Universitat de València 26 27th November, Universitat Politècnica de València 26 27 November 2012, Valencia Sapin 1
1. INTRODUCTION In recent years, fish species richness is used to assess the influence of human alterations on aquatic ecosystems (Clavero, et al., 2004; Hermoso, et al., 2011) Habitat alteration and invasive species are considered the main hazards for the conservation of the freshwater native fish species (Didham et al., 2007; García Berthou et al., 2005; Moyle, 1995). Some researchers question the importance of invasive species (Didham et al., 2005; MacDougall and Turkington, 2005) they suggest that invasive species are passengers and not drivers of the ecological change. The previous affirmation is supported by the fact that the existence of a positive correlation between invasive species and the decrease in native species cannot be interpreted as a dependence relationship (Gurevitch and Padilla, 2004). 2
2. OBJECTIVES To evaluate the relative importance of biological factors versus habitat in the reduction of native fish species richness (NFSR) in 3 Mediterranean Rivers in the Júcar River District (Eastern Spain). To assess if invasive fish are drivers or passengers in the decline of native fish 3
3. MATERIALS AND METHODS 3.1 Study area Main stem of the Júcar, Cabriel and Turia River Basins, in the Eastern Iberian Peninsula. We used 90 fish sampling sites in the 3 rivers, to build and test the models. Fish samples collected by electrofishing in spring and summer from 2005 to 2009. Study area showing the distribution of the 90 sampling sites in the three rivers. 4
We considered two groups of environmental variables: Biological variables Habitat variables Available BIOLOGICAL variables : Biological variables Method Code Invasive fish species richness In situ IFR Number of invasive fish which may affect the physical habitat In situ IFH Number of invasive fish predators In situ IFP number of families of benthic macroinvertebrates BMN BMF 5
Available PHYSICAL HABITAT variables : Habitat variables Method Code Dissolved oxygen (mg/l) MN DIS Habitat variables Method Code Biological Oxygen Demand (mg/l) MN BOD Hydromorphological units: Total phosphorus (mg/l) MN TOP Pools (%) In situ POO Nitrites (mg/l) MN NIT Glide (%) In situ GLI ph MN PH Riffle (%) In situ RIF Suspended solids (mg/l) MN SUS Rapid (%) In situ RAP Conductivity (µs/cm) MN CON Run (%) In situ RUN Water temperature (ºC) MN WAT Channel length without artificial barriers (km) GIS CWB Iberian Biomonitoring Working Party (IBMWP) index BMN Mean IBMWP width of water surface (m) In situ WID Mean Annual flow rate ( m 3 /s) MN FMA Inter annual Habitat mean variables flow (calculated for 5 years) ( m 3 /s) Method Code MN FIA Altitude (m a.s.l.) Coefficient of variation of mean monthly flows (fish sampling GIS year) ALT MN FIM Drainage area (kmcoefficient 2 ) of variation of mean annual flows (calculated GIS for 5 years) DRA MN FCV Distance from headwater source (km) GIS DHS Index of riparian habitat quality QBR BMN QBR 6
3.2 Modelling approach We built three ANN models (MultiLayer Perceptron) to analyze the effect of habitat alteration and invasive species on native fish richness: 1 2 3 Inputs: biological variables Inputs : habitat variables Inputs: biological and habitat variables Variables selection: forward step by step method We assumed that the best performance involve a better control on native species prediction than the models with lower performance. 7
Model performance: correlation coefficient (r) & mean square error (MSE). To assess predictive performance in validation: k fold cross validation method. Contribution of input variables estimated with Partial Derivatives method (PaD) (Dimopoulos et al., 1999). Partial derivatives method (PaD) Method consists in calculating the partial derivatives of the output according to the input variables Procedure for PaD method Two results can be obtained by this method Sensitivity Input relative importance Variations in Outputs for changes of each predictive variable Classification of the relative contributions of each variable to the output 8
Procedure for model selection Step 1 Database N= 90 Dataset preprocessing and input selection (discard collinearity) K=3 cross validation K=5 cross validation K=6 cross validation K=10 cross validation Step 2 Selection of the best ANN (with lowest MSE in validation) Partial Derivatives method (PaD) 9
4. RESULTS a a) Best ANN with biological variables: 5 4 1 nodes c b) Best ANN with habitat variables: 7 6 1 nodes b c) Best combined model biological & habitat variables: 7 6 1 nodes ANN for predicting fish richness NFSR 10
Model with habitat variables was the best, but the combined model presented similar performance. The worst model was that one using only biological variables as inputs. Correlation coefficient (r) and mean square error (MSE) values of the three models in validation Models Validation r MSE ANN with biological variables 0.67 1.06 ANN with habitat variables 0.81 0.62 ANN with both kind of variables 0.78 0.64 11
a N Invasive Predators 34..2% 27.3% Drainage Area Percent. Riffles b 20.1% 16.4% 20.7% Mean Ann. Flow Water Quality c Percent. Riffles Length of River Without Barriers 20.7% 26.7% Relative contribution of each input variable to native fish richness prediction. a) model with biological variables, b) model with habitat variables, c) model with both kind of variables. 12
Partial derivatives (PaD) indicate in general Positive influence of Percentage of riffle (RIF) River length without barriers (RWB) Partial derivatives in function of the most important environmental variables (combined model biological + habitat variables) 13
DISCUSSION CONCLUSION The 3 ANN and the PaD method identified habitat degradation as the main factor threatening native fish species richness in the Júcar, Cabriel and Turia rivers. Riffle proportion and IBMWP, most important variables for Richness; in best model (Input: habitat variables) their relative importance: 20.72% and 20.18% respectively. Mean Annual Flow, Flow variability, Length Without Barriers also very important. Corbacho & Sánchez (2001) indicated that HABITAT is a critical factor for natives decline in Guadiana River Basin (Iberian Peninsula). Godinho & Ferreira (1998); Hermoso et al. (2011); Light & Marchetti (2007) in California : INVASIVES as drivers of natives decline. Different variables, combinations of techniques make difficult comparison. Some techniques good for cause effect analyses (SEM) but require linear relations, etc. 14
CONCLUSIONS AND RECOMMENDATIONS ANN models provided ideas to improve the ecological status of native fish communities in altered rivers. Some potential ecological restoration actions are: 1) Removing dams and weirs that are not in operation in order to increase the river length without barriers and reduce the presence of meso lentic habitats (see Olaya Marín et al., 2012, Science of the Total Environment). 2) Design and Implementation of Environmental Flows imitating natural variability of the hydrological regime of Mediterranean rivers. 15
ACKNOWLEDGMENTS This study was partially funded by the Spanish Ministry of Economy and Competitiveness with the projects SCARCE (Consolider Ingenio 2010 CSD2009 00065) and POTECOL Evaluación del Potencial Ecológico de Ríos Regulados por Embalses y Desarrollo de Criterios para su mejora según la Directiva Marco del Agua (CGL2007 66412). We thank to Confederación Hidrográfica del Júcar (Spanish Ministry of Agriculture, Food and Environment) for the data provided to develop this study. 16
THANK YOU FOR YOUR ATTENTION Francisco Martínez Capel (fmcapel@dihma.upv.es) http://personales.gan.upv.es/fmcapel 17