Defining the ecological hydrology of Taiwan Rivers using multivariate statistical methods

(1)

Deﬁning the ecological hydrology of Taiwan Rivers using multivariate

statistical methods

Fi-John Chang

a,*

_{, Tzu-Ching Wu}

a

_{, Wen-Ping Tsai}

a

_{, Edwin E. Herricks}

_b,* a

Department of Bioenvironmental Systems Engineering, National Taiwan University, Taiwan

b

Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, USA

a r t i c l e

i n f o

Article history:

Received 5 November 2008

Received in revised form 27 April 2009 Accepted 11 July 2009

This manuscript was handled by L. Charlet, Editor-in-Chief, with the assistance of Jose D. Salas, Associate Editor

Keywords: Ecohydrology

Taiwan ecological hydrology indicator system

Flow regime River restoration

Multivariate statistical methods

s u m m a r y

The identification and verification of ecohydrologic flow indicators has found new support as the impor-tance of ecological flow regimes is recognized in modern water resources management, particularly in river restoration and reservoir management. An ecohydrologic indicator system reflecting the unique characteristics of Taiwan’s water resources and hydrology has been developed, the Taiwan ecohydrolog-ical indicator system (TEIS). A major challenge for the water resources community is using the TEIS to provide environmental flow rules that improve existing water resources management. This paper exam-ines data from the extensive network of flow monitoring stations in Taiwan using TEIS statistics to define and refine environmental flow options in Taiwan. Multivariate statistical methods were used to examine TEIS statistics for 102 stations representing the geographic and land use diversity of Taiwan. The Pearson correlation coefficient showed high multicollinearity between the TEIS statistics. Watersheds were sep-arated into upper and lower-watershed locations. An analysis of variance indicated significant differences between upstream, more natural, and downstream, more developed, locations in the same basin with hydrologic indicator redundancy in flow change and magnitude statistics. Issues of multicollinearity were examined using a Principal Component Analysis (PCA) with the first three components related to general flow and high/low flow statistics, frequency and time statistics, and quantity statistics. These principle components would explain about 85% of the total variation. A major conclusion is that managers must be aware of differences among basins, as well as differences within basins that will require careful selec-tion of management procedures to achieve needed flow regimes.

Introduction

Hydrology is recognized as a critical factor in the maintenance of the ecosystem integrity of streams and rivers (Karr, 1981). In a systems view of streams and rivers, hydrologic events are known to form and maintain channel planform and substrate while inter-actions among flow and channel structure create habitats for aqua-tic organisms. The understanding of the relationships between the flow regime characteristics of a river and its ecological functioning are crucial to the developing techniques to manage integrity and develop designs for restoring damaged streams (Hughes and Hann-art, 2003; Sanborn and Bledsoe, 2006). This recognized connection between flow and ecosystems has resulted in the use of hydrologic statistics to characterize environmental flows and supports the

development of ecohydrology as a focus for applications in hydro-logic research.

Recently, there has been a call to develop comprehensive, river-specific methods to define and refine environmental flows (Arthington et al., 2006). The environmental flow standard ap-proach advocated by Arthington et al. calls for the use of hydrolog-ical statistics to develop an appreciation of natural patterns of flow variability and connecting this definition of flow to biological data. They have proposed the development of hydrological classifica-tions based on statistical analysis of hydrologic data combined with a connection to ecosystem characteristics for a range of natu-ral and modified flow conditions. Although they provided an out-line of analytical requirements, they did not provide an examination of data sets or a demonstration of environmental flow rules with a case study. This paper explores the approach advo-cated byArthington et al. (2006) using the extensive hydrologic data collected in Taiwan as a case study.

The objective of this research is to use a knowledge of the eco-logical hydrology of natural ﬂows from relatively undisturbed watersheds, mainly headwaters locations (Chang et al., 2008), with

* Corresponding authors. Tel.: +886 2 23639461; fax: +886 2 23635854 (F.-J. Chang).

E-mail addresses:changfj@ntu.edu.tw(F.-J. Chang),herricks@illinois.edu(E.E. Herricks).

Contents lists available atScienceDirect

Journal of Hydrology

(2)

analysis of additional hydrologic data from watersheds with higher levels of human activity, mid to lower-watershed locations, to develop ecological hydrology characteristics for Taiwan. The connection between hydrology and ecology is achieved through the use of the Taiwan ecohydrological indicator system (TEIS) (Suen, 2005) and autecology information for Taiwan fish species (Suen and Eheart, 2006). The development of ecological hydrology characteristics is based on the connection among flow, habitat, and organisms. Channel planform and substrate conditions interact with flow to produce the habitats occupied by aquatic organisms. With flow variation, habitat conditions change. Aquatic organisms have evolved to accommodate natural changes in flow/habitat, but alteration of the magnitude, timing, or frequency of flow change may exceed the capacity of organisms to adjust to changing envi-ronments leading to aquatic community alteration or extirpation of species. In modern watershed management the emphasis on procedures that mimic natural flow regimes is intended to enhance native fauna using a regime approach. The regime of more natural flows then provides a reasonable target for flow management. There is a clear need for a management focus on natural flow re-gimes (Poff et al., 1997) suggested that natural flows provide hab-itat suited for indigenous fauna and maintenance of more natural fish communities. A number of hydrologic statistics have been pro-posed for use as hydrologic indicators for river restoration and water resources management (Hughes and James, 1989; Poff et al., 1996; Richter et al., 1996, 1997; Clausen and Biggs, 2000; Olden and Poff, 2003).

Taiwan presents an ideal opportunity to evaluate hydrological statistics for local and regional analysis of ecohydrologic indicators. Taiwan’s land area is approximately 36,000 km2_{with mountains}

reaching 3952 m. In this relatively small area, hydrologic monitor-ing has been conducted for over 50 years, providmonitor-ing a rich resource of hydrologic data from a dense network of gauging stations. Phys-iographic conditions and climate, along with available data re-sources, make Taiwan an ideal location for the examination of natural flow issues in a sub-tropical climate. Taiwan annual mean rainfall is about 2.6 times the world annual mean rainfall, but nearly 78% of the rain occurs during the wet months of May to October. In addition, due to the presence of high mountains and limited land area the rivers are short and steep leading to rapid runoff of precipitation. Precipitation can occur in typhoon events where rainfalls in excess of 2000 mm per day have been recorded. The resulting natural flow regimes show marked seasonal period-icity punctuated by events that may produce massive flows. Flow data for Taiwan is abundant, existing for 102 stations located in watersheds with different physiography and land use. Discharge records of up to 50 years are available.

In addition to environmental conditions and flow data, previous research has provided tools to examine natural flow regime issues in Taiwan. The TEIS used hydrologic statistics identified byOlden and Poff (2003)and indicators of hydrologic alteration identified by Richter et al. (1996) with the ecological and environmental requirements of Taiwan freshwater fish species to develop an eco-hydrologic indicator system that supports new water resources management approaches in Taiwan (Suen and Eheart, 2006). At is-sue is how hydrologic statistics used in the TEIS indicators can pro-vide a useful addition to existing hydrologic analysis. In particular, a demonstration is needed that relates TEIS statistics to a better understanding of flow pattern, timing, frequency, and variability, which relates aquatic community habitat needs to well understood hydrologic conditions. Further, it is known that the actual values of hydrologic statistics can vary over a relatively small watershed areas, which effect discharge and flow concentration time. Taiwan provides the ideal location to address whether this variability influences the interpretation of regional or local ecohydrological conditions because of its dense network of flow monitoring

sta-tions and a reasonable ﬂow history for that network. In addition, the calculation of hydrological statistics has brought the recogni-tion that there is potential for redundancy in informarecogni-tion provided to decision makers. It should be possible to reduce the number of measures and still provide an accurate characterization of ﬂow re-gimes with a reduced set of hydrologic statistics that are more eas-ily understood by watershed managers.

This paper examines data from an extensive network of ﬂow monitoring stations in Taiwan and applies multivariate statistical methods to examine TEIS hydrological statistics for 102 stations representing the geographic and land use diversity of Taiwan. Analysis included use of a two-way analysis of variance (ANOVA) to determine if there were differences in TEIS statistics between upper and lower gage station locations in the same basin. Issues of multicollinearity were examined using Principal Component Analysis (PCA). The value of using an ecohydrologic approach was investigated to better connect aquatic organism requirements with a management alternatives.

Methods Two-way ANOVA

The analysis of variance (ANOVA) is a widely used collection of statistical procedures, in which the observed variance is parti-tioned into components due to different explanatory variables (Lindman, 1974). The two-way analysis of variance is an extension to the one-way analysis of variance. There are two independent variables, called factors. Each factor will have two or more levels (means) within it. Three signiﬁcance tests were used: a test of each of the two main effects and a test of the interaction of the variables. The two-way ANOVA model used in this study is of the form:

EðYijkÞ ¼

l

þ

a

iþ bjþ

a

bij ð1Þ

where Yijkis the kth observation in the ijth group,

ai

, i = 1,2

repre-sent factor A (upper and lower positions) main effects, bj, j = 1, 2,

. . .9 represent factor B (nine basins) main effects, and

a

bij, is the

AB interaction effects.

The total sum of squares SST is partitioned as:

SST ¼ SSA þ SSB þ SSAB þ SSE ð2Þ

SSA & SSB are the sum of squares for factors A and B, respectively, SSAB the sum of squares for the interaction between factor A and factor B, and SSE is the error sum of squares.

Principal component analysis, PCA

Principal component analysis (PCA) is a classical statistical method that could reduce the high dimensionality of the analyzed data while retaining most of the variation in the data set through orthogonal linear transformation of the correlated variables into a small number of uncorrelated variables called principal compo-nents. It reduces the dimension of a data set to reveal its essential characteristics and the principle components capture maximal var-iance. Principal component analysis is a popular data processing and has numerous applications in various science and engineering problems. The success of PCA depends on two important properties:

1. Principal components sequentially capture the maximum vari-ability among data set, thus guaranteeing minimal information loss.

2. Principal components are uncorrelated, so one can deal with one component without referring to others.

(3)

Case study

Taiwan ecohydrology indicator system (TEIS)

The TEIS provides a means to integrate hydrologic, ecological, and human management inﬂuences using a new synthesis of hydrologic statistics and provides a useful tool for the ecosystem-based water resources management in Taiwan (Chang and Her-ricks, 2005; Chang et al., 2008). In the TEIS, the hydrologic statistics can be related to ecological conditions through consideration of general habitat conditions (Changet al., 2008). For example, organ-ism response and ecological conditions are related to trends in ﬂow that provide an indication of how habitat needs for spawning, juve-nile rearing, or adult maintenance are met. Rate of change statistics provide measures of habitat disruption or the duration stability of habitat needed to complete organism life history. Using the mean of all positive and negative differences between consecutive values provides a measure of the rate of change in habitat, supporting the analysis of the general the suitability of those conditions for the maintenance of a target aquatic community. What is important to recognize is that individual hydrologic statistics are a useful sur-rogate for ecological condition because these statistics relate to habitat conditions of a species or a guild. The implications for water resources management, when there is a focus on information redundancy, is that some statistics may provide redundant

hydro-logic information, but are still needed as part of a comprehensive ecohydrologic analysis (Chang et al., 2008). The TEIS hydrologic statistics for magnitude, frequency, duration, and rate of change were used in this analysis,Table 1.

Description of data

For a small island, Taiwan has a relatively dense network of flow monitoring stations. The watershed characteristics of 430 stations were reviewed and only the stations with long data records (i.e. greater than 20 years) were chosen for use in this study. A final set of 102 stations from 23 river basins were identified, which dis-tributes 52 stations in the upper reaches of watersheds and 50 sta-tions in the lower reaches. The 66 hydrologic statistics of TEIS for these 102 stations were calculated using the daily average flow, which was obtained from ‘‘The Stream flow Database of Taiwan”, maintained by Water Resources Agency, Taiwan.

Selection of stations for two-way ANOVA

Two-way ANOVA was developed so that factor A addressed dif-ferences between upper and lower positions in the same basin, fac-tor B the difference between basins and facfac-tor C the interaction between location and basin characteristics. The ANOVA thus simul-taneously addressed three questions:

Table 1

The TEIS hydrologic statistics used in this analysis.

TEIS statistics Abbreviation

Group 1 – differences between consecutive values (m3

/s)

1. Mean of all positive differences between consecutive values in dry season Frate d

2. Mean of all positive differences between consecutive values in wet season Rrate w

3. Mean of all negative differences between consecutive values in dry season Rrate d

4. Mean of all negative differences between consecutive values in wet season Frate w

Group 2 – high/low ﬂow event magnitudes (m3

/s)

5. Dry season 1-day minimum Q1daymin d

9. Dry season 1-day maximum Q1daymax d

12. Wet season 1-day minimum Q1daymin w

15. Wet season 1-day maximum Q1daymin d

16. Wet season 3-day maximum Q3daymax w

Group 3 – frequency of high/low ﬂow events and reversals

19. Number of low ﬂow events with in each dry season (times) Nlow d

20. Number of low ﬂow events with in each wet season (times) Nlow w

21. Number of high ﬂow events with in each dry season (times) Nhigh w

22. Number of high ﬂow events with in each wet season (times) Nhigh w

23. Number of low ﬂow events with in consecutive three years (times/year) Nlow 3year

24. Number of high ﬂow events with in consecutive three years (times/year) Nhigh 3year

25. Number of hydrological reversals in dry season (times/year) RVd

26. Number of hydrological reversals in wet season (times/year) RVw

Group 4 – high/low ﬂow event duration (day/time)

27. Mean duration of low ﬂow events in each dry season Dlow d

28. Mean duration of low ﬂow events in each wet season Dlow w

29. Mean duration of high ﬂow events in each dry season Dhigh d

30. Mean duration of high ﬂow events in each wet season Dhigh w

Group 5 – mean 10-day ﬂows (m3

/s)

31.66. The 36 mean 10-day ﬂows Q1;Q2. . .Q36

Wet season: from May to October. Dry season: from November to next April. Low flow event: low than 25% of mean daily flow. High flow event: high than 200% of mean daily flow.

(4)

Does station location (upper or lower position in the watershed) have a signiﬁcant effect on TEIS values?

Do different river basins have significantly different TEIS values? Is there a significant interaction between location and basin that reflects different flow patterns due mainly to rainfall patterns and watershed time of concentration?

In the two-way ANOVA, there are three F statistics that are cal-culated and then used in significance tests. Two of these statistics test for the main effects and one tests for interactions. From the flow records analyzed, nine river basins with a total of 63 gauge stations were selected. Each basin had at least five gauge stations. The selected basins and physiographic characteristics (watershed area and river length) and number of stations in upper and lower locations are shown inTable 2.

Results and discussion

ANOVA of TEIS statistics from nine basins

Table 3shows the results of the two-way ANOVA. In the table the UL column presents the F-value for the difference in group means between upper and lower position in the watershed where the F-value determined for is the between-group variability/with-in-group variability, which is presented in bold when statistical signiﬁcance, p < 0.05. The difference between basins is presented in the BASIN column, and UL_{BASIN column provides the results}

for the interaction difference between UL and BASIN.

In the upper/lower (UL) watershed position assessment statisti-cal significance is found for nearly 60% of the hydrologic statistics with the majority of rate statistics and flow magnitudes showing the expected differences between upper and lower locations. In the magnitude statistics indicators of short term change are not significant. Other significant differences for UL comparisons sug-gest the number of high flow events and flow reversals in the dry season, and the low flow duration differed. These results are expected because of known concentration time and watershed area influences on flow and confirm the importance of recognizing fundamental differences between upstream locations with smaller watershed areas and steeper terrain while downstream reaches are often affected by flow management structures.

The comparisons between basins (BASIN) indicate significant differences in event frequency and duration with only the mean duration of low flow events in each dry season significantly differ-ent between basins. These results suggest that the TEIS hydrologic statistics reflect different patterns in rainfall and the hydro-geol-ogy in different basins, including the incidence of typhoon-related flows based on the significance of positive differences in consecu-tive values. Flow volume and event magnitude statistics did not suggest significant differences among basins. Different basins do have different values for TEIS statistics that reflect flow station

location and the characteristics of the hydro-geology in each basin. In this case, upper and lower location issues are subsumed in dif-ferences in the hydrology of each basin. These results will help fo-cus management to the speciﬁc needs of each basin while alerting managers to the different needs between upper and lower loca-tions in a watershed.

Few signiﬁcant differences were identiﬁed for UL_BASIN

com-parisons with only the 10 day minimum, duration of low flows in the wet season and duration of high flows in the dry season signif-icant. This result suggests little consistent interaction among basin location and location within basins except for statistics that relate to low flow and event description (seeFig. 1).

The results of these analyses are specific for Taiwan, but are consistent with similar analyses on continental scales, so applica-tion to other water resources management programs is appropri-ate. These results indicate that flow regimes will differ between upper and lower locations in watersheds in ways that are consis-tent, and predictable. Rates of change and magnitude of flows will be sufficiently different so as to require specific management ap-proaches to assure a comprehensive, watershed-scale regime-based management approach. This finding can be considered intu-itive, but the demonstration of how consistently a range of hydro-logic statistics, selected for ecohydro-logical relationships, shows this difference is an important foundation for ecological flow regime development in Taiwan, and are expected to be applicable globally. These results also demonstrate that management schemes should also be basin specific because there are significant differences in frequency and duration among basins. The relationships

summa-Table 2

The selected nine river basins and their gauge stations for two-way ANOVA.

Basins No. of stations Drainage (sq. km) Length (km)

Up Down

Danshui River Basin 9 6 2726 159

Dajia River Basin 5 2 1236 124

Wu River Basin 1 4 2026 119

Jhuoshuei River Basin 5 5 3157 187

Beinan River Basin 3 2 1603 84

Siouguluan River Basin 2 3 1790 81

Hualien River Basin 3 3 1507 57

Lanyang River Basin 2 3 978 73

Da-an River Basin 3 2 758 96

Table 3

Two-way ANOVA results of TEIS.

TEIS characteristic UL BASIN UL*_BASIN

F value p F value p F value p

Group 1 – differences between consecutive values

1. Rrate d 12.42 0.000 1.77 0.106 0.63 0.755

2. Rrate w 5.27 0.026 2.61 0.019 0.39 0.918

3. Frate d 14.71 0.000 1.72 0.118 1.66 0.133

4. Frate w 5.75 0.020 1.00 0.445 0.45 0.880

Group 2 – high/low ﬂow event magnitudes

5. Q1daymin d 2.23 0.141 0.69 0.695 1.32 0.257 6. Q10daymin d 3.44 0.070 0.64 0.743 1.40 0.222 7. Q30daymin d 4.74 0.034 0.60 0.776 1.35 0.242 8. Q90daymin d 6.41 0.015 0.67 0.711 1.24 0.295 9. Q1daymax d 8.85 0.004 0.75 0.646 0.46 0.880 10. Q10daymax d 12.01 0.001 0.29 0.967 0.68 0.704 11. Q30daymax d 12.62 0.000 0.24 0.980 0.82 0.592 12. Q1daymin w 3.76 0.058 1.05 0.413 1.80 0.101 13. Q10daymin w 6.39 0.014 1.52 0.176 2.32 0.034 14. Q30daymin w 7.50 0.008 1.34 0.247 1.70 0.124 15. Q1daymin d 4.43 0.040 1.21 0.312 0.40 0.913 16. Q3daymax w 5.18 0.027 1.37 0.236 0.44 0.893 17. Q10daymax w 6.45 0.014 1.25 0.293 0.52 0.833 18. Q30daymax w 5.52 0.023 1.04 0.419 0.75 0.645

Group 3 – frequency of high/low ﬂow events and reversals

19. Nlow d 2.25 0.140 2.95 0.009 2.08 0.057 20. Nlow w 0.02 0.881 2.42 0.026 2.03 0.062 21. Nhigh w 0.01 0.917 6.35 0.000 1.91 0.081 22. Nhigh w 4.43 0.042 3.31 0.004 1.69 0.125 23. Nlow 3year 0.44 0.509 2.87 0.011 1.80 0.100 24. Nhigh 3year 1.28 0.264 4.04 0.001 1.74 0.114 25. RVd 5.31 0.025 10.32 0.000 3.93 0.001 26. RVw 0.91 0.346 8.62 0.000 1.84 0.092

Group 4 – high/low ﬂow event duration

27. Dlow d 4.82 0.033 1.68 0.128 2.33 0.034

28. Dlow w 15.11 0.000 4.89 0.000 4.05 0.000

29. Dhigh d 0.00 0.977 2.82 0.012 4.57 0.000

30. Dhigh w 1.67 0.203 3.66 0.002 1.27 0.283

(5)

rized inFig. 2provide a guide to the consideration of the types of hydrologic statistics useful in capturing regime characteristics for different project basins and locations.

Redundancy of Taiwan ecohydrological indicators

The correlation coefﬁcient matrix of the 66 TEIS indicators was based on all 102 stations. The Pearson correlation coefﬁcient, r, showed high multicollinearity between the TEIS statistics (Table 4). There are 36% of the r-values higher than 0.8, and about 65% of the r-values higher than 0.6. In the analysis negative correlations of 11% and 10% were determined for r-values 0 to 0.2 and 0.2, respectively. This result is also consistent with other studies (Olden and Poff, 2003; Monk et al., 2007). What is important is that there is a consistency in the pattern for a group of statistics selected for

ecological significance. These results support the notion that when choosing hydrologic statistics for ecological flow regime analysis, it is possible to identify statistics that effectively characterize hydro-logic variability and provide statistics important to regime selec-tion and ecological importance. There is no need for choice between hydrologic and ecohydrologic indicators, rather there is a need to recognize which hydrologic indicators have ecological significance.

Principal component analysis (PCA) used results extracted from the 66-by-66 correlation matrix (i.e. 66 TEIS indexes) to identify subsets of statistics that describe the major sources of variation while minimizing redundancy (i.e. multicollinearity). In addition the PCA supports further assessment of the transferability of the statistics by identifying those statistics that consistently explain dominant patterns of variation related to station location

(6)

(upstream or downstream). This analysis is intended to further illustrate issues of redundancy and improve our understanding of how to best select statistics that adequately describe the hydrology while providing sufﬁcient information to ecological ﬂow regime development. To investigate the applicability and consistency of principle component analysis, three different cases have been examined: (1) 102 gauge stations, (2) 52 gauge stations in up-stream areas, and (3) 50 stations in downup-stream areas.

The results from the PCA show that the three cases have similar explained variability in first three principle components (PC1– PC3), shown inTable 5. The first principle component is related to general flow and high/low flow statistics, explaining approxi-mately 66% of the variation observed. The second principle compo-nent is related to frequency and time statistics, explaining approximately 10% of the total variation. The third principle com-ponent is related to quantity statistics and explains about 7.5% of the variation. These first three principle component would explain about 85% of the total variation. In other words, if we use the first three principle components to replace 66 TEIS statistics, we would only lose 15% hydrologic information. The relationship between TEIS statistics and the principal components can be illustrated by the principal components loading. The loading is given between 1 and 1 where higher loading values are related to increased

inﬂuence of a variable.Fig. 3shows the loading of the principal components calculated using all 102 stations and the 66 TEIS sta-tistics, including the loading for PC1, PC2, PC3, respectively. There are generally opposite results in the loading of PC1 and PC2 with the separation is due to indicators 19–26, which are frequency sta-tistics. PC3 shows the higher loading for some frequency and time indicators and different loading for the 36 10-day ﬂow average, which suggests potential seasonality.

These results suggest that is possible to replace 66 TEIS statis-tics with three principle components (PC1–PC3) and represent hydrologic variability. This simplification of the number of statis-tics that will describe hydrologic variability is often needed to make further modeling and optimization of water resources and flow possible. InChang et al. (2008)a cautionary note was pro-vided that there is a difference between adequately describing hydrologic variability, and the need for statistics that are related to flow conditions important to aquatic life in an ecological flow re-gime determination.

Conclusions

The importance of an ecological flow regime approach is recog-nized in modern water resources management, particularly in river restoration and reservoir management. A major challenge for the water resources community is developing a reasonable set of hydrologic statistics that effectively use data from an extensive network of flow monitoring stations to generate ecohydrologic indicators that improve existing water resources management. The Taiwan Ecohydrological Indicator System (TEIS) includes hydrologic statistics that reflect unique characteristics of Taiwan’s water resources and ecology. This paper examined data from an extensive network of flow monitoring stations in Taiwan to exam-ine the hydrological statistics recommended in the TEIS. In this analysis multivariate statistical methods were used to examine the TEIS hydrological statistics from 102 stations representing the geographic and land use diversity of Taiwan. Analyses focused on the following questions:

1. How is the hydrologic regime of Taiwan captured in the TEIS hydrologic statistics?

2. Does station location (upper or lower position in the watershed) have an effect on TEIS values?

3. Do different river basins have different TEIS values? 4. Is there an interaction between location and basin?

The results of this research showed that TEIS hydrologic statis-tics provided a detailed characterization of hydrologic conditions, providing information on ﬂow change, magnitude, frequency, and

4/4 1/4 0/4 General flow variables

11/14 0/14 1/14 High/low flow variables

Significant indexes / total indexes

UL BASIN UL*BASIN TEIS characteristic Frequency variables 2/8 8/8 0/8 Time variables 2/4 3/4 2/4 UL BASIN

Fig. 2. The tendency of two-way ANOVA results, the linked arrow represent high correlation between the factor with the TEIS statistics.

Table 4

Correlation coefﬁcients between 66 TEIS parameters.

Range of correlation >0.8 0.8 0.6 0.4 0.2 0 <0.2 Sum

0.6 0.4 0.2 0 0.2

Number 1582 1288 140 122 286 504 434 4356

Ratio 0.36 0.29 0.03 0.03 0.06 0.11 0.10 1.00

Table 5

Explain variances of principle component analysis of TEIS.

Principle component (%variance) First three PCA variances

I II III

Upstream (n = 52) 65.9 10.5 7.6 84.0

Downstream (n = 50) 65.9 10.6 7.1 83.6

(7)

duration. This summary of statistics in the TEIS is useful in provid-ing a detailed characterization of hydrologic regime for basins throughout Taiwan. The detailed statistics of the TEIS were exam-ined for redundancy. The results of PCA analysis suggested the 66 hydrologic statistics in the TEIS could be replaced by three compo-nents and represent hydrologic variability. What is lost in this replacement is the utility of hydrologic statistics that relate to eco-logical requirements of aquatic life. The TEIS provides a fuller char-acterization of ﬂow regime elements that is essential in the development of management strategies intended to provide eco-logical ﬂow regimes.

ANOVA results indicated differences between upstream, more natural, and downstream, more developed, locations in the same basin with a corresponding hydrologic indicator redundancy in ﬂow change and magnitude statistics. Correlation of TEIS indica-tors with basin area, elevation, and slope indicated the impor-tance of ﬂow frequency-related indicators in ecohydrology. Watersheds were separated into headwaters and mid-watershed locations. An analysis of variance (ANOVA) indicated differences between upstream, more natural, and downstream, more devel-oped, indicators in the same basin. Of importance in these re-sults is that differences between upper and lower locations in

(8)

watersheds in ways that are consistent, and predictable. The differences in rates of flow change and the magnitude of flows is sufficiently different between upper and lower locations that management approaches should be site specific to meet ecolog-ical needs. This argues for careful watershed management that considers local needs as a part of any basin management scheme.

ANOVA results also indicated that in the nine watersheds as-sessed, each basin has individual characteristics that make it diffi-cult to generalize for all basins. Although the cause of differences between basins has not been identified, it is likely that differences in rainfall patterns and the unique hydro-geology of each basin contributes to the observed differences in hydrological statistics. The major conclusion is that managers must be aware of differ-ences among basins, as well as differdiffer-ences within basins that will require careful selection of management procedures to achieve needed flow regimes.

A final conclusion that can be drawn from this analysis is that well designed ecological indicators provide a range of hydrologic statistics that adequately represent hydrologic conditions. Although the TEIS was developed for Taiwan, its foundation included indicators proposed for North America, so with modification for tim-ing related to life history of target species the conclusions about hydrologic statistics should be applicable outside of Taiwan. Unfor-tunately, simply selecting non-redundant hydrologic statistics to represent flow, magnitude, frequency, or duration do not provide sufficient information to develop ecological flow regimes. The selection of hydrologic statistics for modern regime analysis must account for species needs as is accomplished in the TEIS. In this analysis, we have provided a comprehensive review of a case study and have addressed a number of the issues raised byArthington et al., 2006.

Acknowledgments

This study is funded by the Water Resources Planning Institute, Water Resource Agency, MOEA, Taiwan, ROC. The gauging data provided by the Water Resources Agency and ﬁsh community data base provided by the Academia Sinica, Taiwan, ROC are very much appreciated.

References

Arthington, A.H., Bunn, S.E., Poff, N.L., Naiman, R.J., 2006. The challenge of providing environmental ﬂow rules to sustain river ecosystems. Ecological Application 16 (4), 1311–1318.

Chang, F.J., Herricks, E.E., 2005. Integration of Ecohydrology in Managing Water Resources. Water Resources Planning Institute Technical Report (ISBN 986-00-3534-3), Water Resources Agency of MOEA, Taiwan.

Chang, F.J., Tsai, M.J., Tsai, W.P., Herricks, E.E., 2008. Assessing the ecological hydrology of natural ﬂow conditions in Taiwan. Journal of Hydrology 354 (1–4), 75–89.

Clausen, B., Biggs, B.J.F., 2000. Flow variables for ecological studies in temperate streams: groupings based on covariance. Journal of Hydrology 237 (3–4), 184– 197.

Hughes, D.A., Hannart, P., 2003. A desktop model used to provide an initial estimate of the ecological instream ﬂow requirements of rivers in South Africa. Journal of Hydrology 270 (3–4), 167–181.

Hughes, J.M.R., James, B., 1989. A hydrological regionalization of streams in Victoria, Australia, with implications for stream ecology. Australian Journal of Marine Freshwater Research 40 (3), 303–326.

Jolliffe, I.T., 2002. Principal Component Analysis, second ed. Springer, NY. Karr, J.R., 1981. Assessment of biotic integrity using ﬁsh communities. Fisheries 6

(6), 21–27.

Lindman, H.R., 1974. Analysis of variance in complex experimental designs. W.H. Freeman & Co, San Francisco.

Monk, W.A., Wood, P.J., Hannah, D.M., Wilson, D.A., 2007. Selection of river ﬂow indices for the assessment of hydroecological change. River Research and Applications 23 (1), 113–122.

Olden, J.D., Poff, N.L., 2003. Redundancy and the choice of hydrologic indices for characterizing streamﬂow regimes. River Research and Applications 19 (2), 101–121.

Poff, N.L., Tokar, S., Johnson, P., 1996. Stream hydrological and ecological responses to climate change assessed with an artiﬁcial neural network. Limnology and Oceanography 41 (5), 857–863.

Poff, N.L., Allan, J.D., Bain, M.B., Karr, J.R., Prestegaard, K.L., Richter, B.D., Sparks, R.E., Stromberg, J.C., 1997. The natural ﬂow regime; a paradigm for river conservation and restoration. BioScience 47, 769–784.

Richter, B.D., Baumgartner, J.V., Powell, J., Braun, D.P., 1996. A method for assessing hydrologic alteration within ecosystems. Conservation Biology 10 (4), 1163– 1174.

Richter, B., Baumgartner, J., Wigington, R., Braun, D., 1997. How much water does a river need? Freshwater Biology 37 (1), 231–249.

Sanborn, S.C., Bledsoe, B.P., 2006. Predicting streamﬂow regime metrics for ungauged streams in Colorado, Washington, and Oregon. Journal of Hydrology 325, 241–261.

Suen, J.P., 2005. Ecologically Based Methods for Multi-Objective Water Resources Management in Taiwan, Ph.D. Dissertation, University of Illinois at Urbana-Champaign.

Suen, J.P., Eheart, J.W., 2006. Reservoir management to balance ecosystem and human needs: incorporating the paradigm of the ecological ﬂow regime. Water Resources Research 42 (3), W03417.