Microsoft word - peÑasetal_fullpaper_ise2012_v2.doc

DO RIVER REACHES DIFFER IN HABITAT-FLOW RELATIONSHIPS
ACCORDING TO HYDROLOGIC CLASSIFICATION AND RIVER SIZE?
Environmental Hydraulics Institute ‘‘IH Cantabria’’, Universidad de Cantabria. C/ Isabel Torres nº 15. Santander, Cantabria, 39011, Spain Environmental Hydraulics Institute ‘‘IH Cantabria’’, Universidad de Cantabria. C/ Isabel Torres nº 15. Santander, Cantabria, 39011, Spain National Institute of Water and Atmospheric Research. 10 Kyle Street Riccarton ,Christchurch, PO Box 8602 , New Zealand National Institute of Water and Atmospheric Research. 10 Kyle Street Riccarton ,Christchurch, PO Box 8602 , New Zealand Environmental Hydraulics Institute ‘‘IH Cantabria’’, Universidad de Cantabria. C/ Isabel Torres nº 15. Santander, Cantabria, 39011, Spain Environmental Hydraulics Institute ‘‘IH Cantabria’’, Universidad de Cantabria. C/ Isabel Torres nº 15. PCTCAN Santander, Cantabria, 39011, Spain The overexploitation of water resources has led to a progressive impairment of fluvial ecosystems worldwide and the definition and establishment of environmental flow regimes has the objective of avoid further ecosystem deterioration. Nevertheless, when rapid assessments must be done over large areas, at national or regional scale, the definition of these environmental flows are usually based on simple rules that can be easily uniformly applied. These rules do not take into account the hydrological and environmental spatial variability and therefore, they may lead to obtain incoherent results and inconsistent consequences for the protection of the ecosystems. In this study we assess the consequences that minimum environmental flow estimated through a hydrological method have in the habitat quantity and quality, measured as weighted usable area (WUA), of adult Salmo trutta and Luciobarbus graellsii in rivers belonging to different hydrologic classes, rivers of different size and the combination of both characteristics. This analysis was carried out in Atlantic (Cantabrian sea) and Mediterranean (Ebro) catchments in the northern third of Spain. We developed and used a classification of natural flow regimes based on daily mean flow series measured in unmodified gauge stations. Random forest technique was used to model hydrological indices to all the river segments based on environmental predictor variables and then segments were classified according to these indices. Our analysis showed differences in the proportion of WUA that fish species retained between several hydrological classes which was greatly determined by the magnitude of rivers mean and high flows. However, differences were not as strong as expected. The combination of hydrologic classes with the river size allowed establishing clearer patterns between reduction of flow and habitat modification. Nevertheless, contrary trends were observed between the selected species which indicates that further research must be carried out to the yield more conclusive results. INTRODUCTION
The overexploitation of the water resources has led to a progressive impairment of fluvial ecosystems worldwide [1][2]. The importance of the flow regime to maintain the ecosystems functions and processes has been widely recognized [3][4] and thus, the definition of environmental flow regimens (EFR) is a central issue to limit the further deterioration of rivers [5]. There exist over 200 methodologies to calculate the extent to which the natural flow regime can be altered while maintaining a certain degree of ecological integrity [6]. In situations where water resource usage is intense and ecological values are high, EFRs are generally established through site specific studies and precise methods [7]. However, EFR definition for large jurisdictional regions is generally based on more simplistic rules that can be uniformly applied, as occurs with many hydrological methods [8]. This is the case of Spain and other states of the European Union, where relatively new water legislation proposed a generic methodology to establish a temporal distribution of environmental flows through hydrologic methods. In addition, in the case of Spain the results of the hydrologic method must be adjusted through the application of a habitat simulation model. However, the habitat simulation models have to be applied in just the 10% of the water bodies and there is any procedure neither to carry out this validation nor to analyze the consequences of the establishment of the proposed environmental flow in the other 90% of the water bodies. Given the importance of hydrologic variability in the function of river ecosystems, the application of the same hydrologic methods and simple rules to establish EFR in hydrological and environmental variable regions, as is the Iberian Peninsula, may lead to obtain incoherent results between different rivers. Moreover, if the hydrological and ecological variability is not taken into account a deterministic EFR method may generate large differences in the ecosystem protection level. The natural flow regime classification of rivers arises as a useful tool to assess the repercussions that the establishment of a certain EFR will have in the ecosystem. Several authors have pointed out the usefulness of those hydrologic classifications to define homogeneous water management units, predict the hydrological characteristic of unmeasured river reaches or to define environmental flow regimes [5][9][10]. The natural flow classification search the definition of a reduced number of homogeneous classes which also account with an environmental meaning and sense, and the extrapolation of those classes based on the relation with a set of environmental predictor variables, capable to be measured at a large scale. There are several approaches to develop hydrological classifications, although the procedures based in the analysis of measured or modelled flow series and the definition of indices related with the different ecological relevant aspects of the regime are considered the most precise and useful. We hypothesised that EFRs estimated through hydrologic methods in rivers with different hydrological character would produce contrasting effects on environmental protection level, measured as fish habitat enhancement or reduction. Several authors [11][12] have analysed the relation between the optimum habitat quality, measured as the PHABSIM outcome Weighted Usable Area (WUA), and several environmental variables, highlighting the importance of flow magnitude in their predictions. Moreover, Snelder et al. [7] observed a reverse relation between the river mean annual flow and the habitat loss when an environmental flow regime calculated with hydrologic methods was proposed for the New Zealand Rivers. On the other hand, habitat-flow relationships are not only influenced by hydrological variability but also by the physical structure of the river reach which is greatly determined by the size of the river. Thus, the variability of the rivers size within homogeneous hydrological classes might be exerting an important effect in the relationship between flow magnitude and habitat quality. Moreover, the habitat suitability model differs from one species to other, and thus the modification of the habitat in one hydrological class and the variability among classes has not to be the same for different species. Thus the present study has two objectives: 1) Analyze the consequences of the establishment of an environmental flow calculated through a hydrologic method in the habitat of adult Salmo trutta and Luciobarbus graellsii, 2) analyze the variability of these consequences between rivers belonging to different hydrologic classes and rivers of different size and 3) analyze the variability in the response of S. trutta and L. graellsii. 2.1 Study area
The study area comprised the northern third of the Iberian Peninsula, including both Atlantic and Mediterranean river basins (Figure 1). There exist a wide range of hydrologic conditions, mainly related to the influence of the humid temperate Atlantic and the dry Mediterranean climate in the north and south regions of the study zone, respectively. It can be also found high contrasting topographic conditions, from large mountainous systems to vast plain areas. Figure 1. Map of the study area showing the situation of the unaltered gauge station (Grey triangles) used in the hydrologic classification and in-stream habitat sites (Salmo trutta: Black squares; Luciobarbus graellsii: White 2.2 Hydrologic data and environmental flow values
The available hydrologic data comprised time series of daily mean flow (m3s-1) measured in 389 gauge stations distributed trough the study area (Figure 1). We selected only those stations whose flows records were unmodified, i.e., those that did not present impoundments or important abstractions in the upstream catchment and for which data was available in the period between 1976 and 2006. Afterwards a visual inspection of the complete flow series was conducted and several criteria were applied in order to eliminate those year that did not presented an optimal quality. Lastly, those years with data gaps longer than 30 days were discarded. Once those criteria were applied we selected the gauges with flow series longer than 7 years. Finally, a total of 172 gauges were retained. For each flow series we computed a set of 106 hydrological indices that characterize the central tendency (mean) and the inter-annual variation (standard deviation) of ecologically relevant aspects of the flow regime including magnitude of mean monthly and annual flows, magnitude and duration of annual extreme flows, timing and predictability of flows, frequency and duration of high and low flow pulses and rate and frequency of changes of flows. Additionally, the retained flow series were used to calculate the minimum monthly mean daily flow (Qmin) and the minimum environmental flow (Qb) according to one of the methods required by the current Spanish legislation. Specifically, we applied the basic flow method [4] which studies the discontinuities in the variation of the minimum flows through the application of moving averages to annual daily flow series to obtain the. 2.3 Theoretical River Network and environmental predictors
The spatial framework for the analysis was a digital representation of the river network derived from a 25 m elevation model comprising over 660.000 segments with lengths ranging from 200 to 1000 m. 17 environmental variables describing topography (catchment area, elevation and gradient), climate (mean annual precipitation, temperature, evapotraspiration and intraanual precipitation variability), land uses (surface occupied by agriculture, pastures, urban zones, broadleaf forest, coniferous forest, scrubs and shrubs and denuded area), geology (catchment rock hardness and permeability) and river network configuration (confluence density and drainage density) were derived for all segments from existing GIS databases. 2.4 Hydrologic classification and river size definition
In this study we developed a hydrologic classification using the predict-then-classify strategy which follows a set of successive steps (Figure 2). The approach commenced by performing an principal component analysis (PCA) based on the correlation matrix of the 106 indices to derive a parsimonious set of non-correlated synthetic indices that represent several uncorrelated dimensions of hydrological variability [13]. The broken stick method was used to define the optimal number of PCA axes to summarize the original indices. Then the gauge coordinates on each retained PC were used as synthetic indices. The next step was to fit regressions models to establish quantitative relations between the synthetic indices and the environmental predictor variables in each gauge station. We used Random Forest (RFM) [14] which comprises an ensemble of individual regression trees develop individually with a bootstrap sample of the training data and only using a small random sample of the predictor variables to define the split in each node. The introduction of this random variation increases the prediction accuracy respect other regression models. Then, predictions of the synthetic indices were made for the whole network based of the RFMs and the predictor variables associated to each segment. Lastly, based on the values of the predicted indices a classification was produced by clustering the river segments using the Partionioning Around Medoids (PAM) [15] to define the hydrological classes (Figure 2). Figure 2. Diagram showing the procedure followed in the predict-then-classify classification approach. Beyond the hydrology character of a river segment, the habitat configuration is highly dependant on the relationship between the hydraulic variables and flow modification [12]. Furthermore this physical is highly influenced by the size of the river (p.e. large rivers present different types, sequences and density of mesohabitats to small rivers) [16][17][18]. River width was used as an indicator of the river size. We have calculated river width coefficients following Eq (1): Bankf_W were A= Drainage area (km2) and P= Mean annual precipitation (mm). Then we obtained the median width of all the study sites and classify as large or small rivers if their width was higher or lower than the median value, respectively. 2.5 Minimum monthly and environmental flows prediction
RFM were also developed to predict the values of the Qmin and the Qb in all the segments of the river network. The RFM were developed based on the monthly and environmental flow values calculated from the flow series of the unaltered gauge stations and the environmental predictor variables associated to each gauge. 2.6 Weighted Usable Area (WUA) and Flow curves
Some of the most extended methods to define minimum environmental flows are based on the physical habitat simulation models for several fish species and life stages [6]. These methods link hydraulic model predictions of water velocity, depth, bottom hydraulic stress or substrate size with habitat suitability models to predict the availability of suitable habitat at various discharge rates. The availability of suitable habitat is expressed as weighted physical area (WUA) in m2. Habitat models can be used to assess the WUA over a desired range of flows and therefore make predictions of the change of habitat with the change of flow. In this study we collected 242 WUA-Flow curves of adult Salmo trutta and Luciobarbus graellsii included in different reports and works carried out by different research groups, consultants or water agencies in the last 10 years. We selected those curves situated in minimally hydrological and morphological altered river reaches. Thus, we discarded those sites situated downstream dams and reservoirs with a capacity greater the 3 Hm3. However we kept those sites downstream reservoirs in which a tributary of the same order join the mainstem upstream the site. Finally, we used 91 WUA-Flow curves for adult S.trutta and 48 WUA-Flow curves for adult L. graellsii to evaluate the habitat loss produced for a certain reduction of flow (Figure 1). 2.7 Analysis
Based on each WUA-Flow curve we calculate the WUAs at the Qmin and at the Qb previously predicted for each river segments and calculate the WUA variation between these two values. To allow comparisons we expressed the reduction of flow at Qb as a percentage of the WUA at Qmin. Box-plots diagrams were used to analyze descriptively the differences in the reduction of WUA between hydrological classes, large and small rivers and the combination of characteristics. Only classes with more than five sites were used in the analysis. RESULTUS
3.1 Hydrologic Classification
The first five axes of the PCA retained the 91% of the total hydrologic variability. The hydrological indices with the highest loadings in the PCs indicated that they represented: 1) the magnitude and duration of mean and high flows (66% of the variability); 2) the frequency of high flows and magnitude of base flow (11%); 3) Frequency and duration of flood events (7%); 4) the interannual variability of flow change ( 5%); 5) the intra and interannual variability and predicatibility of mean and high flows (3.6%). The r2 values for the fitted RFMs of the synthetic indices used to define the classification was 68% for the PC1 and then decreased with subsequent PC to PC5 (7%). The mean r2 for the five models was 44%. The most important predictor variables were topographic variables (catchment area and elevation), climatic (mean annual precipitation and temperature and intraanual precipitation variability) and land uses (surface occupied by agriculture and pastures. We selected a hydrological classification comprised by 7 classes. However, not all the classes retained at least 5 habitat simulation sites. The hydrologic classes that contained more than 5 habitat simulation models for S. trutta were class 1, 2 and 6, while for L. graellsii were the classes 2, 3 and 5. The class 1 was conformed by rivers with low annual mean flows but with frequent and short high flows and floods events. Most of the rivers of the class 1 were situated in the Atlantic basin. Class 2 and 3 comprised rivers draining to the Ebro basin (Mediterranean Sea). They presented similar mean annual flows but high flows events were more frequent in Class-2, probably due to the situation of the rivers in the mountainous areas. The rivers of the class 5 also drain to the Ebro and they presented the lowest annual mean flows which was related to the situation of the rivers in the aridest areas of the study zone. The class-6 presents the highest annual flows, although there is a high variability within the class. The rivers belonging to class 6 are situated in the eastern part of the study zone in the Pyrenees mountainous zone. In addition, the 57 % and 29 % of the habitat simulation sites for S. trutta and L. graellsii, respectively, were situated in large rivers. 3.2 Minimum environmental flows prediction
min and at the Qb obtained a model fitted r2 value greater then 0,6 in both 3.3 Analysis
None of the analysis has shown clearly differences in the habitat reduction between classes or river sizes and a high variability was founded within each river type. However general patterns can be extracted from the results. The quantity of usable habitat for S. trutta under the environmental flows condition suffered a median reduction of 20 % in the hydrologic classes 1 and 6 while this value dropped until 12 % in the class 2. The analysis of the reduction of habitat in regard to the river size did not show much differences between the sites situated in large and small rivers, however the coupling of the hydrologic classes with the river size accentuate the differences between classes. Thus, the big rivers of the class-1 and 6 exhibited a reduction in the habitat loss producing values close to class 2, while small rivers of these classes suffered a slight increase in the habitat reduction. The usable habitat reduction of L.graellsii was generally higher than for S.trutta although an objective comparison could be just carried out in class 2. Classes 2 and 3 suffered a similar reduction of WUA reaching values around 27 %. The sites in the class 6 suffered the highest reductions with a median value of 37 %. In this case, the analysis of the reduction of habitat considering the river size highlighted differences between large and small rivers. The environmental flow condition generates a median reduction of 24 % and 34 % for small and large rivers, respectively. The combination of hydrologic classes and river size enlarge the patterns described. Hence, the reduction of the WUA in the classes 2 and 3 rose up to 33 % and 36 %, respectively, when just large rivers where considered. In the class 6 this variation was not so pronounced. Nevertheless, it must be taken into account the limited number of sites to discriminate between hydrologic and river size classes Thus in 5 fives cases there were less than five sites in each group. Fig 3. Box-plots showing the variation in the reduction of WUA for Salmo trutta (A) and Luciobarbus graellsii (B) under the environmental flows conditions in the 1) different hydrological classes 2) different size rivers and 3) the combination of hydrological classes and river size. (Solid line: median reduction values; Box: 25th and 75th percentiles; Black dots: 10th and 90th percentiles; C: Class; Lg: Large Rivers; Sm: Small River). DISCUSSION
Our study has highlighted the utility of classifying rivers according to their hydrological character and size to evaluate the consequences of applying homogeneous environmental flow rules to large jurisdictional regions. The segregation of rivers to assess the effects of EFR is especially useful in regions where hydrologic conditions present a high variability [7], as it is the case of the northern third of the Iberian Peninsula. However, it must be noted that the potential usefulness of this approach is determined by the classification method utilized. In this regard, the classification strength (CS), i.e. the capacity to discriminate between classes, and the ability to interpret the environmental meaning of the classes will vary greatly from one classification method to other. The use of data-driven classification and powerful statistics methods have been proved to increase the discrimination capacity compared to other classification based solely in climatic variables and catchment configuration [19] or in non-empiric criteria [20]. Therefore, several frameworks developed in recent years recommend the use of this kind of classifications to optimize the definition of environmental flows and other water planning issues [5]. Other key task in the classification procedure is the selection of the number of classes which can be somehow a subjective routine, as the CS of different classification levels does not present significant differences in many cases [19]. In this regard, Poff et al. [5] and Kennard et al. [9] recommended the use of classification with a reduced number of classes in order to facilitate the interpretation of the classes. Thus, we considered that a classification comprising 7 classes was an optimal solution to represent the hydrological variability of the study zone and to ensure keeping a sufficient number of habitat sites in several classes for further analysis. Our results have highlighted general patterns in the reduction of habitat of two fish species and life stages according to the hydrologic classes. The most important variable in the hydrologic classification developed in this study was the mean annual flow and, in general, we observed an opposite relationship between annual flow and habitat reduction, which is consistent with a number of results obtained for multiple taxa in different countries [7][11][12]. Nevertheless, our results have also highlighted an important uncertainties associated to the prediction of WUA-flow relationships as it has been pointed out in those studies [7][12]. On one hand the results has shown an important variability in the WUA-Flow relationships within the classes and in the other, that not all the classes are in accordance with the general pattern described. In this regard, we observed a great variability in the shape and gradient of the WUA-Flow curves, and many of them do not present the classical PHABSIM bell-shaped curves as found by other authors [11]. This fact indicates the high specificity of the curve to the river reach where it was developed and not only to the hydrologic class that does belonged to. In this regard, we consider that the classification of the WUA-Curves attending to their shape and gradient, including bell-shaped, asymptotic, constantly increased, high and low gradient curves, etc, will allow us to obtain more conclusive results to relate flow modification and habitat loss. In addition, it has been demonstrated that the combination of the hydrologic classes with the river size reduced the intra-classes variability and clarified some unexpected results. For example, given the high mean annual flows of the class 6 and the general patterns described above, we expected a lower reduction of habitat than the observed. The inclusion of the river size demonstrated that most of the sites of this class were located in small reaches, which in part explained the high reduction of the habitat of S.trutta in this class. However, the two species analyzed showed an opposite response in regard to the river size. S.trutta presented a higher reduction of WUA in small reaches, only when size and hydrologic classes were combined, while L.graellsii showed an increase in the WUA reduction in large rivers. This contrasting result might be associated to the different habitat preferences of the selected species and life stages. Lamouroux et al. [12] also founded divergent relations between S.trutta and L.graellsii WUA-Flow relationship in French rivers which was attributed to the preference of L.graellsii for deep and fast flowing habitats. Moreover, Lamouroux et al. [12] showed that in the reaches dominated by riffle-type habitats all the species that they analyzed presented the same response to flow modification except for S.trutta, for which the reverse behavior was observed. In addition, the observed differences between S.trutta and L.graellsii founded in the present study could also be attributed to differences in the morphology of the sites used to analyze each species. In this sense, all the sites where the habitat of L.graellsii was analyzed were located in low reaches of Mediterranean rivers which drain to the Ebro basin. These reaches commonly present lower gradients and a minor degree of entrenchment than the Atlantic basin rivers. Given that the variation of the hydraulic characteristics in response to a flow changes differs from one type of river to the other according to its physical structure [12], this might be conditioning the final WUA-Flow relationship. Finally, it must be noted that we used the river reach width as an indicator of the habitat structure, given the widely recognized relation between channel width and habitat structure and composition [16][21]. However, our analysis did not yield conclusive results and hence, further research must be done in this issue. In this sense, two approaches can be followed in the future. Firstly, the approach developed by Lamouroux et al. [12] according to which hydraulic geometry characteristic, such as depth, velocity and particle size are predicted for a range of flows in each river reach or secondly, the approach developed by Barquin et al. [17], in which the mesohabitat composition and density of 200-1000 m length river reaches were predicted for a whole network based on large scale catchmement variables. There are several potential uses of the approach followed in this study. Firstly, it may help to establish more adequate water management rules in order to avoid the ecosystem deterioration. Given the differences between classes it would be logical whether water resource plans and legislation included different habitat reduction thresholds that take into account the hydrological and morphological variability of streams. This would have two major advantages. On one hand, it would avoid the establishment of too low environmental flows in rivers where the reduction of WUA is more pronounced and, thus, reduce the risk of produce greater ecosystem deterioration. On the other in those cases where the environmental flows calculated through simple rules resulted too environmentally conservative, other water management options could be suggested. Furthermore, our approach allows evaluating the consequences that a reduction of flow would have in the fish habitat using the methodological principles of physical habitat simulation models at large scale. Habitat simulation models are one of the most extended and scientific accepted methods to calculate environmental flows, but they required high amount of economic and personal resources. The discrimination of rivers according to elements important for the estimation of habitat quality and quantity would allow defining more reliable thresholds in each river type without developing the habitat simulation models in all the locations, which will greatly alleviate the efforts and cost, especially if studies must be done at large scale. ACKNOWELEDGMENT
This study was partly funded by the Spanish Ministry of Science an Innovation as part of the MARCE (Ref: CTM-2009-07447) project. Mario Álvarez-Cabria is supported by a Program of Postdoctoral Fellowships for Research Activities of the University of Cantabria (published by resolution on 17 January 2011). REFERNCES
[1] Postel, S. & Ritcher, B. D., Rivers for life: Managing water for people and life, Island Press, Washington, [2] Naiman, R. J., Latterell, J. J., Pettit, N. E. & Olden, J. D., Flow variability and the biophysical vitality of river systems, Geoscience, Vol. 340 (2008), pp 629-643 [3] Poff, N. L., Allan, J. D., Bain, M. B., Karr, J. R., Prestegaard, K. L., Richter, B. D., Sparks, R. E. & Stromberg, J. C., The natural flow regime. A paradigm for river conservation and restoration, BioScience, Vol. 47, No. 11, (1997), pp 769-784. [4] Palau, A. & Alcázar, J., The basic flow method for incorporating flow variability in environmental flows, River Research and Applications, Vol. 28, (2012), pp 93-102. [5] Poff, N. L., Richter, B. D., Arthington, A. H., Bunn, S. E., Naiman, R. J., Kendy, E., Acreman, M., Apse, C., Bledsoe, B. P., Freeman, M. C., Henriksen, J., Jacobson, R. B., Kennen, J. G., Merritt, D. M., O'keeffe, J. H., Olden, J. D., Rogers, K., Tharme, R. E. & Warner, A., The ecological limits of hydrologic alteration (ELOHA): a new framework for developing regional environmental flow standards, Freshwater Biology, Vol. 55, No. 1, (2010), pp 147-170. [6] Tharme, R. E., A global perspective on environmental flow assessment: emerging trends in the development and application of environmental flow methodologies for rivers, River Research and Applications, Vol. 19, No. (2003), pp 397-441. [7] Snelder, T., Booker, D. & Lamouroux, N., A Method to Assess and Define Environmental Flow Rules for Large Jurisdictional Regions, Journal of the American water resources association, Vol. 47, No. 4, (2011), pp 828–840 [8] Acreman, M., Linking science and decision-making: features and experience from environmental river flow setting, Environmental Modelling & Software, Vol. 20, No. 2, (2005), pp 99-109. [9] Kennard, M. J., Pusey, B. J., Olden, J. D., Mackay, S. J., Stein, J. L. & Marsh, N., Classification of natural flow regimes in Australia to support environmental flow management, Freshwater Biology, Vol. 55, No. 1, (2010), pp 171-193. [10] Belmar, O. & Martinez-Capel, F., Hydrological Classification of Natural Flow Regimes to Support Environmental Flow Assessments in Intensively Regulated Mediterranean Rivers, Segura River Basin (Spain), Environmental Management, Vol. 47, (2011), pp 992-1004. [11] Hatfield, T. & Bruce, J., Predicting Salmonid Habitat-Flow Relationships for Streams from Western North America, North American Journal of Fisheries Management, Vol. 20, No. 4, (2000), pp 1005-1015. [12] Lamouroux, N. & Capra, H., Simple predictions of instream habitat model outputs for target fish populations, Freshwater Biology, Vol. 47, No. 8, (2002), pp 1543-1556. [13] Olden, J. D. & Poff, N. L., Redundancy and the choice of hydrologic indices for characterizing streamflow regimes, River Research and Applications, Vol. 19, No. 2, (2003), pp 101-121. [14] Cutler, D. R., Edwards, T. C., Beard, K. H., Cutler, A. & Hess, K. T., Random forests for classification in ecology, Ecology, Vol. 88, No. (2007), pp 2783-2792. [15] Legendre, P. & Legendre, L., Numerical Ecology, Elsevier, (1998). [16] Wohl, E. & Merritt, D., Prediction of mountain stream morphology, Water Resources Research, Vol. 41, [17] Barquín, J., Snelder, T., Booker, D., Álvarez-Cabria, M., Peñas, F. J. & Fernández, D., Modelling physical characteristics of habitats from rivers reaches to entire river networks in northern Spain,, 7th Symposium for Freshwater Sciences, Girona. June 27-July 1, (2011). [18] Peñas, F. J., Fernandez, F., Calvo, M., Barquín, J. & Polo, L., Theoretical river networks structure and spatial accuracy. Do they matter for inferring in river habitat characteristics?, Limnetica., Vol. 30, No. 2, (2011), pp 197-216. [19] Snelder, T. & Booker, D., Natural flow regime classifications are sensitive to definition procedures, River [20] Peñas, F. J., Barquín, J., Snelder, T. H., Booker, D., Alvarez-Cabria, M. & Fernandez, D., Comparison of natural flow regime classifications based on different strategies, 7th Symposium for Freshwater Sciences, Girona. June 27-July 1 2011, (2011). [21] Flores, A. N., Bledsoe, B. P., Cuhaciyan, C. O. & Wohl, E. E., Channel-reach morphology dependence on energy, scale, and hydroclimatic processes with implications for prediction using geospatial data, Water Resources Research, Vol. 42, No. 6, (2006).

Source: http://www.ise2012.boku.ac.at/papers/14956_2.pdf