raster stack was the result of a larger dataset already tested for multicollinearity. This is an important stage in the data preparation. The WorldClim dataset comprises 19 continuous variables, and the HWSD numerous soil characteristics. Correlations exist between variables and needs to be removed, in this case using a Pearson’s correlation (r