identifying optimal sets of ecosystem indicators: a ... · icesc m2 015/b:19%...

2
ICES CM 2015/ B:19 Identifying optimal sets of ecosystem indicators: A comparative study of data analysis methods and regional results Danielle Dempsey (1), Wendy C. Gentleman (1), Mariano KoenAlonso (2), Pierre Pepin (2), Michael Fogarty (3) (1) Department of Engineering Mathematics, Dalhousie University, Halifax, Canada; (2) Fisheries and Oceans Canada, Northwest Atlantic Fisheries Centre, St. Johns, Canada; (3) NOAA Fisheries, Northeast Fisheries Science Center, Woods Hole, USA. Presenter contact details: [email protected], Phone 1 902 494 1905 Summary We are investigating quantitative approaches for identifying optimal sets of ecosystem indicators and comparing our findings for different regions. “Optimal” is defined as indicator sets that best predict stakeholderdefined ecosystem state with the least data requirements. Here we present our analysis using data from 1985 – 2013 for the Grand Banks off Eastern Canada. Time series of dozens of indicators were calculated and categorized as ecosystem drivers, pressures or states. Correlations within each category were used to identify and justify the removal of redundant indicators. The remaining indicators were then combined to predict ecosystem state indicators using multivariate multiple regression, and optimal predictor sets were identified from the results. We discuss these findings and outline our future plans to explore neural network analysis and compare results for Georges Bank, which straddles the US and Canadian borders in the Northwest Atlantic. Introduction Implementation of Ecosystem Based Fisheries Management (EBFM) requires information on ecosystem status and trends, which can be provided by databased indicators (Fogarty, 2014). Dozens of indicators have been developed to characterize environmental conditions, the ecological community, and fishing pressure, but no single metric captures complex ecosystem dynamics. Hence, indicator sets are evaluated to provide information to scientists and decisionsmakers (Methratta and Link 2006). Recommendations on how to select indicators that can address EBFM objectives are vague and largely qualitative. We are engaged in a project designed to aid EBFM by testing quantitative approaches that combine indicators to predict stakeholderdefined multivariate measures of ecosystem status while accounting for interconnections among biological, environmental and human factors. These methods facilitate determination of optimal indicator sets, defined as those with the highest predictive power but least data requirements. We are comparing application of different methods to different datarich regions with the objectives of better understanding individual ecosystems, examining the generality of our findings, and guiding choices for datapoor areas. Here we present a case study from the first phase of work, which applies multivariate multiple regression (MMR) as a tool for indicator selection for the Grand Banks of Newfoundland Canada. This analysis was structured using the DriverPressureStateImpactResponse (DSPSIR) framework (Pirrone et al., 2005), focusing on human and environmental forcings and on fish functional groups as metrics of ecosystem state. Materials and Methods We collected an array of biotic and abiotic data for the Grand Banks from a variety of sources, including Canadian and US government fisheries and climate organizations. Indicators were calculated from 1985 – 2013 and classified as a driver, pressure or state, following DPSIR. Indicator trajectories were compared between and within these categories, and Spearman correlations were used to identify redundant

Upload: others

Post on 04-Sep-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Identifying optimal sets of ecosystem indicators: A ... · ICESC M2 015/B:19% indicators!and!eliminatethem.!The!remaining!driver!andpressure!indicators!were!analyzedusing!MMR,! which!is!an!extension!of!simple!linear

ICES  CM  2015/  B:19  

Identifying  optimal  sets  of  ecosystem  indicators:    A  comparative  study  of  data  analysis  methods  and  regional  results  

Danielle  Dempsey  (1),  Wendy  C.  Gentleman  (1),  Mariano  Koen-­‐‑Alonso  (2),  Pierre  Pepin  (2),  Michael  Fogarty  (3)  (1)  Department  of  Engineering  Mathematics,  Dalhousie  University,  Halifax,  Canada;  (2)  Fisheries  and  Oceans  Canada,  Northwest  Atlantic  Fisheries  Centre,  St.  Johns,  Canada;  (3)  NOAA  Fisheries,  Northeast  Fisheries  Science  Center,  Woods  Hole,  USA.  Presenter  contact  details:  [email protected],  Phone  1  902  494  1905  

Summary  We   are   investigating   quantitative   approaches   for   identifying   optimal   sets   of   ecosystem   indicators   and  comparing   our   findings   for   different   regions.   “Optimal”   is   defined   as   indicator   sets   that   best   predict  stakeholder-­‐‑defined  ecosystem  state  with  the  least  data  requirements.  Here  we  present  our  analysis  using  data  from  1985  –  2013  for  the  Grand  Banks  off  Eastern  Canada.  Time  series  of  dozens  of  indicators  were  calculated  and  categorized  as   ecosystem  drivers,  pressures  or   states.  Correlations  within  each   category  were   used   to   identify   and   justify   the   removal   of   redundant   indicators.   The   remaining   indicators  were  then  combined  to  predict  ecosystem  state  indicators  using  multivariate  multiple  regression,  and  optimal  predictor  sets  were  identified  from  the  results.  We  discuss  these  findings  and  outline  our  future  plans  to  explore   neural   network   analysis   and   compare   results   for   Georges   Bank,   which   straddles   the   US   and  Canadian  borders  in  the  Northwest  Atlantic.    

Introduction  Implementation  of  Ecosystem  Based  Fisheries  Management   (EBFM)   requires   information  on   ecosystem  status  and  trends,  which  can  be  provided  by  data-­‐‑based  indicators  (Fogarty,  2014).  Dozens  of  indicators  have   been   developed   to   characterize   environmental   conditions,   the   ecological   community,   and   fishing  pressure,  but  no  single  metric  captures  complex  ecosystem  dynamics.  Hence,  indicator  sets  are  evaluated  to  provide  information  to  scientists  and  decisions-­‐‑makers  (Methratta  and  Link  2006).  Recommendations  on  how  to  select  indicators  that  can  address  EBFM  objectives  are  vague  and  largely  qualitative.  

We   are   engaged   in   a   project   designed   to   aid   EBFM   by   testing   quantitative   approaches   that   combine  indicators  to  predict  stakeholder-­‐‑defined  multivariate  measures  of  ecosystem  status  while  accounting  for  interconnections   among   biological,   environmental   and   human   factors.   These   methods   facilitate  determination  of  optimal  indicator  sets,  defined  as  those  with  the  highest  predictive  power  but  least  data  requirements.  We  are  comparing  application  of  different  methods  to  different  data-­‐‑rich  regions  with  the  objectives  of  better  understanding  individual  ecosystems,  examining  the  generality  of  our  findings,  and  guiding  choices   for  data-­‐‑poor  areas.  Here  we  present  a   case   study   from   the   first  phase  of  work,  which  applies  multivariate  multiple  regression   (MMR)  as  a   tool   for   indicator  selection   for   the  Grand  Banks  of  Newfoundland   Canada.   This   analysis   was   structured   using   the   Driver-­‐‑Pressure-­‐‑State-­‐‑Impact-­‐‑Response  (DSPSIR)   framework   (Pirrone   et   al.,   2005),   focusing   on  human   and   environmental   forcings   and  on   fish  functional  groups  as  metrics  of  ecosystem  state.      

Materials  and  Methods    We  collected  an  array  of  biotic  and  abiotic  data  for  the  Grand  Banks  from  a  variety  of  sources,  including  Canadian  and  US  government  fisheries  and  climate  organizations.  Indicators  were  calculated  from  1985  –  2013  and  classified  as  a  driver,  pressure  or  state,  following  DPSIR.  Indicator  trajectories  were  compared  between   and   within   these   categories,   and   Spearman   correlations   were   used   to   identify   redundant  

Page 2: Identifying optimal sets of ecosystem indicators: A ... · ICESC M2 015/B:19% indicators!and!eliminatethem.!The!remaining!driver!andpressure!indicators!were!analyzedusing!MMR,! which!is!an!extension!of!simple!linear

ICES  CM  2015/  B:19  

indicators  and  eliminate  them.  The  remaining  driver  and  pressure  indicators  were  analyzed  using  MMR,  which  is  an  extension  of  simple   linear  regression  that  uses  two  or  more  explanatory  variables  to  model  multiple   dependent   variables,   and   accounts   for   covariance   among   the   dependent   variables   when  assessing   explanatory   power.   All   possible   combinations   of   the   explanatory   indicators   were   processed  using  MMR,  and  their  adjusted  R2  values  were  compared  to  determine  the  optimal  indicator  sets.  

Results  and  Discussion  The  driver  and  pressure  indicators  exhibited  diverse  trajectories,  from  nearly  flat  to  pronounced  decadal-­‐‑scale   trends   and   dramatic   inter-­‐‑annual   fluctuations.   The   North   Atlantic   Oscillation   (NAO)   climate  indicator   was   uncorrelated   to   climate   drivers   or   environmental   pressures.   High   correlations   for   other  pressures   suggested   redundancy,   including   among   temperatures   at   different   depths,   between   salinity  and  stratification,  and  among  various  landings.  Based  on  these  results,  we  justified  a  subset  of  indicators  to  use  as  predictors  for  the  MMR  analysis.  Ecosystem  state  indicators,  represented  in  our  case  study  by  three  key  functional  groups,  also  exhibited  contrasting  decadal-­‐‑scale  trends  (Fig  1).  Correlations  of  these  ecosystem  state  indicators  with  drivers  and  pressures  reveal  large  differences  among  indicators  (Table  1),  which  illustrates  the  importance  of  considering  multiple  metrics  of  ecosystem  state.    

Preliminary  results  of  the  MMR  analysis  reveal  that  the   human   drivers   and   pressures   have   more  explanatory   power   than   the   environmental   drivers  and  pressures,  and   that  not  all  of   the   indicators  are  required   to   optimize   explanatory   power.   Future   work   will   contrast   MMR   with   neural   network   (NN)  analysis,   a   type   of   machine   learning   that   implicitly   models   non-­‐‑linear   relationships.   Our   preliminary  work   with   NN   illustrates   that   it   has   less   pre-­‐‑processing   requirements   and   better   or   equivalent  explanatory  power  than  MMR.  We  will  apply  both  methods  to  Georges  Bank  to  investigate  the  extent  to  which  the  optimal  indicator  sets  established  for  Grand  Banks  can  explain  the  ecosystem  stage  of  Georges  Bank   and   vice   versa.   We   predict   that   human   drivers   and   pressures   will   be   more   influential   than  environmental  drivers  and  pressures  in  both  of  these  economically  and  ecologically  important  regions.  

References  Fogarty,   M.   2014.   The   art   of   ecosystem-­‐‑based   fishery   management.   Canadian   Journal   of   Fisheries   and   Aquatic  

Sciences,  71:  479–490.  Methratta,   E.,   &   Link,   J.   2006.   Evaluation   of   quantitative   indicators   for   marine   fish   communities.   Ecological

Indicators,  6:  575–588.    Pirrone,  N.,  Trombino,  G.,  Cinnirella,  S.,  Algieri,  A.,  Bendoricchio,  G.,  &  Palmeri,  L.  2005.  The  Driver-­‐‑Pressure-­‐‑State-­‐‑

Impact-­‐‑Response   (DPSIR)   approach   for   integrated   catchment-­‐‑coastal   zone   management:   Preliminary  application  to  the  Po  catchment-­‐‑Adriatic  Sea  coastal  zone  system.  Regional  Environmental  Change,  5:111–137.