Download Cloud-Based Exploration of Complex Ecosystems for Science

Cloud-Based Exploration of Complex Ecosystems for Science,
Education and Entertainment
Ilmi Yoon1,2, Sangyuk Yoon2, Gary Ng1, Hunvil Rodrigues1, Sonal Mahajan1, and Neo D. Martinez2
1
Computer Science Department, San Francisco State University, San Francisco, California, USA
2
Pacific Ecoinformatics and Computational Ecology Lab, 1604 McGee Avenue, Berkeley, California, USA
Abstract
Recent computational advancement has opened up a new
field of computational analyses of complex ecological
networks where the nonlinear dynamics of many interacting species can be more realistically modeled and understood. Cloud computing supports ecological network research to extend to more complex and realistic modeling of
ecosystems. Considering that simpler models underpinning
much natural resource extraction policy leads to less biomass, and biodiversity than predicted by those simple
models, more realistic models are desirable for sustaining
extraction of biomass (e.g., fish, forests, fiber) from ecosystems that account for ecological complexity and computing for maximizing economic profit, employment and
carbon sequestration by ecosystems. These realistic ecology computation models in Clouds support not only the scientific computation, but also support Multiplayer online
game, “World of Balance” for players to entertain the experience of nurturing, managing and sustaining ecosystems. The game intends to educate players about the significant factors about ecosystems – non linear impact of one
species over others within ecosystem. Clouds computing
makes it feasible to supports the multifaceted usages of
ecological network computational models for science, education and entertainment seamlessly; it will effectively
circulate the meaningful data created from game play
(simulation) to scientific research, and from research data
to more realistic ecosystem managing experience to game
players.
Ecology Research
The destruction of biodiversity and ecological productivity
continues to degrade ecosystems' abilities to sustain human
and non-human life (Millennium Ecosystem Assessment
2005). Ecosystems that provide such services are complex
systems comprised by networks of diverse interdependent
interacting species. These networks or "food webs" depict
interconnected food chains within habitats such as lakes or
forests (Dunne 2009). Ecologists need to better understand
ecosystems in order to help manage threats to them. Only
recently has this understanding progressed to the point that
realistically complex ecosystems can be computationally
modeled. This advance emerged from computational studies of ecosystems that uncovered how such nonlinear high
dimensional systems may dynamically persist despite their
mathematical improbability. This improbability, described
four decades ago, directly contradicted the dominant paradigm at the time that held that diversity and complexity
stabilize ecosystems. The instability was based on some of
the earliest computational studies that found that large random networks did not persist because increasing their
complexity by increasing the number of nodes and links
decreases the probability that the network would return to
equilibrium following a small disturbance (Gardner &
Ashby 1970). This led to a famously coined challenge for
supporters of the paradigm to "elucidate the devious strategies that make for stability in enduring natural systems"
(May 1973).
Over thirty years later, computational research addressing this challenge found that regularities in food web structure (Martinez et al. 2006), predator-prey body-size ratios
(Otto et al. 2007), and feeding behavior (Williams & Martinez 2000) put ecosystems in a highly non random parameter space where diversity begets stability (Brose et al.
2006). Whereas before, realistically complex models failed
to persist, the new insights enabled such systems modeled
as nonlinear, high dimensional, coupled ordinary differential equations to characterize the bioenergetic feeding and
biomass dynamics of complex networks of persistently
interacting species (Berlow et al. 2009, Gross et al. 2009).
Such advances led to a resurgence of basic research on
ecological stability and initiated new and highly active
computational research focused on the ecological effects of
species loss and pollution (Brose et al. 2005, Berlow et al.
2009).
The seriousness of the many stressors on ecosystems,
which provide services critical for human life on Earth, is
recognized by an increasingly large number of people
across the world. Climate change and habitat degradation
due to human activity are chief among these challenges
(Millennium Ecosystem Assessment 2005). Some immediate and dramatic results of such perturbations are the loss
of species native to ecosystems (Hughes et al. 1997), the
invasion of ecosystems by species alien to them (Williamson 1996) and degradation of ocean productivity due to
overfishing (Worm et al. 2009). Interacting species within
ecosystems including humans form highly complex, nonlinear, dynamically coupled systems, but scientists are only
beginning to understand how this interdependence impacts
the fundamental structure, dynamics, function, and stability
of complex ecological systems (Carpenter et al. 2009).
Computational analyses of ecological systems that ignore humans have illuminated powerful structural regularities in the consumer-resource networks that comprise food
webs. Knowledge of this structure in turn helped illuminate network dynamics that describe how species' abundances and feeding rates vary over time and the dynamic
consequences of species loss, invasion, and environmental
change for ecosystems. For example, computational studies of species loss can now quantitatively predict the effects on the abundance of other species in field experiments (Berlow et al. 2009) and suggest which additional
species to eliminate to prevent extinction cascades resulting from the initial loss (Sahastrabudhe and Motter 2011).
The urgency of environmental problems and complexity
involved in solving them require new advances to computational approaches to these problems. More usable approaches are needed to enable ecologists and other noncomputational experts to conduct computational research.
More powerful approaches are needed to explore more and
larger networks of increased complexity that reflect more
of the variability, interactions, and environmental problems
found in nature.
Computational approaches are helping to continue such
research by modeling specific habitats (e.g., coral reefs,
lakes, forests, etc.), and we describe advances in these approaches that improve the power and usability of modeling, data management, and visualization (Fig. 1).
The Niche Model (William & Martinez 2000, 2008)
generates the initial structure of the food webs and has two
input parameters: the number of species S and connectance
𝐶 (Martinez 1992) where 𝐶 = 𝐿/𝑆 ! and L is the number of
trophic links. The model assigns a uniformly random
"niche value" (0 ≤ 𝑛! ≤ 1) to each of S species. Consumer
i eats only species whose niche values are contained within
a range 𝑟! with a center of 𝑐! < 𝑛! . 𝑐! is randomly chosen
from a uniform distribution between 𝑟! /2 and 𝑚𝑖𝑛(𝑛! , 1 −
𝑟! /2). 𝑟! = 𝑥𝑛! , where x is a random variable defined on
[0,1] with a beta-distributed probability density function
𝑝 𝑥 = 𝛽 1 − 𝑥 !!! with 𝛽 = 1/(2𝐶) − 1.
We use 17 network properties to describe food-web
structure (William & Martinez 2000, 2008, Dunne et al.
2002, 2008): Top, Int, Bas are the proportions of species
that are respectively without predators (top), with both
predators and preys (intermediate), and without preys (basal); Can, Herb, Omn and Loop are the fractions of species
that are cannibals, herbivore (only basal preys), omnivores
(i.e. feeding on multiple trophic levels) and involved in
loops (apart from cannibalism); ChLen, ChSD and ChNum,
the mean length, standard deviation of length and log number of the food chains; TL, the mean short-weighted trophic
level of species (Williams and Martinez 2004); MaxSim,
the mean of the maximum trophic similarity of each species; VulSD, GenSD and LinkSD are the normalized standard deviations of vulnerability (number of predators), generality (number of preys) and total links; Path is the mean
shortest food-chain length between two species and Clust
is the clustering coefficient (Watts & Strogatz 1998).
Fig. 1 – Network3D provides user friendly browser based interface and visualization and analysis tools for computa-­‐
tional ecology models The population dynamics within the food webs were
simulated using (Berlow et al. 2009): 𝐵! = 𝑟! (1 −
!∈!"#$#%$&!!
𝐵!
)𝐵 −
𝐾 !
𝐵! = −𝑥! 𝐵! +
𝑥! 𝑦!" 𝐵!
!∈!"#$%&'($
𝑥! 𝑦!" 𝐵! 𝐹!" −
!∈!"#$%!&"#
−
𝐹!"
(1)
𝑒!"
𝑥! 𝑦!" 𝐵!
!∈!"#$%&'($
𝐹!"
𝑒!"
𝑞! 𝐸!" 𝐵! (2)
!∈!"#$%
Eq. 1 and 2 describe the changes in the biomass densities
of, respectively, an autotroph and a heterotroph species. In
these equations, 𝑟! is intrinsic growth-rate of basal species
i, K is the carrying capacity shared by all the basal species,
𝑥! is i’s metabolic rate (𝑥!"#"$ = 0, Brose et al. 2006), 𝑦!" is
the maximum consumption rate of i eating j, 𝑒!" is i’s assimilation efficiency when consuming j. Current implementation of this population dynamics has 8 system parameters, 8 node parameters and 15 link parameters. As
number of node and links increases rapidly and each parameter has large parameter spaces, so endless variations
of simulation are possible and computationally it becomes
heavily complicated. With the price of the complexity,
extraction of biomass from realistically complex ecological
systems can result in profoundly different effects that extraction from more simplistically modeled systems. The
high dimensional, nonlinear, and nonrandom nature of these networks largely prohibit more analytical approaches
from shedding much light on their behavior. Cloud computing that provides as-needed access to computing with the
appearance of unlimited resources makes it feasible. However, while cloud computing has found widespread application in business environments, it has yet to demonstrate
many scientific accomplishments (Fox 2010). This makes
the harnessing of cloud-based approaches to scientific
problems a novel computer science challenge deserving of
innovative research efforts.
Gamification Approach
Multiplayer online game is a new multifaceted medium of
communication to reach the masses (players) effectively
that fosters healthy interaction and team cohesion. Recent
multiplayer online game, FoldIT enabled ordinary game
players to play for leisure while their intuitive and collaborative efforts has lead to unlocking of the structure of an
AIDS-related enzyme that scientific community had been
unable to unlock for a decade (Cooper et al., 2010, Firas
Khatib et al., 2011). These approaches make use of social
interaction and competition tendencies to engage massive
players to work together to achieve the intended objectives.
“World of Balance” is an entertaining educational multiplayer online game promoting the concept of ecosystem
nurturing inspired by “Farmville”, farm nurturing game
that are leisurely played by over 100 million players (Babcock, C., 2011). Inside World of Balance game, players are
given an empty land within the Serengeti eco environment
of Africa (first level eco environment), develop ecosystems
by adding plants, animal, and insect species one by one up
to 95 real representative species from the latest paper (S.
Visser et al., 2011, Dobson 2009). Players’ level increases
as they unlock new species while maintaining the balance
of the ecosystem. New player starts as an apprentice of one
of three types of mastery of special power (Weather, Animal Reproduction, or Plant Growth Rate) and increases the
influence over ecosystems. Inside World of Balance, players can build and nurture complex balanced ecosystems by
collaborating with each other; termed as “Player vs. Environment (PvE)” mode — or battle out to destroy and unbalance the ecosystems created by other players; termed as
“Player vs. Player (PvP)” mode. In both modes players
experience the process of ecosystem creation from producers (plants) to herbivores and carnivores, and the complex
prey-predator relationship that exists between them; along
with the inter-dependence essential to strike a balance in
the ecosystem (Fig. 2). The game also facilitates players to
share their progress in the game with their friends by posting it on their Facebook wall.
This game opens a mutually educational communication
channel between ecologists and masses (players). Players
benefit by learning important aspects of ecosystem development and food-web stability while producing prolific
and significant amount of scientific data useful for ecologists to analyze population dynamics model which is unfeasible for ecologists to produce on their own merits.
Game handles biological environment consisting of biotic
(living) components (organisms) and abiotic (non-living)
components (eg. air, soil, water, sunlight). Players are
managing both components within a game while playing
with features like creating a new food web, adding new
species to the food-web (invasion), adding more species of
the same type (proliferation), decreasing number of species
of the same type (exploit), removing species from the foodweb (removal), modifying parameters such as carrying
capacity, growth rate and metabolic rate; viewing user’s
manipulations and networks; updating biomass of nodes
(species) in the network etc.
Fig. 2 – screen capture from World of Balance game. Population dynamics of computational models deal with
effect of birth and death rates, immigration and emigration,
population decline etc. on the existing population. There
are 3 different types of parameters, system parameter, node
parameter and link parameter. And the current population
dynamics engine supports 8 system parameters, 8 node
parameters and 15 link parameters. In case of Serengeti
web, there are 95 nodes and 547 links, then there are 8 system parameters, 760 node parameters and 8205 parameters
that are all interdependent. For a human scientist or computational method alone, the parameter space is too huge to
be manipulated individually. Using the game interface, the
huge parameter space can be intuitively tuned and guided
by players. Human players can manipulate parameters
based on the knowledge and intuition such as carrying capacity, metabolism rate, and growth rate. Also distinctions
among plants, invertebrates, endotherms and ecotherms
vertebrates according to their body size to predict both
metabolic and maximum assimilation rates of organisms,
changing feeding rate of a predator in response to variations in prey density, predator divides its time searching for
prey or processing captured prey, attack rate as a function
of density of the prey, searching time in interacting with
members of their own species are parameters to be tuned to
simulate more realistic and balanced ecosystem. Simulation results (for example, reduction in biomass of species)
on each time step (equivalent to single game day) are feed
to game engine to map them into meaningful game activities (species are hunted down for the amount of reduction
if predator exist or die with other reasons) that user can
experience and interact further. World of Balance game
uses same game engine to support both collaboratively
nurturing world (inspired by Farmville) and competitively
battling world (inspired by Starcraft) by mainly altering the
game day scale. And the game engine uses the same cloud
computing infrastructure to predict the ecosystem for each
players, resulting fun and educational experience to players
and scientific simulation data carefully driven by human
players. This may be just the first of a series of ecosystem
management games that more realistically depict exploitation of forests, lakes, grasslands and oceans. Such games
may contribute to ecology and conservation biology much
of what Foldit has contributed to molecular biology. That
is the widespread inclusion of the complementary talents of
players and computers in solving important research questions (Khatib et al. 2011).
Clouds Computing as Solution
NIST has defined Cloud Computing as follows: "Cloud
Computing is a model for enabling ubiquitous, convenient,
on-demand network access to a shared pool of configurable
computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned
and released with minimal management effort or service
provider interaction.” This means Cloud computing creates
perfect environment to support computational ecology
models for science, education and entertainment approach
we have designed.
(1) On-demand resource: usage pattern of ecology research is not predictable as researchers from world wide
intensely use up memory and computation cycle during
active simulation period, but sparsely during other data
collection period. Also game is about to be deployed without having good estimation of near-future user community
size. Each player runs one or more population simulation
engines for each ecosystem player owns. cloud computing
can also store the times series from the simulations allowing analyses to focus on the data rather than rerunning
simulations to output subsets of simulation data thought to
be able to help answer research questions. On-demand re-
source reduces burden of purchasing expensive computational resource to support the whole infrastructure.
(2) Each simulation engine is independent to each other
and the large parameter sweeps to be conducted in short
times using vast computational resources in parallel makes
our analyses perfect application gaining the great potential
of cloud computing approaches.
(3) Minimal management effort or service provider interaction enables all the effort to focus on the problems to
solve, not the system maintenance.
(4) Researcher or game players from all over the world
will have convenient access to the resource.
These benefits could greatly accelerate scientific progress
on critically important problems.
Fig. 2-­‐ Diagram of cyberinfrastructure for ecological network analysis. A tool called Network3D is used to simulate the com-­‐
plex non-­‐linear systems that characterize these problems (i.e., Equations 1-­‐3). We ported Network3D to Azure. The Network3D engine uses Windows Workflow Foundation to implement long-­‐
running processes as workflows. Given requests which contain a number of manipulations, each manipulation is delivered to a worker instance to execute. The result of each manipulation is saved to SQL Azure. A web role provides the user with an inter-­‐
face where the user can initiate, monitor and manage their ma-­‐
nipulations as well as web services for other sites and visualiza-­‐
tion clients. Once the request is submitted through the web role interface, the manipulation workflow starts the task and a ma-­‐
nipulation is assigned to an available worker to process. The Network3D visualization client communicates through web ser-­‐
vices and visualizes the ecological network and population dy-­‐
namics results. Here we briefly discuss the process to port the existing
codes (Network3D Windows application) to Microsoft
Window Azure Cloud Computing environment. Population
dynamics engine of Network3D has been imported to
Windows Azure Cloud Computing as core processing unit.
User Interface of Network3D has been transformed to
Network3D Silverlight Web application as web role of
Windows Azure. Network3D Rest Web Services (web
role) serve as core management unit and manages Network3D tasks. User request which contains a number of
manipulations sent from Network3D Silverlight web application to Network3D web services is delivered to worker
instance (worker role) to execute. Manipulations are executed by Network3D Population Dynamics module and
results of each manipulation are being saved to SQL Azure
during the process. N3D web services access SQL Azure
to perform network or manipulation query or retrieve user
authentication. N3D population dynamics access SQL Azure to perform network or manipulation related operation.
Windows Azure Tables(Windows Azure Storage) which is
only used at Windows Azure is not used for compatibility
issue when N3D needs to be ported to regular windows
operating system. And the completed structure that supports Network3D browser interface and World of Balance
multiplayer social game is shown in figure 3.
Performance has been radically improved on Clouds.
Batch process for running population dynamics on nine
different food webs shown at table 1 on 3000 time steps on
windows application on PC vs. Clouds computing. The
table 1 shows the performance gain.
90000
PC
80000
.)c
se
(
e
m
ti n
io
ta
t
u
p
m
o
C
Windows A zure
70000
60000
50000
40000
Results such as those presented here have helped motivate
significant developments of cyber infrastructure to address
complex ecological networks. The cloud computing provides invaluable resources for infrastructure to support
Complex Ecosystems Computational Models for Science,
Education and Entertainment
References Babcock, C., "Lessons From FarmVille: How Zynga Uses The Cloud". InformationWeek (UMB): 29–34, 57. (2011). Berlow, E.L. et al. Simple prediction of interaction strengths in complex food webs. Proceedings of the National Academy of Sciences of the USA, 106: 187-­‐191. (2009) Brose, U. et al. Scaling up keystone effects from simple modules to complex ecological networks. Ecology Letters, 8: 1317-­‐1325. (2005) Brose, U. et al. Allometric scaling enhances stability in complex food webs. Ecology Letters, 9: 1228-­‐1236. (2006) Carpenter, S. et al. Science for managing ecosystem services: Beyond the millennium ecosystem assessment. Proceedings of the National Academy of Sciences USA, 106: 1305-­‐1312. (2009) Cooper et al, 2010 Predicting protein structures with a multi-­‐
player online game, Nature 466, 756–760 (2010). Dobson, A., Food-­‐web structure and ecosystem services: insights from the Serengeti, Phil. Trans. R. Soc. B (2009) 364, 1665–1682 Fox, A. 2011. Computer science. Cloud computing-­‐-­‐what’s in it for me as a scientist? Science 331:406-­‐7. Firas Khatib et al. Nature Structural & Molecular Biology 18, 1175–1177 (2011) Gross, T. et al. Generalized models reveal stabilizing factors in food webs. Science 325:747-­‐50. (2009) 30000
20000
10000
0
0
1
2
3
4
5
6
7
8
9
10
Index No.
Index No. 1 2 3 4 5 6 7 8 9 𝑪𝒐𝒏𝒄𝒍𝒖𝒔𝒊𝒐𝒏
Food web St.Martin Island Coachella Glass Bridge Brook Lake Flack Hawkins Seregenti Everglades Elverde Node No. 44 30 75 75 48 87 86 65 156 Link No. 218 290 113 553 702 126 547 652 1510 Table 1. – food webs in different complexity were used to com-­‐
pare the performance of Clouds computing vs. earlier Windows application on single PC. Gardner, M.R. & Ashby, W.R. Connectance of large dynamic (cy-­‐
bernetic) systems: critical values for stability. Nature 228, 784 (1970). Hughes, J.B. Et al. Population diversity: its extent and extinction. Science, 278: 689-­‐692. (1997) Millennium Ecosystem Assessment. Ecosystems and Human Well-­‐Being: Current State and Trends: Findings of the Condition and Trends Working Group. Island Press. (2005) Sahasrabudhe, S. & Motter, A.E. Rescuing ecosystems from ex-­‐
tinction cascades through compensatory perturbations. Nature communications, 2, 170. (2011) S. Visser et al, The Serengeti food web: empirical quantification and analysis of topological changes under increasing human impact, Sara N. de Visser*, Bernd P. Freymann and Han Olff, Journal of Animal Ecology 2011, 80, 484–494 Watts, D. J., and Strogatz, S. H. Collective dynamics of “small-­‐
world”networks. Nature 393, 440-­‐442. (1998) Williamson, M. Biological Invasions. Chapman & Hall. (1996) Williams, R. J., and N. D. Martinez. Simple rules yield complex food webs. Nature, 404: 180-­‐183. (2000) Williams, R.J., and N.D. Martinez. Stabilization of chaotic and non-­‐permanent food-­‐web dynamics. European Physical Journal, B 38: 297-­‐303. (2004) Worm, B., et al. Rebuilding Global Fisheries. Science, 325: 578. (2009)