=Paper=
{{Paper
|id=Vol-2380/paper_72
|storemode=property
|title=Plant Recommendation using Environment and Biotic Associations
|pdfUrl=https://ceur-ws.org/Vol-2380/paper_72.pdf
|volume=Vol-2380
|authors=Sara Si-Moussi,Mickael Hedde,Wilfried Thuiller
|dblpUrl=https://dblp.org/rec/conf/clef/Si-MoussiHT19
}}
==Plant Recommendation using Environment and Biotic Associations==
https://ceur-ws.org/Vol-2380/paper_72.pdf
Plant recommendation using environment and
biotic associations
Sara Si-Moussi1,2,3[0000−0002−0519−8699] , Mickaël Hedde1[0000−0002−3510−0701] ,
and Wilfried Thuiller2[0000−0002−5388−5274]
1
Eco&Sols, Univ Montpellier, CIRAD, INRA, IRD, Montpellier SupAgro, F-34398
Montpellier, France
2
University Grenoble Alpes, CNRS, Univ. Savoie Mont Blanc, CNRS, LECA,
Laboratoire d’Ecologie Alpine, F38000 Grenoble, France
3
sara.si-moussi@inra.fr
Abstract. Automatically predicting species make-up in geographic lo-
cations is of great importance in the context of the current conversation
about biodiversity. Inspired by the ecological concepts of Grinnellian and
Eltonian niches, we investigate two neural network architectures that aim
to that aim to exploit the respective features of these two types of niches
in order to tackle the plant recommendation task. The first proposal
uses environmental rasters and leverages advanced feature extraction
techniques based on distributed representations and convolutional neu-
ral networks. The second proposal relies on neighboring co-occurrences
of plants and organisms from an expert-curated list of taxa. We find
that the former solution outperforms the latter in prediction accuracy,
yet the second solution provides interesting and more interpretable in-
dicators. Both approaches yield promising results on the GeoLifeCLEF
2019 challenge.
Keywords: Distributed representations · Convolutional neural networks·
Species distribution models · Ecological niche theory · Plant ecology.
1 Introduction
Predicting the most likely species in a given location is of great importance in
biodiversity studies. This age-old task in biogeography consists in learning a
density function of the species over the geographic space from a set of observed
geolocalized occurrences. In practice, due to sampling bias, limited examples and
local habitat heterogeneity, geographic coordinates are not used directly as pre-
dictors. Instead, species abundance is modeled as a function of the environmental
conditions at the given locations. Such models are called Species Distribution
Copyright c 2019 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
CLEF 2019, 9-12 September 2019, Lugano, Switzerland.
Models (SDMs) or Habitat Suitability Models (HSMs).
The local environment is usually described by abiotic features such as cli-
mate and pedology. Recently, more studies include biotic covariates in the form
of other living beings abundance, motivated by the need to account for the de-
pendencies between species that can affect their co-distributions. Indeed, two
species distributions may be correlated indirectly through a latent abiotic vari-
able or directly if they interact in some way that creates a dependency between
their respective populations as in the case of plant-pollinators, host-parasites,
predator-preys, etc.
The set of locations with suitable abiotic conditions for a given species de-
fine its Grinnellian niche[6]. On the other hand, the role it occupies within its
community through feeding on other organisms or interacting with them defines
its Eltonian niche[5] or its biotic requirements. The locations at the intersection
between the two niches constitute the species’ Fundamental niche. What we ob-
serve is an accessible subset of it called the Realized niche [14].
In the context of the GeoLifeCLEF 2019 challenge, part of the LifeCLEF
evaluation campaign[8], we evaluate two models on location-based plant rec-
ommendation. The first one relies purely on abiotic features while the second
harnesses regional level co-occurrences of the target species with other organisms
selected with expert assistance. We describe both solutions and discuss results
obtained during training, validation and test phases.
2 Dataset description
The task organizers provided a training dataset containing about 280K obser-
vations from the Global Biodiversity and Information Facility database. Nearly
2M plant occurrences from automatic species identification of pictures produced
in 2017–2018 by the smartphone application Pl@ntNet were added. We used the
complete provided plant occurrences involving almost 3.5K plant taxa. Besides,
10M occurrences from other biodiversity kingdoms were also included. Finally,
33 environmental rasters covering the French territory were provided. They de-
scribe the climate, topology and pedological landscape. They were constructed
from various open datasets as explained in the protocol note [2].
3 Task and proposals
The task consists in training a model that predicts at a geographic location the
dominant plant species. We formulate this task as a multi-class classification
problem where the output class is the dominant plant identity. We evaluate two
architectures for solving the task:
– GrinnellNet: A convolutional neural network architecture using environmen-
tal rasters. It aims to learn features of species Grinnellian niche.
– EltonNet: A species embedding network leveraging associations with nearest
non-plant taxa occurrences. Its purpose is to identify community composition
patterns that are positively associated with a specific plant species, possibly
related to the Eltonian niche concept.
Hereafter, we describe both proposals and motivate our architecture, preprocess-
ing and optimization choices.
3.1 GrinnellNet: a CNN with categorical rasters embedding
GrinnellNet’s architecture (illustrated in Figure 1) is organized in a stack of
components trained end-to-end: input preprocessing, feature extraction, feature
interaction and classification. It takes environmental rasters as inputs and re-
turns the identity of the dominant plant for each location.
Input preprocessing Observing a species in a given location does not nec-
essarily imply the suitability of the abiotic environment. Indeed, some species
survive in locations with unfavorable conditions (known as sink locations) as
long as new individuals continuously join the population from a nearby suitable
habitat (source locations), through seed dispersal for instance via wind currents.
These so called source-sink dynamics [7] ensure an indefinite sustainability of the
sink populations despite unfavorable abiotic conditions. Consequently, failing to
account for these spatial processes may result in overestimation of the species’
environmental niche’s breadth. In addition to the effects of such stochastic as-
sembly processes, the geographic coordinates provided by smartphone devices
come with measurement errors, which represent a further source of uncertainties
that need to be accounted for.
For these reasons, it is necessary to consider a landscapewise rather than a
pointwise description of the environment around the observation geolocation by
using environmental patches instead of local values.
We divide the environmental rasters into three groups based on their semantic,
the resolution at which they vary and their data type (quantitative, ordinal and
categorical).
– TopoHydroClimate group: quantitative variables describing global biocli-
mate (CHBIO, ETP), hydrology (water proximity) and topology (altitude).
– Pedology group: ordinal and categorical variables describing the physico-
chemical structure of the soil.
– Land use group: includes Corine Land Cover class.
Embedding categorical features
Part of the soil’s physico-chemical properties are described by categorical or
pseudo-ordinal features such as texture, land cover, erodibility and crusting class.
Therefore, for each of the fc categorical features, before feeding it to the CNN
(a) SepGrinnellNet: separate feature extraction modules
(b) JointGrinnellNet: joint feature extraction
Fig. 1. GrinnellNet architecture
layers, we replace each of its nc categories by a real vector representation of
size kc , where kc is a tunable hyperparameter, typically chosen in the interval
[[2, n2c ]]. In practice, this is implemented by a feature-specific embedding lookup
layer parameterized by a (nc , fc ) matrix Ec , such that Ec [i, ] is the kc sized
embedding of the ith value of fc . We apply this transformation batchwise in par-
allel to all patch cells (light grey module in Figure 1). For an input of dimension
(batchSize, patchRadius, patchRadius, 1), the embedding layer of fc returns a
(batchSize, patchRadius, patchRadius, kc ) tensor. These vector representations
are trained along with other network parameters.
Categorical embeddings capture richer relationships than raw categories.
They are also considered as a dimensionality reduction technique, more practical
than one-hot-encoding when dealing with high-cardinality yet sparse features (a
typical example in our case is land cover). Note that each embedding dimension
could have multiple meanings that do not necessarily line up with ordinal di-
mensions. In the end, categories with similar representations translate a similar
effect on the target variable.
Resulting preprocessed inputs
Pedological features embeddings resulting from the previous step are concate-
nated into
Pa single tensor of dimension: (batchSize, patchRadius, patchRadius, dc )
s.t dc = c∈C kc (brown module in Figure 1). C refers to the set of categorical
features. In our case, dc = 18. Land cover is embedded separately (khaki box in
Figure 1).
Climate, topological and hydrological features are input to the neural net-
work as batches of (batchSize, patchRadius, patchRadius, dnum ) multi-channel
images, dnum = 22 being the number of features.
Feature extraction We investigated two modes of feature extraction on the
preprocessed inputs:
– JointGrinnellNet: A joint mode where all input tensors are concatenated
into a single multi-channel image. The result
P with dimension
(batchSize, patchRadius, patchRadius, c∈C kc + nnum )) is fed into a fea-
ture extraction module, as presented in Figure 3.1.
– SepGrinnellNet: A separate mode where the three groups of features are
treated separately by dedicated feature extraction modules then merged af-
terwards, as shown in Figure 3.1.
Handling spatial data requires to use appropriate feature extraction tech-
niques that are able to harness the spatial structure of such inputs. Convolu-
tional neural networks[10], a class of artificial neural networks inspired from the
virtual cortex of animals, constitute an ideal choice as they allow to extract
features from spatially-structured inputs within an end-to-end learning process.
They have been previously shown to provide substantial improvements in pre-
dicting species abundance [1,4].
Each of our CNN-based feature extraction modules comprises a two-block
architecture similar to VGG[16]. Each block contains two 3×3 convolution layers
set to extract 256 features, followed by MaxPooling then a Leaky Rectified Linear
Unit activation. The latter function choice prevents the model from falling into
a dying ReLu problem (experienced during first tests with ReLu activation)[11].
Retained embedding sizes, patch radius and resulting number of channels are
shown in the detailed architecture are provided in the source code.
Feature interaction and classification components Extracted features
from the different components are flattened and eventually concatenated into
a single large vector. This vector is then fed into a fully connected neural net-
work dedicated to learning the separation of the plant classes in the learnt feature
space. This feature interaction component (green box in Figure 1) comprises 3
dense layers of respectively 8192, 4096 and 3353 neurons. We applied a 0.75
dropout rate on the intermediary layers to prevent overfitting. The classification
layer consists on a softmax activation applied on the output to determine the
probabilities of each class. Class probabilities sum to one by definition. Naturally,
the class with the highest probability is attributed to the instance.
3.2 EltonNet: a species embedding network
Here we propose to rely purely on associations between plants and other taxo-
nomic groups. The goal is to predict the dominant plant from knowledge of the
occurrences of other taxa around it, up to a certain radius. In order to reduce
the number of co-occurring organisms, address the rarity of some of them and
capture stronger associations with plants, the following processing steps were
applied to the records of non-plant occurrences.
Taxonomic grouping and biogeographical filtering We aggregated taxa
according to ecological knowledge on the taxonomic level where biogeographical
correlations to plants are meaningful. This level differs from one group to an-
other, thus different preprocessing schemes were applied to different taxonomic
groups. Then, for some groups we used domain-knowledge heuristics to filter
irrelevant groups. Proportions of retained groups in the taxa list are illustrated
in Figure 2. Finally, we assigned internal codes, i.e. unique identifiers, to each
group.
Fungi selection . We grouped fungal species by genus. Then, we used the Fun-
GUILD database[13] to select fungi from guilds (groups with similar diets and
functions in the ecosystem) that are dependent on plants for feeding. We kept the
following guilds: Pathotrophs (parasites of plants) and Symbiotrophs (involved in
positive associations with plants such as mycchorizea). We deleted Saprotrophs
(organic matter decomposers) as they do not depend on plants whatsoever. In
the end, we retained 195 out of 531 genuses.
Insects selection . We aggregated insects to the order level except for Coleoptera
and Orthoptera which were grouped in families as they exhibit significant in-
traorder variability in terms of habitat preferences and diet. We chose among
insect orders those with known co-evolution history with plants (such as Hy-
menoptera) and/or established potential for direct interaction with plants (such
as pollinators and herbivores)[15]. The intuition is that some insects have strong
affinities or preferences (such as specialist pollinators) towards specific plants
which leads to a greater chance of co-existence. This process led to the selection
of 464 families of Coleoptera and Orthoptera in addition to 9 other orders.
Aves selection . Most birds breed in their preferential habitat during the pe-
riod spanning from March to July. The rest of the year, during their migration
phase, they travel through other areas where they can occasionally be observed.
We considered these observations as spurious and removed them to avoid es-
tablishing false associations to plants. We then aggregated birds to the genus
level. Afterwards, we used www.oiseaux.net to identify and remove some intro-
duced/invasive genuses. We ended up with 240 bird taxa.
Amphibians, mammals and reptiles aggregation . Lacking expert knowledge on
these phyla, we simply grouped them to the genus level, yielding 21 amphibians,
93 mammals, and 33 reptiles.
Fig. 2. Non plants taxa proportion in the retained list.
Biotic context calculation To accelerate the training phase, we precomputed
for each training example i, given its coordinates, the set Vi of non plant ob-
servations that occur within a radius of at most 8 Km. Starting with 500m, we
iteratively doubled the radius until we identified a non-empty set of neighbor
species, up until at most 8km. Afterwards, we randomly draw with repetition
w observations from Vi with a uniform probability. That way, more abundant
taxa (present multiple times in Vi ) have a higher probability of being included.
At the end of this process, we had associated each training example to its biotic
context made of w observations of organisms from other kingdoms.
The species embedding network architecture The learning model used is
an adaptation of the Continuous Bag of Words model first introduced in [12].
The architecture, illustrated in Figure 3, is based on a neural network composed
of 3 layers:
- The input layer of size W receives the identifiers of the biotic context compo-
nents.
- An embedding layer that associates a real-valued vector of size knp for each
taxa (non plants). This embedding vector captures the effect of observing this
organism on the odds of each plant class.
- A lambda averaging layer that aggregates the biotic context embeddings.
- A dense layer that computes for each target plant species the dot products of
its weight vector to the aggregated context embedding. This layer uses a soft-
max activation to return the probabilities of each target plant to occur given the
observations of the surrounding non plants.
4 Training and evaluation
4.1 Optimization and evaluation metrics
Given:
- P : the set of plant classes (here species-level identifications).
- c: the expected or true class.
- sc : the neural network output probability for the true class.
- wc : weight of the true class.
In both proposals, we optimize the class-weighted sparse categorical cross-
entropy loss given by equation Eq 1
expsc
CE = −wc log( PP ) (1)
sj
j exp
Some species were observed more often than others leading to a class-imbalance
problem within the training set. To address this issue, we weighted each training
example by the weight of its expected class (see Eq 2). This strategy allowed
us to give more importance to the misclassification of rare classes observations
Fig. 3. EltonNet architecture. Each box represents a layer described by its name and
type as well as its input and output dimensions. None refers to the undefined batch
size. The context size and embedding dimension shown here are respectively w = 50
and knp = 100. The last layer (bottom level) gives the probabilities of each plant class,
here 3353 classes.
(correcting for false negatives). Each class c with a frequency of occurrence pc
on the training set was attributed a weight computed as the ratio of its points
of absence to its points of presence.
1 − pc
wc = (2)
pc
This process is particularly useful for endemic species of undersampled locations.
4.2 Implementation and learning setting
We implemented GrinnellNet and EltonNet in Python4 using Keras deep learn-
ing framework with Tensorflow5 backend. We trained the models using multigpu
data parallelism on a single computing node equiped with 4 GPUS V100 with
NVlink6 . We used adam optimization algorithm[9] with a decaying learning rate
4
Source code: https://github.com/SoccoCMOS/GeoLifeCLEF2019-GrinnEltonNet
5
https://www.tensorflow.org/guide/keras
6
Ciment cluster, UMS GRICAD, Grenoble Alpes University
starting from 0.001 and reduced by a 10 factor whenever the validation loss stops
improving after 5 epochs.
4.3 Evaluation
We sampled 80% of the dataset for training and kept the remaining 20% for val-
idation. We used a stratified cross-validation split procedure to ensure coverage
of all classes in the training set. At the end of every epoch, we evaluated the
prediction accuracy of the models on the validation set. Figure 4 summarizes
the performances obtained during training (left axis) along with results on test
set (right axis). Note that different evaluation metrics are used. As a result, we
can only compare the models ranking.
ELT100 and ELT300 correspond to EltonNet applied to occurrences of test
species (evaluated in the test set) with the embedding size parameter knp set
respectively to 100 and 300. GRIN SEP and GRIN SEP+ (trained longer) apply
GrinnellNet on occurrences of all plant species whereas in GRIN SEP TEST the
model is trained only on test species. GRIN SEP uses separate feature extraction
components for each feature group while GRIN JOINT uses the joint feature ex-
traction mode.
Fig. 4. Performance of GrinnellNet and EltonNet variants on validation and test set.
Unsurprisingly, GrinnellNet performs much better than EltonNet. Indeed, we
would expect such results as covariates used in the former are richer and un-
biased. Besides, biogeographical theory recognizes the superiority of the abiotic
filter in selecting species[3], as it is directly related to their physiological traits.
On the other hand, EltonNet resulted from a series of arbitrary domain heuris-
tics. Nevertheless, it still performs better than random with relatively strong
associations learnt between plants and other taxa, a non-negligible insight for
community ecologists.
In the case of GrinnellNet, the choice of the feature extraction mode clearly
affects its predictive performances. Results show that treating the feature groups
separately leads to better performances. This can be explained by the nature
of the data encoded in the rasters that were created/interpolated from differ-
ent data collection protocols. Indeed, pedological characteristics for instance are
mainly determined by subjective field observations whereas climate data are
calculated with advanced mathematical models. Another possible reason to sep-
arate the feature extraction processes is the scale at which the rasters were
constructed. While bioclimatic variables are interpolated to the kilometer in
regular grids, soil data are aggregated using anthropo-topological polygons to
the landscape level, which translates in several kilometers. Consequently, we
only submitted GRIN SEP runs for the clef challenge.
We also observed that that the weighting strategy yielded significant per-
formance improvements over the unweighted variant (not shown here). Further-
more, the runs ranking on the validation set is roughly the same as in the test set
results except for GRIN SEP and GRIN SEP TEST. During validation, we found that
GrinnellNet performs better when it is trained solely on test species than when it
uses occurrences of all plant species. At test time, the order was reversed which
might be a sign of overfitting in GRIN SEP. But also, because GRIN SEP TEST is
exposed to more observations it probably learns more robust features.
5 Conclusion
We presented two proposals for the location-based species recommendation prob-
lem. The first solution leverages the concept of Grinnellian niche by building its
predictions on only abiotic features, automatically extracted from environmental
rasters using convolutional neural networks. This approach can be extended to
any taxonomic group beyond plants. Moreover, we investigated the use of dis-
tributed representations as a means to reduce feature dimensionality as well as
to capture rich semantic associations.
In the second proposal, we attempted to learn the Eltonian niche of the plants
by embedding the biotic contexts where they are observed. We relied heavily on
domain knowledge with expert assistance to filter co-occurrences in order to
learn strong associations. Although the assumptions and rules used to select
non plants species were specific to plant modeling, the learning architecture it-
self can be used for any taxonomic group. Additionally, this approach suffered
from the heterogeneity of the sampling effort. Ideally, one could use projection
maps predicted by species distribution models when available as input to a con-
volutional neural network.
Overall, our proposed CNN solution outperformed the species embedding
approach. But the latter allowed us to identify associations between plants and
other taxa which can be used to develop bioindicators. In the end, one could
train both models jointly with shared layers that can capture the interactions
and possible feedbacks between biotic and abiotic variables.
Acknowledgments
The authors would like to thank Tanguy Daufresne for considerable guidance
in selecting relevant vertebrate taxa and Esther Galbrun for comments on draft
versions of this document. SS is supported by a joint PhD fellowship of the
French National Institute of Agricultural Research (INRA) and the French Re-
search Institute for digital sciences (INRIA). The research was also supported by
funding from the French Agence Nationale de la Recherche (ANR) through the
GlobNets grant (ANR-16-CE02-0009). All the computations presented in this
paper were performed using the GRICAD infrastructure 7 .
References
1. Botella, C., Joly, A., Bonnet, P., Monestiez, P., Munoz, F.: A deep learning ap-
proach to species distribution modelling. In: Multimedia Tools and Applications
for Environmental & Biodiversity Informatics, pp. 169–199. Springer (2018)
2. Botella, C., Servajean, M., Bonnet, P., Joly, A.: Overview of geolifeclef 2019: plant
species prediction using environment and animal occurrences. In: CLEF working
notes 2019 (2019)
3. Boulangeat, I., Gravel, D., Thuiller, W.: Accounting for dispersal and biotic in-
teractions to disentangle the drivers of species distributions and their abundances.
Ecology letters 15(6), 584–593 (2012)
4. Deneu, B., Servajean, M., Botella, C., Joly, A.: Location-based species recommen-
dation using co-occurrences and environment-geolifeclef 2018 challenge. In: CLEF
2018 (2018)
5. Elton, C.S.: Animal ecology. University of Chicago Press (2001)
6. Grinnell, J.: The niche-relationships of the california thrasher. Auk 34(4), 427–433
(1917)
7. Holt, R.D.: Population dynamics in two-patch environments: some anomalous con-
sequences of an optimal habitat distribution. Theoretical population biology 28(2),
181–208 (1985)
7
https://gricad.univ-grenoble-alpes.fr
8. Joly, A., Goau, H., Botella, C., Kahl, S., Servajean, M., Glotin, H., Bonnet, P.,
Vellinga, W.P., Planqu, R., Stter, F.R., Mller, H.: Overview of lifeclef 2019: Iden-
tification of amazonian plants, south & north american birds, and niche prediction
9. Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. arXiv preprint
arXiv:1412.6980 (2014)
10. LeCun, Y., Bengio, Y., et al.: Convolutional networks for images, speech, and time
series. The handbook of brain theory and neural networks 3361(10), 1995 (1995)
11. Maas, A.L., Hannun, A.Y., Ng, A.Y.: Rectifier nonlinearities improve neural net-
work acoustic models. In: Proc. icml. vol. 30, p. 3 (2013)
12. Mikolov, T., Chen, K., Corrado, G., Dean, J.: Efficient estimation of word repre-
sentations in vector space. arXiv preprint arXiv:1301.3781 (2013)
13. Nguyen, N.H., Song, Z., Bates, S.T., Branco, S., Tedersoo, L., Menke, J., Schilling,
J.S., Kennedy, P.G.: Funguild: an open annotation tool for parsing fungal commu-
nity datasets by ecological guild. Fungal Ecology 20, 241–248 (2016)
14. Peterson, A.T., Soberón, J., Pearson, R.G., Anderson, R.P., Martinez-Meyer,
E., Nakamura, M., Araújo, M.B.: Ecological niches and geographic distributions
(MPB-49), vol. 56. Princeton University Press (2011)
15. Sauvion, N., Calatayud, P.A., Thiéry, D., Marion-Poll, F.: Interactions insectes-
plantes. Editions Quae (2013)
16. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale
image recognition. arXiv preprint arXiv:1409.1556 (2014)