Comparative hydrology of Mediterranean shrublands : promo 1990-1992
Item Type Thesis-Reproduction (electronic); text
Authors Andreassian, Vazken Paul Jerome.
Publisher The University of Arizona.
Rights Copyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
Download date 27/09/2021 19:02:01
Link to Item http://hdl.handle.net/10150/192070 COMPARATIVE HYDROLOGY OF
MEDITERRANEAN SHRUBLANDS
by
Vazken Paul Jérôme Andréassian
A Thesis Submitted to the Faculty of the
SCHOOL OF RENEWABLE NATURAL RESOURCES
In Partial Fulfillment of the Requirements For the Degree of
MASTER OF SCIENCE WITH A MAJOR IN WATERSHED MANAGEMENT
In the Graduate College
THE UNIVERSITY OF ARIZONA
1992 2
STATEMENT BY AUTHOR
This thesis has been submitted in partial fulfillment of requirements for an advanced degree at The University of Arizona and is deposited in the University Library to be made available to borrowers under rules of the Library.
Brief quotations from this thesis are allowable without special permission, provided that accurate acknowledgement of source is made. Request for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the head of the major department or the Dean of the Graduate College when in his or her judgement the proposed use of the material is in the interests of scholarship. In all other instances, however, permission must be obtained from the author.
SIGNED:
APPROVAL BY THESIS COMMITTEE
This thesis has been approved on the date shown below:
Vicente L. Lopes Date Assistant Professor of Watershed Management
i. "/E)wvt 10/1/5 2_ Richard H. Hawkins Date Professor of Watershed Management
(t tTZ tojqi D. PilIip Guertin Date Assistant Professor of Watershed Management 3
ACKNOWLEDGEMENTS
I wish to express sincere thanks to all the faculty members who helped me during the realization of this thesis. Dr Martin Fogel played a special role by helping me to get into the watershed management program, and provided a safe and quiet study haven in his office. Dr Vicente Lopes was helpful from the first day I came on campus. He followed closely the elaboration of my research plan, providing a continuous guidance and showing much interest and enthusiasm for my research. Dr Richard Hawkins was also helpful by providing his own software and unpublished studies for this research, and discussing rugby issues with me. Dr Philip Guertin initiated my interest for chaparral and encouraged me to look for foreign sources of data, he has therefore a special responsibility in this thesis. Dr Peter Ffolliott gave very useful advice in the early phase of this research, and helped me define better its topic.
Dr Eric Parent at the ENGREF Paris, provided very efficient and strategic support of my project to come to study at the University of Arizona, which without him would have probably not been successful.
The help of all the research institutions who made available their data for this study is gratefully acknowledged. Data were provided by Dr Jacques Lavabre (CEMAGREF, France), Mr David Scott (Jonkershoek Forestry Research Centre, South Africa), Dr Philip Riggan (USDA Forest Service, Riverside, California), Dr Malchus Baker (USDA Forest Service, Tempe, Arizona) and Dr J. Pifiol (CREAF, Spain).
Financial support to study at the University of Arizona was provided by the Ministère de l'Agriculture et de la Forêt, and the Ministère des Affaires Etrangères (Programme des Bourses Lavoisier).
However, the most important support of all has been the friendship of Michael Ellingson, of the members of the U of A rugby team, and the continuous encouragement of my family. My aunt Aghavni and my cousin Ida were a great support for the defense of this thesis. 4
TABLE OF CONTENTS Page
LIST OF ILLUSTRATIONS 9
LIST OF TABLES 10
ABSTRACT 1 2
CHAPTER ONE: INTRODUCTION 13
1.1 Hydrology and Ecology : a missing link? 13
1.2 Objective 15
1.3 Methods 16
1.4 Benefits achieved 14
CHAPTER TWO: LITERATURE REVIEW 19
2.1 Mediterranean-type shrubland ecosystems and their geographic repartition 19
2.1.1 Definition of mediterranean-type shrublands 19
2.1.2 Repartition of mediterranean-type shrublands 20
2.2 Evidences of ecological similarities between mediterranean shrublands 21
2.3 The need for Comparative Hydrology 24
2.4 Is there really a link between Ecology and Hydrology? 26 5
Table of Contents -- Continued Page
CHAPTER THREE: NATURE AND ORIGIN OF THE ANALYZED DATA. 28
3.1 Review of the watersheds used in the comparison 29
3.1.1 Mediterranean-type shrubland watersheds 29
3.1.1.1 France: Mal Collobrier (RC) Research Watersheds 31
3.1.1.2 South Africa 32
3.1.1.3 California (USA): San Dimas Experimental Forest (SDEF) 33
3.1.2 Non-mediterranean chaparral watershed 34
3.1.3 Mediterranean-type shrubland watershed used in validation 36
3.2 Limits to the analysis imposed by the data 37
CHAPTER FOUR: TESTING THE FIRST HYPOTHESIS 39
4.1 Characterization of Watershed Behavior using a two-slope linear rainfall-runoff model 39
4.1.1 Methodology used to separate baseflow using daily rainfall-runoff data 39
4.1.1.1 Use of Storm Data in hydrological analysis 39
4.1.1.2 Generation of stormflow data 40 6
Table of Contents -- Continued Page
4.1.2 Computation of model parameters and reasons why Hawkins' method was not used for comparative analysis 46
4.2 Establishment of regression-type relationships to describe the watersheds' dynamic hydrological behavior 48
4.2.1 Research of the best regression equations to describe mediterranean shrubland watersheds' hydrological behavior 48
4.2.2 Relationship between annual precipitation and streamflow values (Model 1) 49
4.2.3 Relationship between precipitation and streamflow during the wet season (Model 2) 54
4.2. 4 Relationship between precipitation and streamf low during the dry season (Model 3) 58
4.2.5 Watershed characterization using stormf low data 61
4.2.6 Conclusion relative to the 4 relationships chosen to describe mediterranean shrubland watersheds' hydrological behavior 65
4.3 Test of results homogeneity for Models 1 through 4 66 7
Table of Contents -- Continued Page
4.4 Selection of relevant and non-correlated identifiers by Principal Components Analysis (PCA) 69
4.4.1 Selection of identifiers 70
4.4.2 Discussion of the PCA results 74
4.4.3 Stepwise multiple regression analysis 76
4.4.4 Validation of Model 1 using an additional watershed 82
4.4.5 Conclusions relative to Hypothesis 1 83
CHAPTER FIVE: TESTING THE SECOND HYPOTHESIS 85
5.1 Assessment of botanical similarities between interior and coastal chaparral 85
5.2 Comparison of the hydrological behavior of chaparral in Arizona and California 86
5.3 Conclusions relative to Hypothesis 2 89
CHAPTER SIX: SUMMARY, CONCLUSIONS AND RECOMMENDATIONS 90
6.1 Summary of the procedures used and conclusions on the two hypotheses tested 90
6.2 Conclusions relative to the selected identifiers 91
6.3 Limits of the study 92
6.4 Recommendations 92 8
Table of Contents -- Continued Page
APPENDIX A: Addresses of the Institutions that provided the data used in this study 95
APPENDIX B: Topographic maps of the watersheds used in this study 99
APPENDIX C : Plots of the four descriptive relationships used for watershed behavior characterization 107
LITERATURE CITED 136 9
LIST OF ILLUSTRATIONS
Figure Page
2.1 World Map of the Areas with Mediterranean-type Shrublands (Di Castri, 1981) 22
2.2 Degree of Ecological Similarity between the Different Mediterranean-type Shrublands of the World (Di Castri, 1981) 25
4.1 Separation of Baseflow and Stormflow 41
4.2 Baseflow Separation Using Daily Hydrographs 44
4.3 Two-straight-lines model for the rainfall-runoff relationship 47
4.4 Model 1 52
4.5 Model 2 56
4.6 Model 3 60
4.7 Model 4 63
10
LIST OF TABLES
3.1 Summary of the general characteristics of the mediterranean-type shrubland watersheds used in this study 30
3.2 General characteristics of the non-mediterranean chaparral watershed used in this study 35
3.3 General characteristics of the mediterranean-type shrubland watersheds used for validation 36
4.1 Summary of the storm events analyzed in this chapter 45
4.2 Coefficient values and significance level for Model 1 53
4.3 Coefficient values and significance level for Model 2 57
4.4 Coefficient values and significance level for Model 3 59
4.5 Comparison of Model 4 coefficients obtained when using original and matched (ranked) pairs 62
4.6 Climatic Variables Selected to Analyze Mediterranean Shrubland Watersheds' Behavior 71
4.7 Morphometric Variables Selected to Analyze Mediterranean Shrubland Watersheds' Behavior 72
4.8 Dynamic and Static Hydrological Output Variables Selected to Describe Mediterranean Shrubland Watersheds' Behavior 73
11
List of Tables -- continued Page
4.9 Multiple Regression Equations Obtained for the Hydrological Behavior Variables of the Mediterranean Shrubland Watersheds 78
4.10 Computed and estimated values of Model l's coefficients for Placer County Watershed A 82
5.1 Parameters Values of the Models 1 to 4 for the Interior and Coastal Chaparral Watersheds 87
5.2 Predicted Parameters Values for the Interior and Coastal Chaparral Watersheds 88 1 2
ABSTRACT
A comparative hydrology approach was proposed to analyze the hydrological behavior of small mediterranean shrubland watersheds and provide a potential basis for regionalization studies. Four simple regression- type models were used to describe the behavior of nine mediterranean shrubland watersheds from France, California and South Africa. The values of the models' paramaters were found significantly different. To explain the differences, a set of morphometric and climatic variables was selected, and a stepwise regression analysis was done. The results show that most of the variability encountered among the parameters was the effect of the watersheds' morphometry and of the summer rainfall amount in each region.
Additional watershed data is needed to validate the parameters' explanatory equations. However, the approach proposed here seems to present valuable opportunities for the comparative study of other vegetation types, the
regionalization of hydrological models parameters, and the study of the effects of watershed management practices. 13
CHAPTER ONE:
INTRODUCTION
1.1 Hydrology and Ecology : a missing link?
At first sight, ecology and hydrology seem to have little in common.
However, when studied at a watershed level, it becomes obvious that these two scientific domains overlap each other, and that they could probably gain much if studied together.
Watershed hydrology has a long history, but the quantitative estimation of a watershed's regime parameters was not initiated until the
1850s with two French scientists, Edmé Mariotte and Charles Perrault. They estimated the discharge of the Seine river at two different points, and had the idea to compare it to the total precipitation falling on the catchment area
(Biswas, 1970). Their purpose was to prove that the amount of precipitation falling on the drainage basin is enough to sustain the flow of rivers. Since the beginning of the century, watershed experiments have been used to study the effects of vegetation (and its management) on water yield (Bosch and Hewlett, 1982). They have been somewhat deceiving, because of the wide variation in observed results, which led Hibbert (1969) to conclude that
"response to treatment is highly variable, for the most part, unpredictable."
Therefore, generalization of results from experimental watersheds to other 14 areas, or even only to other ungaged catchments in the same region remains a difficult task.
Comparative hydrology is a subfield of hydrology which attempts to find the interactions between a watershed's ecosystem and its hydrology
(Falkenmark, 1989). It could be a very useful tool for hydrological regionalization. Chapman (1989) states that "the ideal scheme for hydrological regionalization would be based on an ecological classification, since it is the natural ecology of an area which most effectively integrates the different features of the hydrological regime." If it was possible to define the hydrological regime of a watershed mainly through its ecological characteristics, the regionalization of hydrological parameters and patterns would be easier to conduct. But in order to follow Chapman's recommendation and use the ecological classification as a basis for regionalization and generalization of watershed studies' results, the following question must be answered: is there a common hydrological behavior associated with similar ecological conditions?
Mediterranean-type shrublands are particularly interesting to study the assumption of hydrological similarity in related ecosystems. Mediterranean- type climates around the world have generated evergreen sclerophyllous shrublands, which all present interesting structural, morphological and
successional similarities (Specht, 1969; Di Castri and Mooney, 1973). Called 1 5 chaparral in California, maquis and garrigue in the Mediterranean basin, fynbos in South Africa, these shrublands are all important for the water supply of coastal areas where they occur. Extensive studies have been made and experimental watersheds have been established, making comparative hydrological studies possible.
1.2 Objective
The objective of this study is to assess the possibility of defining a global hydrological behavior on the basis of the ecological characteristics of a watershed. Mediterranean-type shrublands were used because given their striking ecologic convergence, it may be possible to demonstrate a parallelism between ecology and hydrology. The behavior patterns of mediterranean shrublands' hydrology were first described and analyzed. A second step involved testing hypotheses relative to the similarity of regimes
among mediterranean shrubland watersheds. 16
1.3 Methods
Two hypotheses were successively tested:
Hypothesis 1
All mediterranean shrubland watersheds have a similar
hydrological behavior.
This hypothesis allowed to verify whether the climate is the major controlling factor in the hydrology of mediterranean shrublands. The hypothesis was tested by comparing a sample of mediterranean shrubland watersheds with each other. Identifiers that describe best the hydrological behavior of the watersheds were first selected. Emphasis was put on retaining those identifiers that integrate the watersheds dynamic hydrological response, rather than using 'static' means. Then, an attempt was made to establish predictive equations for the main hydrological characteristics of the mediterranean shrubland watersheds.
Hypothesis 2
Botanically similar chaparral shrublands occurring under a
different climate have a similar hydrological behavior. 17
This hypothesis allows to test with more detail the relationship between watersheds' ecology and hydrological regime. It helped to decide which identifiers allow to discriminate between botanically similar but climatically different vegetation types. Since the first hypothesis hadn't been rejected, it was interesting to know at what level the ecology-hydrology link had to be defined. As an example, a comparison between mediterranean and non-mediterranean shrublands (California and Arizona chaparral) was attempted, in order to establish if the similarity of hydrological behavior observed among mediterranean shrublands was not the attribute of a broader "shrubby" vegetation type. The choice of interior (Arizona's) chaparral for this comparison is due to its geographical proximity and to some similarities in botanical composition with coastal (California's) chaparral. When shrublands are botanically similar, it should be easier to demonstrate similarities in the hydrological behavior of shrublands under different climates.
1 .4 Benefits achieved
Benefits were achieved both in the theoretical and applied domains:
.From the theoretical standpoint, benefits came from the trial of a
comparative hydrology approach for regionalization purposes, and also from
an improved understanding of the mediterranean shrubland watersheds' 18 hydrological behavior.
.Applied results (predictive equations) also were produced. These equations could be useful for engineering applications to small ungaged watersheds, even though the limited size of the studied sample can limit their accuracy.
These benefits are of real value, because of the importance of mediterranean-type shrublands in the water supply of densely populated coastal areas, such as southern California, the Mediterranean basin and Cape
province, and because of the present lack of an efficient approach to
regionalization of research results. Even the best watershed experimental
results are of little interest without a theoretical basis for their
regionalization. 19
CHAPTER TWO:
LITERATURE REVIEW
2.1 Mediterranean-type shrubland ecosystems and their geographic repartition
2.1.1 Definition of mediterranean-type shrublands
It is necessary to define what has been considered to be a mediterranean-type shrubland in this study. For Di Castri (1981), mediterranean-type shrublands are "scrub formations, found primarily within the xerothermic range of mediterranean climates, characterized by the dominance of woody shrubby plants with evergreen, broad and small, stiff and thick (sclerophyll) leaves, an overstory of small trees sometimes being present, and with or without an understory of annuals and herbaceous perennials." If all the watersheds used in the present study fall within this definition, the shrublands types represented on the studied watersheds are broader than those described by Di Castri:
(1) the south african fynbos has been considered as a shrubland and
not as a heathland (Di Castri classified fynbos as a heathland). This was justified by the fact that the term fynbos is not commonly used with the very restricted meaning of a "heathland occurring on oligotrophic soils under
a mediterranean climate." More often, fynbos is considered to be "the south 20 african equivalent of maquis and chaparral" (Di Castri, 1981). Furthermore, as Di Castri recognizes it, the "differentiation between mediterranean shrublands and mediterranean heath/ands of the same region is not easy." It was therefore decided to include the fynbos watersheds in the group of the mediterranean-type shrubland watersheds, because the literature showed that they were subject to a mediterranean climate and had mostly a shrubby cover (van der Zel and Kruger, 1975; van Wilgen and Kruger, 1985).
(2) all the vegetation type classified in France as maquis has been considered to correspond to a mediterranean shrubland, because it was not possible to establish a clear limit between shrubland and woodland.
To complete the definition of mediterranean-type shrublands, it is useful to define the mediterranean climate itself: according to Nahal (1981), it is a climate where "rainfall is concentrated in the cold or relatively cold seasons of the year, summer, the hottest season, being dry". All the watersheds used in this study were characterized by a typical mediterranean climate.
2.1.2 Repartition of mediterranean-type shrublands
The repartition of mediterranean-type shrublands is presented in Figure
2.1 (Di Castri, 1981). They can be found on the five continents and in
Australia, in narrow strips along coasts. It must be noted that the map 21 doesn't show the precise areas where the mediterranean shrublands actually exist, but the zone where they might be found. Mediterranean shrublands have a very discontinuous distribution, due to the relief which is usually very pronounced (the only exception being Australia), and to strong anthropic modification of the environment (especially in the mediterranean basin).
As Figure 2.1 shows, mediterranean shrublands occur in three heavily populated areas of the world: the Mediterranean basin, the Cape province and southern California. The water supply from mediterranean shrubland watersheds has a great importance, and a better knowledge of their
hydrological behavior is needed.
2.2 Evidences of ecological similarities between mediterranean shrublands
Similarity among species can either be the consequence of a common
descent (the species are then said to be homologous), or it can occur among
unrelated species which developed analogous structures under the same
evolution pressure; this phenomenon is called convergent evolution
(Johnson, 1973). At an ecosystem level, the concept of convergence applies
principally to structures and functions inside the ecosystem, and not to
organisms.
Evidences of ecological similarities among mediterranean - type
shrublands have been presented by numerous authors. Specht (1969), who 22 23 studied the sclerophyllous vegetation of the mediterranean-type climates of
France, California and Australia states : "The vegetation of the drier mediterranean-type environments has many structural and phenological characteristics in common. All show similar life-form spectra and possess evergreen sclerophyllous leaves and lignotubers. Maximum flowering and shoot growth occur in spring and are usually terminated by the onset of hot, dry summer conditions."
Di Castri and Mooney (1973), P refacing a compilation of studies covering all the aspects of mediterranean type ecosystems' ecology, assess that "no other disjunct pieces of land present such striking similarities as the widely separated regions with a mediterranean type of climate, that is, the territories fringing the Mediterranean sea, California, Central Chile, and the southernmost strips of South Africa and Australia. Similarities are not confined to climatic trends, but also reflected in the physiognomy of the
vegetation, in land use patterns, and frequently in the general appearance of
the landscape".
It seems well accepted by the scientific community, that the
mediterranean-type vegetation scattered around the world is more than
superficially alike, and that the observed convergence takes place not only at
the species level, but also at the ecosystem level (Johnson, 1973).
However, Barbour and Minnich (1990) criticize what they call the "myth of 24 chaparral convergence." According to them, "the similarities among the five chaparral vegetation types scattered on five continents are dramatic, but ultimately they are seen to be rather few and superficial in comparison to the differences." This must remind us that, in spite of the similarities, differences still exist, and that all the five mediterranean regions of the world don't share the same extent of similarity with each other. Di Castri (1981),
using numerical classification methods identified 2 clusters of greater affinity
on the basis of phylogenetic relations between taxa, physiographic, climatic,
and phenologic characteristics. California, Chile and the Mediterranean basin
formed the first group of closest similarity, and Australia and South Africa
the second (Figure 2.2).
2.3 The need for Comparative Hydrology
The phenomenon of convergence in ecosystems' structure and
function led some hydrologists to ask themselves if convergence also could
extend to the hydrological behavior of watersheds. This type of problem is
dealt with by Comparative Hydrology. Falkenmark (1989), defines it as "the
study of the character of hydrological processes as influenced by climate
and the nature of the earth's surface and subsurface. Emphasis is placed on
understanding the interactions between hydrology and the ecosystem, and
determining to what extent hydrological predictions may be transferred from 25
CALI ;ORNIA 4.111=111. MED ITERRANEAN BASIN
/
CHILE 11111MIMMI SOUTH AFRICA 4111111. AUSTRALIA
Figure 2.2 : Degree of Ecological Similarity Between thé Different Mediterranean-type Shrublands of the World (Di Castri, 1981) 26 one area to another." Indeed, transferring results from an experimental watershed to ungaged drainage basins is one of the major difficulties faced by hydrologists, and also one of the most crucial processes for the generalization of research results.
Comparative hydrology also deals with more theoretical aspects of watershed behavior. One of its purposes is "to delineate regions where sufficient hydrological similarity can be assumed to justify the application of techniques for characterizing hydrological processes" (Chapman, 1989).
Again, it seems that using ecologically based information could be of great help in improving hydrologists' knowledge of drainage basins operation.
Could we eventually extend or transfer hydrological results from one mediterranean type shrubland to another? The unique and "striking similarities" reported by Di Castri and Mooney (1973) make mediterranean shrublands an ideal study area for Comparative Hydrology.
2.4 Is there really a link between Ecology and Hydrology?
An attempt of grouping watershed behavior on the basis of land use was made by Hawkins (1992), who sought to associate drainage basins' response types and vegetative cover. He found "only mild associations between response type or coefficient values and the four vegetative covers
(Forest, Range, Agriculture and Urban)." This suggests that it is necessary to 27 look for a more accurate and more detailed description of vegetation types, in order to make successful associations.
The former conclusion also is supported by the work of Seyhan
(1982), who concludes from computations using a large number of watersheds, that "runoff was highest correlated with the hydromorphometrical variables." But again, the vegetation classes (cropland, pasture or grassland, woodland, impermeable living quarter and roads, grassland with scattered trees, bare land, ponds and lakes) were so vague, that it is not surprising that they couldn't allow any substantial association with the hydrological regime.
However, Seyhan's conclusion relative to the importance of hydromorphometrical variables is supported by many studies, for example the extensive work of Zavoianu (1985). It shows that in order to analyze the link between ecology and hydrology, it is necessary to understand and identify the influence of morphometric variables on watershed behavior. 28
CHAPTER THREE:
NATURE AND ORIGIN OF THE ANALYZED DATA
To conduct this comparative hydrology study, it was necessary to obtain hydrological data sets from mediterranean-type shrubland watersheds.
The data searched for were primarily daily time series of rainfall - runoff, since they could provide numerous opportunities for hydrological analysis, including the generation of storm hydrographs and the computation of monthly, seasonal and yearly flow statistics. Given the limited amount of time available to contact research Institutions in other countries, it has not been possible to obtain data from all the mediterranean-type shrublands of the world (no data from Chile or Australia), nor to constitute a large sample of watersheds. This is unfortunate, since an extended sample would allow a more extensive study. The core of data came from South Africa, France and
Southern California. Data on a northern California watershed was found in the literature. As only limited information was available for northern
California, these data couldn't be used in the initial analysis, but served as supplementary data to assess the predictive power of the equations derived
during this study. 29
3.1 Review of the watersheds used in the comparison
The main characteristics of the watersheds are presented in Tables
3.1, 3.2 and 3.3. Addresses of the research Institutions which collected and provided the hydrological data are given in Appendix A.
3.1.1 Mediterranean-type shrubland watersheds 30
Table 3.1 - Summary of the general characteristics of the mediterranean shrubland watersheds used in this study.
Watershed Country vegeta- Area Mean Mean tion type (km 2 ) Annual Annual Precipi- Stream- tation (mm) flow (mm)
Langrivier South Africa fynbos 2.46 2261 1557
Bakkerskloof South Africa fynbos 3.56 1397 615
Real Collobrier #5 France maquis 9.25 1233 498 (RC 5)
Réal Collobrier #6 France maquis 8.41 1118 432 (RC 6)
Real Collobrier #7 France maquis 1.52 1088 498 (RC 7)
Real Collobrier #8 France maquis 1.47 1230 715 (RC 8)
San Dimas California chaparral 0.31 678 91 Experimental Forest (USA) #81 (SDEF 81)
San Dimas California chaparral 0.41 678 75 Experimental Forest (USA) #82 (SDEF 82)
San Dimas California chaparral 0.25 678 70 Experimental Forest (USA) #83 (SDEF 83)
San Dimas California chaparral 0.14 678 84 Experimental Forest (USA) #84 (SDEF 84) 31
3.1.1.1 France: Réal Collobrier (RC) Research Watersheds.
The Réal Collobrier (RC) Research Watersheds are located in the
Maures mountains, approximately 6° 20' E and 43 0 15' N, twenty kilometers from the harbor of Toulon on the Mediterranean sea. The sub- watersheds used in this study are the drainage basins #5 (Valescure), #6
(Maurets), #7 (Vaubarnier) and #8 (Rimbaud). These watersheds are covered by a "natural" vegetation (maquis shrubland-woodland), the main species being Quercus spp. and Arbutus spp. The substratum consists of metamorphic rocks, slopes are steep, and altitude varies between 200 and
710 meters a.s.l. (Lavabre, 1989; CEMAGREF, 1990). Soils are mostly sandy and shallow, but the bedrock is fractured and deeply weathered, allowing local accumulation of soil that can be tapped by the deeply - rooted vegetation. According to Emberger's classification of mediterranean climates
(Nahal, 1981), the climate is cool humid.
Data was made available by Dr Jacques Lavabre at the Centre
National du Machinisme Agricole, du Génie Rural, des Eaux et des Forêts
(CEMAGREF) in Aix-en-Provence (France). The data used in this study were daily rainfall-runoff time series, and temperature data from the meteorologic station of Collobrières. Computations were made using the water years
1969 to 1988 for the precipitation and runoff (water year: October to 32
September), and 1970 to 1983 for the temperatures. Maps of all the watersheds described in this chapter are presented in Appendix B.
3.1.1.2 South Africa
The Langrivier and Bakkerskloof watersheds are located in the south- western part of the Cape province of South Africa. The Langrivier catchment is located in the Jonkershoek mountains (18° 58' E, 33° 59' S); its altitude varies between 370 and 470 meters a.s.I.; it is the control catchment in the
Jonkershoek experimental watersheds group. The Bakkerskloof catchment is located in the Klein Drakenstein mountains (19° 03' E, 33° 49' S); its altitude varies between 290 and 850 meters a.s.I.; it is the control catchment in the Zachariashoek experimental watersheds group.
Both watersheds are covered with a fynbos vegetation (van Wilgen and Kruger, 1985; van Wyk, 1987). In Langrivier, the vegetation is a tall open to closed shrubland, dominated by Protea spp., Brunie nod/flora and
Widdringtonia nodiflora. In Bakkerskloof, the vegetation-type is a tall grassland dominated by Restionaceae and Cyperaceae, with numerous lower shrubs and sporadic taller shrubs (Lindley et al., 1988).
The main substratum in both catchments is sandstone, and Langrivier
is characterized by high cliffs in its upper part. Soils are mostly sandy and
deep, except on the cliffs where the bedrock is exposed. According to 33
Emberger's classification (Nahal, 1981), the mediterranean type climate is temperate superhumid in Langrivier and temperate humid in Bakkerskloof.
Data were made available by David Scott at the Jonkershoek Forest
Research Centre in Stellenbosch (South Africa). For Langrivier, the data used were rainfall-runoff time series for the water years 1982 to 1990 (water year: April to March). Though values were daily, we could only generate yearly and seasonal means with the rainfall gages available, using the regression equation given by van der Zel and Kruger (1975). Pan-evaporation data was available for the water years 1977 to 1990 from the
Swartboskloof meteorological station. Stormflow data came from a study by
Hawkins (1992). For Bakkerskloof, the data used were weekly rainfall-runoff time series for the water years 1970 to 1989. Pan-evaporation data was available for the water years 1977 to 1990 from the Jonkershoek weather station.
3.1.1.3 California (U.S.A.): San Dimas Experimental Forest (SDEF)
The San Dimas Experimental Forest (SDEF) is located in the San
Gabriel mountains of California, 45 km northeast of Los Angeles. The four
catchments used in this study are watersheds # 81, 82, 83 and 84. They
are located side by side in the upper part of Bell canyon, at approximately
117° 46' W and 34° 12' N. Their elevation varies between 760 and 1070 34 meters, and their exposure is to the south. The vegetation is chamise chaparral, the main species being Adenostoma fasciculatum and Ceano thus spp (Hill, 1963). The substratum is mainly metamorphic rocks (mostly gneiss), with some intrusions of rhyolite. It is deeply weathered (Crawford,
1962). Soils are shallow to fairly deep, and excessively drained. According to Emberger's classification (Nahal, 1981), the mediterranean-type climate is cool subhumid.
Data for the Bell watersheds were provided by Dr Philip Riggan, at the
Forest Fire Laboratory in Riverside (California). They were daily rainfall - runoff time series (rainfall from the master raingage at Tanbark Flats). Available water years were 1939 to 1944 and 1953 to 1958 for SDEF 81, and 1939 to 1946 and 1948 to 1958 for SDEF 82, 83 and 84. Pan-evaporation data were available for the water years 1936 to 1958.
3.1.2 Non mediterranean chaparral watershed, Arizona (U.S.A.): Three Bar
Watersheds (TB)
The general characteristics of the non mediterranean chaparral watershed used in this comparison are given in Table 3.2 below. 35
Table 3.2 - General characteristics of the non-mediterranean chaparral watershed used in this study.
Watershed Area Mean Annual Mean Annual (km 2 ) Precipitation Streamf low (mm) (mm) Three Bar D 0.33 785.2 103 (TBD)
The Three Bar D watershed is located in Central Arizona, on the
northern slopes of the Mazatzal mountains, near Lake Roosevelt. It is the control catchment of the Three Bar experimental watersheds group.
Elevation varies between 1200 and 1500 meters. Vegetation is dominated
by Quercus turbinella and Cerocarpus betuloides, with Ceanothus spp.
present. Substratum is granitic and deeply weathered (down to 6-12 m), soil
are very gravelly and permeable (Hibbert, 1971).
Data were provided by Dr Malchus Baker at the USDA Forest Service
Rocky Mountain Forest and Range Experiment Station, in Tempe (Arizona).
They were daily rainfall and water heights at the gaging station. The water
years (October to September) used were 1957-1958, 1964 to 1965, and
1967 to 1978. 36
3.1.3 Mediterranean-type shrubland watershed used in validation
One of the mediterranean shrubland watersheds had too much missing information to be used in the initial analysis. It was thus decided to use it for validation of the relationships derived from the sample of mediterranean-type watersheds. The characteristics of this watershed are presented in Table
3.3:
Table 3.3 : General characteristics of the mediterranean-type shrubland watershed used for validation
Watershed Country Vegetation Area Mean Mean type (km2) Annual Annual Precipi- Stream- tation flow (mm) (mm)
Placer U.S.A., Woodland 0.19 623 118 County A Northern chaparral California
The Placer County watershed A is located in Placer County (Northern
California), in the foothills of the Sierra Nevada. Altitude there varies between 120 and 240 meters. Aspect is northern, soils are shallow but the parent material is highly weathered. Vegetation is a woodland chaparral, dominated by Quercus wislizenii and Rhus diversiloba (Lewis, 1968). 37
3.2 Limits to the analysis imposed by the data
Available data impose serious limitations to the present study
(1) Representativeness of the watersheds : all of the data used in this study come from experimental watersheds, monitored for diverse
research goals. Are they actually representative of the mediterranean-type
shrublands in their region?
(2) Intra-regional and inter-regional variability: it seems logical to
think that given their low number, the studied watersheds don't represent
the entire variability of response encountered in each kind of shrubland.
(3) Data available : this study attempted to put forward the link
between ecology and hydrology. The input variables considered in the
analysis were the climate, the drainage basins' morphometry, and the
vegetative cover (kept "constant" as a mediterranean-type shrubland).
Differences in geology and soil cover were not taken in account, due to the
lack of quantitative information. This is obviously a limitation, since it is
clear that the geology can be an important factor in hydrological response
(Hewlett, 1982; Seyhan, 1982).
(4) Limited sample available : to ensure the statistical validity of
this study, a larger sample of watersheds was desirable. This was not
possible, due to the difficulty of obtaining hydrological data from some
research Institutions, and to the lack of information concerning existing 38 gaged watersheds in the mediterranean-type shrubland zones.
In spite of these limitations, it seems that the approach adopted in this study is valid, and that it can be useful for a similar study using a larger watershed sample, or for other comparative hydrological studies. 39
CHAPTER FOUR:
TESTING THE FIRST HYPOTHESIS
4.1 Characterization of Watershed Behavior using a two-slope linear rainfall- runoff model.
4.1.1 Methodology used to separate baseflow using daily rainfall-runoff data
4.1.1.1 Use of Storm Data in hydrological analysis.
Stormflow is defined by Hewlett (1982) as the sum of surface and subsurface stormflow. It is "the term most often used by hydrologists in describing the flood producing characteristics of watersheds". Indeed, individual storm events and the hydrograph they produce are the main object studied by hydrologists and hydraulic engineers who are mainly concerned about flood control. However, stormflow data are also useful to describe the integrated hydrological response of watersheds. According to Chow (1964),
"the hydrograph can be regarded as an integral expression of the physiographic and climatic characteristics that govern the relations between rainfall and runoff of a particular drainage basin."
Stormflow data was used by Hawkins (1992) in an attempt to classify watershed behavior, comprising a broad setting of climate and vegetative conditions. In this study, his method was used to try to compare the
behavior of mediterranean shrubland watersheds with other vegetative 40 covers, and to test the relative homogeneity of the catchments by comparing them with the watersheds studied by Hawkins.
4.1.1.2 Generation of stormflow data.
Stormflow data used in most studies come from very detailed hydrographs and hyetographs. In this study, as these data were not available, it was decided to use the daily rainfall-runoff time series, available for all but two of the watersheds.
Any computation of stormflow implies the separation of stormflow
from baseflow. This separation is difficult, given that numerous methods
exist, all of them empirical (Chow, 1964; Hewlett, 1982). Hewlett (1982)
reminds that "no graphical or mathematical operation performed on a
hydro graph will reveal the source or pathway of stormflow." Traditional
separation methods are contested by some authors, who like Beven (1987),
consider them to be "a nonsense." In this study, the simple approach
described by Hewlett (1982) was used, considering the limited accuracy of
available storm hydrographs. Hewlett's method assumes a linear increase of
basef low contribution to total streamflow from the beginning of the
hydrograph to the end of the flood event (Figure 4.1). Other more
complicated methods were not realistically applicable to baseflow separation
using daily hydrographs. 41
I Flow i
Basef low Separation Slope
1 Time I
Figure 4.1 - Separation of Baseflow and Stormflow 42
A second problem concerning basef low separation is that there is no generally recognized method to compute the baseflow increase slope for a given watershed. This is evident, given the numerous watershed characteristics affecting baseflow slope, including soils properties and soil moisture conditions before the storm. However, there is a need to determine this slope for different watersheds and for different periods of record. Kim
(1989) presented an interesting approach to optimize a yearly baseflow slope for a watershed. He optimized the baseflow increase slope by trying successively different values, and computing for each value the baseflow
percentage. The "best slope" is then defined by a break occurring in the
slope versus baseflow curve. This approach was tested on a number of
basins monitored by the USGS, and a software was developed. Using this
software, a "best separation slope" was obtained for each hydrological year
of data.
The separation of stormflow for individual storm events was done
using the corresponding yearly slopes. This approach seemed to be more
adequate than prescribing a constant value found in the literature. However,
it was still not satisfactory when applied on an event basis. One major
problem found in this study was that the slope was sometimes
underestimated: some events could last for weeks after the rainfall occurred,
although it is known that subsurface stormflow has only a residence time on 43 the order of days (Hewlett, 1982). Therefore, a "mixed" approach was used involving rainfall data to minimize this inconvenience. The following procedure was used: the algorithm for stormflow separation is used when streamflow at day n +1 is greater than streamflow at day n. It starts recording the corresponding rainfall on day n. For each following day, the algorithm adds to the event values the rainfall of the day k, and the stormflow on day k:
Stormflow(k) = Streamflow(k) - [baseflow(k-1) + slope]
A storm event would end on day (k):
if streamflow(k + 1) < baseflow(k) + slope (Figure 4.2.a) or
after 3 dry days (no rain occurred) following the last wet day of the
event (Figure 4.2.13)
The 3 days limit after the end of the rain was chosen arbitrarily to
accommodate a longer time for subsurface stormflow to reach the channel,
while keeping the event length at reasonable limits.
Stormflow data sets were generated for the following mediterranean
shrubland watersheds: 44
Rain
Boseflow separation slope CLÉS the hydrogroph
, ,?!..%:•4: ,..W•11111111
days
jRain
4=7
Event ends 3 days after the last rain
days
Figure 4.2 - Baseflow Separation Using Daily Hydrographs 45
Table 4.1: Summary of the storm events analyzed in this chapter
Watershed Number of Events Hydrological Years used RC 5 211 1969- 1988 RC 6 208 1969- 1988 RC 7 197 1969 - 1988 RC 8 172 1969 - 1988 SDEF 81 62 1939-44, 1953-56 SDEF 82 75 1939-46, 1948-49 SDEF 83 81 1939-46, 1948-49 SDEF 84 . 79 1939-46, 1948-49 Note:
RC: Réal Collobrier (France). SDEF: San Dimas Experimental Forest (California, USA).
Note on South Africa's fynbos watersheds:
.Bakkerskloof: as only weekly rainfall data were available, no stormflow was generated.
.Langrivier: as the daily rainfall data available were not representative of the total precipitation on the watershed, no stormflow data were generated. Instead, a data set (312 events) from Hawkins (1992) was used.
This data set was used with reserve, since it contained only events caused by a rain > 20 mm, it was based on more detailed hydrographs, and no information was available on the method used for baseflow separation. 46
4.1.2 Computation of model parameters and reasons why Hawkins' method was not used for comparative analysis.
Hawkins (1992) developed an algorithm fitting a two straight lines model for the purpose of classifying different watershed behavior on the basis of stormflow response. The computer program optimized the following coefficients, presented in Figure 4.3
a: threshold rainfall depth
Il l : slope of stormflow response to rainfall P (when P < a)
B2: slope of stormflow response to rainfall P (when P a) or,
6.1 a + E2 (P - a) for P > a Stormflow = (4.1) fl i P for P < a
Storm data from eight mediterranean shrubland watersheds were analyzed using this method. All watersheds showed a very low 11 1 value.
However, SDEF watersheds showed that stormflow response slopes are extremely sensitive to extreme rainfall events. The results showed that, if there is not enough data available at the upper range of rainfall values, the one or two values with the highest rainfall will determine the watershed classification: adding two low-response storm events (occurring in 1954 and
1956) to the data for watersheds SDEF 82, 83 and 84 changed their 47
0 E t
i T . r Rainfall
Figure 4.3 - Two-Straight-Line Model for the Rainfall-Runoff Relationship 48 classification from violent to standard behavior (Hawkins, 1992). Therefore, an absence of extreme events can produce questionable results when using this method for comparative purposes.lt was preferred to decline the use of this method (and the comparison with the different watersheds used in
Hawkins' study, since the storm data available for the mediterranean shrubland watersheds was lacking of extreme events. It was however noticed that all but one (Langrivier) of the watersheds had a very small reaction to low rainfalls (a l was almost negligible). This is consistent with the observations made by Fogel and Duckstein (1971) regarding the hydrological response of semi-arid watersheds.
4.2 Establishment of regression-type relationships to describe the watersheds' dynamic hydrological behavior
4.2.1 Research of the best regression equations to describe mediterranean shrubland watersheds' hydrological behavior
Rather than characterizing hydrological behavior by mean values, it seemed more interesting and more useful to describe the interannual and
interseasonnal variability of responses by defining relationships between the
main climatic parameters and some key hydrological parameters. The
regression coefficients obtained were then included in a Principal
Components Analysis, allowing an analysis of the dynamic behavior of the 49 watersheds. This approach was used with the conviction that statistical analysis wouldn't yield easily interpretable results if the data were not prepared in a meaningful way. Moreover, defining hydrological behavior relationships prior to any analysis was a good way to test for the homogeneity of the sample, and identify outlier watersheds, which fitted poorly with a model that was adequate for the rest of the group.
Numerous potential relationships were investigated, using data that were derived from the initial sets. Available data covered stormflow, potential evapotranspiration (derived from temperature or pan data, depending on the availability) and rainfall-runoff. An important criteria in the choice of a descriptive relationship was either its potential use for hydrological design, or the contribution it could bring to a better understanding of the hydrological functioning of mediterranean-type shrublands.
4.2.2 Relationship between annual precipitation and streamflow values
(Model 1)
There is often a good linear relationship between the annual precipitation and flow at the outlet of a watershed (Linsley et al. 1975). This relationship has been used for a long time to study the variability of water yield. It is especially valuable to describe hydrological behavior in semi-arid 50 or more generally mediterranean climates, where the interannual variability of precipitations is high (Shachori and Michaeli, 1965; Naha!, 1981). Diskin
(1970), noticed that for arid and semi-arid watersheds, some years yield little or no runoff, and that the linear relationship between precipitation and stormflow only appears after a given precipitation threshold. Under wetter conditions, the same threshold can be defined, but is never reached (in the
available records at least), and the watershed yields water every year.
Failure to recognize the existence of two distinct domains can be misleading.
Piñol et al. (1991) attempted to define the coefficients of this relation for two woodland-maquis watersheds, but found no good fit. This seems due to the fact that they hadn't enough data points (only 7) to define the two parts
of the model (before and after the precipitation threshold), and that they
were comparing their watershed to a more humid one, where all the data
points were above the threshold; it seems that they were expecting the
same kind of response, and failed to recognize that their data record wasn't
long enough.
The form that was adopted for this model is:
(P - BO for P > B i = (4.2) 0 for P B,
(Figure 4.4)
51
where:
P : annual precipitation (mm) Q : annual streamflow (nrn) A l , B 1 : regression coefficients
This relationship is particularly valuable and instructive for watershed
analysis purposes:
.B 1 represents the minimum amount of precipitation necessary for
runoff to occur. It is related to vegetation characteristics, soil properties, and
to the temporal distribution of rainfall throughout the year (Diskin, 1970).
.A 1 also is related to soil and vegetation characteristics. (1 - A l ) is
particularly interesting for the analysis of dryland ecosystems, because it
represents the ability of the vegetation to adapt its water consumption to
the available moisture. As noted by Pitiol eta!. (1991), the actual
evapotranspiration (computed by the water balance method) varies little in
wetter ecosystems, but considerably in mediterranean woodlands. Equation
4.2 can be adapted to predict the actual evapotranspiration (Ea) of the
vegetation at a watershed scale, using the water balance method (Diskin,
1970):
P(1 - A l ) + A 1 13 1 for P > B 1
Ea = (4.3) for P 15. 13 1 52
a 0 = A1 (P — B1) for P > B1 0 = 0 for P < B1
Yearly Precipitation (P)
Figure 4.4 - Graphical Interpretation of Model 1 53
Model 1 parameters, fitted using a least squares method, are presented in Table 4.2.
Table 4.2 : Coefficient values and significance level for Model 1
Watershed Al Bi R2 Significance level
Langrivier 0.703 - 192 0.29 non significant Bakkerskloof 0.800 619 0.81 0.01
RC 5 0.651 467 0.79 0.01 RC 6 0.696 498 0.76 0.01 RC 7 0.762 435 0.73 0.01 RC 8 0.763 293 0.94 0.01 SDEF 81 0.562 637 0.90 0.01 SDEF 82 0.493 602 0.82 0.01 SDEF 83 0.425 584 0.83 0.01 SDEF 84 0.494 571 0.79 0.01
The significance level presented in Table 4.2 corresponds to the test of the hypothesis Ho : A l =0 against the alternative H 1 : Al < >0. Its value corresponds to the probability of having H o true. Relations were considered to be non-significant when the probability p in the F-test was found greater than 0.05.
All the watersheds, excepted Langrivier, shown a high R 2 , and the 54 relationships were always significant at the 0.01 level. The relationship was not significant for Langrivier (p > 0.05). Plots of the relationship for the ten watersheds are presented in Appendix C.
4.2.3 Relationship between precipitation and streamflow during the wet season (Model 2)
One of the determining characteristics of mediterranean-type climates is the existence of a summer drought and the occurrence of a rainy season from the end of Autumn to the beginning of Spring. It was interesting, in order to understand the hydrological processes involved and also to predict the water yield during both periods, to fit the same relationship as in Model
1, and observe the differences. The equation adopted for the model was:
A2 (Pw - B2) for Pw > B2 Qw = (4.4) 0 for Pw s B2
(Figure 4.5)
where:
Pw : precipitation during the wet period (mm) Qw : streamflow during the wet period (mm) A2, B2 : regression coefficients
For South Africa, the dry period corresponded to the months 55
November to March, and the wet period to the months April to October. For
France and California, the dry period corresponded to the months May to
September, the wet period to the months October to April. For both hemispheres, the duration of the "dry" period was the same (5 months), so that comparisons are possible.
The coefficients of this model may be interpreted similarly as for the precedent equation: A2 is related to the ability of the vegetation to adapt its water consumption (i.e. its growth) to the water available during the wet period. It was thought to be a good descriptor for mediterranean ecosystems, where growth is very seasonal and mainly concentrated in the spring (Specht, 1969). Computed coefficients are presented in Table 4.3. 56
Wet Season Precipitation (Pw)
Figure 4.5 - Graphical Interpretation of Model 2 57
Table 4.3 - Coefficient values and significance level for Model 2
Watershed A2 B2 R2 Significance level
Langrivier 1.043 320 0.52 0.05 Bakkerskloof 0.884 497 0.91 0.01 RC 5 0.735 373 0.83 0.01 RC 6 0.845 429 0.87 0.01 RC 7 0.834 355 0.79 0.01 RC 8 0.830 205 0.92 0.01 SDEF 81 0.525 608 0.87 0.01 SDEF 82 0.404 536 0.83 0.01 SDEF 83 0.363 519 0.86 0.01 SDEF 84 0.414 506 0.81 0.01
Again, all the watersheds, Langrivier excepted, showed a very satisfactory fit (high R 2 , relationship significant at the 0.01 level). The behavior of Langrivier's data was better, and the relationship was significant at the 0.05 level. However, Langrivier shows a coefficient A2 with a value slightly greater than 1. Two explanations can be proposed:
.this can be due to the uncertainty concerning actual rainfall on the watershed, because the seasonal precipitation amount was computed adapting an equation that was initially derived for yearly totals only.
.this A2 value can also be considered to be equal to one. This would then signifies that during the wet season, under the superhumid 58 mediterranean climate of Langrivier, all "additional" water is transformed into runoff (during the wet season, water is not a limiting factor for plant growth).
4.2.4 Relationship between precipitation and streamflow during the dry season (Model 3)
When plotting the values of the dry season's precipitation and runoff against each other, it was observed that the relationship was probably not
linear (see Appendix C). In analogy with Hawkins' model (equation 4.1), it
was decided to fit the data with a two-line model. The model used was of the form (Figure 4.6):
A3.1*B3 + A3.2 (Pd - B3) for Pd > B3
(4.4) Qd = A3.1 * Pd for Pd 15_ B3
where
Pd : precipitation during the dry period (mm) Qd : streamflow during the dry period (mm) A3, B3 : regression coefficients
This change in data behavior is a little surprising. A behavior similar to
the one observed for the Californian watersheds in Model 1 and 2 would
have been expected, (i.e. a threshold under which there is no or little 59 response), since during the dry season characteristic of mediterranean climates, vegetation suffers of a water-stress, and water is obviously limiting. A possible explanation lies in the nature of the summer storms under mediterranean climate: the rare precipitations are caused by very high intensity convective storms, that quickly exceed the infiltration capacity of sometimes poorly wettable soils, and create runoff. The second part of the model (Pd > B3 ) can be explained similarly as the regression lines in Models
1 and 2. The coefficients computed for Model 3 are presented in Table 4.4.
Table 4.4 : Coefficient values and significance level for Model 3
Watershed A3.1 A3,2 B3 R2 Significance (mm) level
Langrivier 0 0.537 320 0.81 0.05 Bakkerskloof 0.139 0.325 497 0.65 0.01
RC 5 0.176 0.298 373 0.67 0.01 RC 6 0.119 0.286 429 0.54 0.01
RC 7 0.324 0.345 355 0.53 0.01 RC 8 0.194 0.466 205 0.75 0.01 SDEF 81 0.0 * * 0.31 *
SDEF 82 0.0 * * 0.38 *
SDEF 83 0.0 * * 0.41 * SDEF 84 0.0 * * 0.31 *
* : None of the regressions that were tried for the SDEF watersheds were 60
for Pd < 83 Gd = A3.1 x Pd for Pd > 133 Gd = A3.1x133 + A3.2(Pd — 83)
%."
i
I B3
Dry Season Precipitation (Pd)
Figure 4.6 - Graphical Interpretation of Model 3 61 significant at the 0.05 level. Therefore, the alternative hypothesis (A 31 =
. the available A3 . 2 -= 0) was accepted. B3 and A3 2 remained undefined with records.
4.2.5 Watershed characterization using stormflow data
Since stormflow response is an important characteristic of watershed behavior, the two-slope rainfall-runoff modelling approach discussed in 4.1 was modified in an attempt to make it less sensitive to the effect of outliers
(extreme values). Ranking rainfall and stormflow values is an interesting alternative, that takes into account the stochastic aspect of stormflow response: when matching rainfall and stormflow of same rank, it is assumed that the return periods of rainfall and stormflow events correspond. This is obviously a simplifying assumption (and the actual rainfall-stormflow pairs show it), since stormflow response is strongly influenced by factors such as antecedent soil moisture and rainfall intensity. However, all attempts to describe realistically stormflow response using the data available, and considering the moisture status at the watershed scale were unsuccessful.
The greatest advantage of ranking rainfall-runoff values is certainly that it minimizes the influence of extreme values on the final fitting result. The model used for stormflow modeling with ranked rainfall and runoff is: 62
A4.1 *B4 4- A4.2 * (PS - B4 ) for Ps > B4 Qs= (4.5) A4.1 *Ps for Ps :s B4
(Figure 4.7) where
Ps : storm precipitation (mm) Qs : storm runoff (mm)
Table 4.5 shows a comparison of the values obtained with the ranked pairs and the original pairs. The regression lines are presented in Appendix C.
Table 4.5 : Comparison of Model 4's coefficients obtained when using original pairs and matched (ranked) pairs.
Original Values Ranked Values
Watershed A4 B41 B42 A4 B41 B4 2 (MM) (mm)
RC 5 30.0 0.035 0.234 79.0 0.053 0.613
RC 6 116.4 0.095 1.000 107.7 0.080 1.000
RC 7 34.1 0.036 0.301 70.7 0.066 0.741
RC 8 22.1 0.054 0.428 41.6 0.078 0.640 SDEF 81 55.7 0.007 0.106 173.3 0.007 0.996 SDEF 82 82.1 0.017 0.815 82.1 0.007 0.841 SDEF 83 107.7 0.033 1.000 103.5 0.015 1.000 SDEF 84 106.1 0.040 1.000 99.7 0.013 1.000 LANGRIV. 29.8 0.182 0.591 29.2 0.110 0.659 63
/ For Ps < B4 Qs n A4.1 x Ps For Ps > B4 Qs n A4.1x84 + A4.2(Ps - B4)
A4 . 1 I I
B4
Precipitation (Ps)
Figure 4.7 - Graphical Interpretation of Model 4 64
Notice that ranking corrects the effects of extreme events on the
SDEF 81 watershed, and gives homogeneous results for the group of the
SDEF watersheds. These four watersheds all fall in the violent category
(Hawkins, 1992), while when using unranked values, SDEF 82, 83 and 84 felt in the violent category, but SDEF 81 in the standard category. The result obtained through ranking is more logical, because the four SDEF watersheds are contiguous and apparently similar to each other, with SDEF
81 only differing from the others in the hydrological years that had been used for storm events extraction (see Table 4.1). Note that all the mediterranean shrubland watersheds considered here fall in Hawkins' violent category.
The matched (ranked) pairs define obviously an "averaged" response.
This average response coefficients can be valuable to those considering to use the simple two-slope linear rainfall-runoff model for hydraulic design purposes. This relation can be used as an indication of average stormflow
response, and the designer can apply a safety coefficient according to the
potential hazard of structure failure. 65
Conclusion on the possibilities of characterization of watershed behavior using storm data:
Storm data are precious to describe a watershed's hydrological behavior, by their capacity to integrate all the watershed conditions in a stormflow hydrograph. However, this integration capacity makes the hydrograph prediction extremely complex, and requires for modeling purposes a lush of detailed observations that are not usually available. Using ranked rainfall-stormflow pairs rather than the original values, we estimate a
"mean" response to "average" watershed conditions. This information is valuable for comparative hydrology studies, and can be useful to engineers in the design of small hydraulic structures, when no other information is available. Therefore, the two-slopes linear model fitted with ranked rainfall
low values, was considered to be a valuable descriptor of the -stormf mediterranean shrubland watersheds' behavior, and its parameters were included in the principal components and multiple regression analysis.
4.2.6 Conclusion relative to the 4 relationships chosen to describe mediterranean shrubland watersheds' hydrological behavior
The four relationships selected to describe watershed behavior fit well the data and are significant for all but one of the watersheds. South Africa's
Langrivier watershed doesn't fit to Model 1, and fits with difficulty to Model 66
2. This may be due to the poor quality of the available rainfall data, which were extrapolated using the regression equation given by van der Zel and
Kruger (1975). It also may be due to the superhumid climate of Langrivier
(with 2261 mm of mean annual precipitation, Langrivier is a very extreme mediterranean climate), its complicated geological features, or the high altitudinal rainfall gradient (van Wyk, 1987). Langrivier would be obviously interesting to include in the data analysis, because it represents the wet end of the mediterranean climate, but given the numerous uncertainties, it was chosen not to be included in the principal component and stepwise regression analyses.
4.3 Test of results' homogeneity for Models 1 through 4
As noticed in the precedent sections, the four models' parameters depend on the characteristics of soils, climate, watershed morphometry and vegetation. The fact that all the 9 mediterranean shrubland watersheds fit the same models was an indication of a similar behavior. Furthermore, all of the nine watersheds fall in the same class of behavior ('violent' behavior) for
Model 4 (Hawkins, 1992). However, before going further, it was important
to test if the values found were significantly different from each other; i.e. if
it was not possible to represent very simply all the computed values by their
mean (or median). For that purpose, the mean and median of the four model 67 parameters were computed. It was important to check the nature of each parameter's distribution, to make sure that the mean is a good descriptor of the global trend. A test for normality (Lilliefors test) was applied. The hypothesis of normal distribution was accepted at the 0.05 level of
confidence for all the parameters excepted A4 . 2. Therefore, it was decided to use the mean value of the parameters in the following test for all the parameters excepted for A 4 2. For this parameter, the median value of the parameters was used.
The following hypothesis was tested:
Ho: the model parameters' values are not significantly different
from their mean against
H 1 : the model parameters' values are different from the mean.
Testing the values of Model 1 parameters:
To test the difference of the two regression parameters with their means, it was important to test them together, rather than alone. A simultaneous F-test was performed, and H o was rejected at the 0.05 level for 7 of the 9 watersheds (H o wasn't rejected for RC 5 and RC 6). 68
Testina the values of Model 2 parameters:
Ho was rejected at the 0.05 level for all of the 9 watersheds.
Testina the values of Model 3 parameters:
To test model 3 and 4 was difficult, because they were not simple linear regressions, but two-slope regressions. To implement the F-test,
Models 3 and 4 were decomposed into two simple linear regressions, and a separated test was performed for each slope. For A31, the nine watersheds were tested for difference with the mean value of the coefficient. But for
A3 . 2, the SDEF watersheds couldn't be tested, as this parameter was not defined for them. Therefore, the mean of the parameter was only based on five watersheds. For A31, Ho was only rejected for RC 7. For A 3 2 , H o was only rejected for RC 8.
Testing the values of Model 4 parameters:
A 4 2, Ho For A 4 1 , Ho was rejected for all of the nine watersheds. For wasn't rejected for the watersheds SDEF 82, 83, and 84.
Conclusions relative to the homogeneity of the Model's parameters'
values.
The tests implemented show us that for three out of the four models,
parameters values were significantly different from their mean. The
watersheds seem to have a very similar hydrological behavior during the dry 69 season (Model 3).
These results suggest that mediterranean shrubland watersheds have a different hydrological behavior. Hydrological behavior is not only a function of the soil-vegetation-climate complex, it also depends on the morphometry of the watersheds, and on some minor climatic characteristics which can vary inside the mediterranean-type shrubland group. The preceding tests show that there may be a greater variability induced by morphometry and climate inside the group, than similarity due to the common vegetation-type.
Only a comparison with a different ecological group could allow to show the real effect of the vegetation.
In an attempt to further analyze the effects of morphometry and minor climatic variations on hydrological behavior, and help identify the actual sources of variability, a multiple regression analysis was implemented.
4.4 Selection of relevant and non-correlated identifiers by Principal
Components Analysis (PCA)
Seyhan (1982) used PCA to identify relevant parameters in the development of runoff equations. He states that this method "is not merely an interesting research entity, but can become a valuable tool in solving multivariate problems."
Identifiers corresponding to the output of the hydrological processes 70 were selected and presented in section 4.2. Identifiers corresponding to the input, i.e. climatic and morphometric variables, were selected from the literature (references are presented in Table 4.7, 4.8 and 4.9). The purpose of principal component analysis was to select the best explanatory variables among the input variables and to determine the degree of correlation among variables. This is a necessary step in the analysis, since strongly correlated variables give usually bad (unstable) results in a stepwise regression analysis. The PCA also attempted to show what were the best descriptors of the watersheds' hydrological behavior among the four models' parameters.
4.4.1 Selection of identifiers
Following Seyhan (1982), the variables to be used in the multivariate analysis were grouped as input variables (climatic and morphometric variables), and output variables (hydrological behavior variables). The following Tables present the hydrological input variables that seemed pertinent to this study, and recapitulate the output variables (model parameters) selected to describe watersheds' hydrological response. 71
INPUT VARIABLES:
Table 4.6 - Climatic variables selected to analyze mediterranean shrubland watersheds' behavior.
CLIMATIC VARIABLES
IDENTIFIER DESCRIPTION REFERENCE 1. P Mean Annual Precipitation Seyhan (cm) (1982)
2. ET° Mean Annual Potential Evapotranspiration
3. Qp, Emberger's Pluviothermic Quotient : Nahal Qpi = 20*P/(M 2-m 2 ) (1981) (cm °C-1 )
4. P' Mean dry months precipitation / Mean Seyhan annual precipitation (1982) (dimensionless)
5. 138 UNESCO ratio of aridity : Re = P / ET° UNESCO (dimensionless) (1977)
Note:
M : Mean Maximum Temperature of the Warmest Month ( ° C) m : Mean Minimum Temperature of the Coldest Month (°C)
Note: Emberger's pluviothermic quotient was divided by one hundred to avoid the problem of large values in regression equations. The original Q pi coefficient has a coefficient 2000 in its formula. 72
Table 4.7 - Morphometric variables selected to analyze mediterranean shrubland watersheds' behavior.
1 MORPHOMETRIC VARIABLES IDENTIFIER DESCRIPTION REFERENCE 6. k Lemniscate coefficient Rice (1969), k = L211 /7/4A Zavoianu (dimensionless) (1985)
7. SB Average watershed slope Rice (1969),
SB 1= (Hmax - Hmin)/L Seyhan (dimensionless) (1982)
8. S. Mean slope of the main channel Rice (1969), (dimensionless) Seyhan (1982)
9. A Area of the watershed Rice (1969), (km 2 ) Seyhan (1982), Zavoianu (1985)
10. D Drainage density Seyhan D = (Total length of channel)/A (1982), (km-1 ) Zavoianu (1985)
Note:
L : major axis of the watershed Hmax : maximal altitude on the watershed Hmin : minimal altitude on the watershed 73
OUTPUT VARIABLES:
Table 4.8 - Dynamic and static hydrological output variables selected to describe mediterranean shrubland watersheds' behavior
HYDROLOGICAL BEHAVIOR VARIABLES
IDENTIFIER DESCRIPTION REFERENCE
11. Fy Mean annual streamflow fraction Kim (1989) occurring as baseflow (dimensionless) 12. F' Mean fraction of the yearly streamflow occurring during the dry season (dimensionless) F' = Qd / Q 13. y Water yield y = Q/P (dimensionless)
14-15. A l , Model 1 coefficients Diskin 13 1 (A1 dimensionless, B, in mm) (1970), Fogel and Duckstein (1971)
16-17. A2, Model 2 coefficients B2 (A2 dimensionless, B2 in mm)
18-20. A3.1 , Model 3 coefficients A3 .2, B3 (A3.1 and A3 .2 dimensionless, B3 in mm)
21-23 A4.1, Model 4 coefficients Hawkins (A4.1 and A4. 2 dimensionless, B4 in mm) (1992) A4.2 , B4
Note :
Q : mean yearly Streamflow (mm) Pd, Pw : Precipitation during the dry and wet periods, respectively (mm) Qd, Qw : Streamflow during the dry and wet periods, respectively (mm) Ps : Storm Precipitation (mm) Qs : Stormflow (mm) 74
4.4.2 Discussion of the PCA results
Principal Component Analysis is a widely used data analysis method; it allows to extract the major features of a data set, assessing which are the variables that contain most of the variability in the set. PCA defines principal components, which are not correlated with each other, and that contain the larger part of the total variability; these components are the best descriptive axes in the space of variables. PCA also puts forward the correlations between variables. Nine watersheds were included in the PCA (Langrivier was set aside). The variables were kept in two groups (input and output variables) for analysis. The conclusions that were drawn from this analysis are presented thereafter.
Input variables: PCA showed that 97% of the variability could be represented by five components.
(1) the first component was strongly correlated with four climatic
variables: Ra, 00, P and P'; it explained 39% of the total variability.
The correlation coefficients were respectively 0.98, 0.98, 0.90 and
0.73.
(2) the second component was strongly correlated with D (correlation
coefficient: 0.97); it explained 15% of the total variability.
(3) the third component was strongly correlated with So (correlation 75
coefficient: 0.96); it explained 15% of the total variability.
(4) the fourth component was strongly correlated with A (correlation
coefficient: 0.92); it explained 16% of the total variability.
(5) the fifth component was strongly correlated with K (correlation
coefficient: 0.96); it explained 12% of the total variability.
The four component which were strongly correlated with morphometric variables explained 58.8% of the total variability, while the component strongly correlated with the climatic variables explained 38.7% of the variability. The correlation matrix showed that in the climatic variables group, Q0, P and Ra were strongly correlated. It was decided to let Ra out of the stepwise regression, since its computation requires knowledge of potential evapotranspiration and it is more difficult to estimate than the two other variables.
Output variables: PCA showed that 95% of the total variability could be represented by five components:
(1) the first component was strongly correlated with A l , A2, A4 . 1 and
Y. It represented 34.4% of the total variability.
(2) the second component was strongly correlated with F y only; it
represented 18.4% of the total variability.
(3) the third component was strongly correlated with A3 . 2; it 76
represented 21.7% of the total variability.
(4) the fourth component was strongly correlated with A4 . 2; it
represented 13.4% of the total variability.
(5) the fifth component was strongly correlated with B i ; it represented
17.6% of the total variability.
PCA results show that the dynamic descriptors (parameters of the four models) encompassed much of the variability encountered among the watersheds. Only one of the static descriptors (Fy ) was strongly correlated with a principal component. Another interesting information brought about by the correlation matrix was that A l and A2 were strongly correlated: this seems logic, since in a mediterranean climate, most of the annual precipitation and the runoff occur during the wet season. Therefore, Model 1 and 2 seem to be equivalent.
4.4.3 Stepwise multiple regression analysis
The purpose here was to establish predictive equations for the
Hydrological Output variables, using the relevant descriptors identified by
Principal Component Analysis, and to assess if it was possible to form statistically significant equations correlating the hydrological behavior
identifiers of the mediterranean shrublands with the climatic and
morphometric variables. 77
The following explanatory variables were used: Qpi , P, P', K, S o , D, A
(see tables 4.7 and 4.8 for the description of each variable). Multiple regressions were developed for the following Hydrological Output variables:
A l , 13 1 , A2, B2, A3 . 1, B4, A4 . 1, A4 . 2, and F. The reason why no equations were
developed for A3 . 2 and B3 was the small number of observations available
(only 5, since these two coefficients values were undefined for the SDEF watersheds) which wouldn't allow a meaningful result.
Backward and forward stepwise regression analysis were used. In most of the cases, they both gave the same result. When different equations were produced, the solution with the highest adjusted R 2 was chosen. When
R2 's were very close to each other, the simplest solution was chosen, because it was thought that, given the limited watershed sample used, simple solutions were probably the most reliable and most meaningful.
Significant regressions were produced for all the selected output variables, excepted F, for which the regression was rejected at the 0.05 level. Among the two other static identifiers (F' and Y), only Y yielded a significant relationship. This is very interesting, since it seems to justify our initial approach consisting in preferring dynamic identifiers to static ones:
PCA showed that dynamic identifiers were preferable to static ones. They describe better the total variance in the sample. Stepwise regression analysis showed that static identifiers couldn't be significantly explained by 78 morphometric and climatic explanatory variables. Regressions equations obtained for the dynamic descriptors are presented in Table 4.10 below.
They are all significant at the 0.01 level.
Table 4.9 - Multiple regression equations obtained for the hydrological behavior variables of the mediterranean shrubland watersheds.
Regression equation Adjusted
R 2
A l = 0.424 - 0.015A + 1.857P' 0.87
B 1 = 545.4 - 97.2D + 57.7K + 429.8S 0 0.93
A2 = 0.332 + 2.389P' 0.91
B2 = 437.8- 97.3D + 73.1K + 431.76 0 0.91
A3.1 = 0.079 - 0.020A - 0.427P + 3.306P' + 0.95
0.3326 0
A4 . 1 = 0.018 + 0.291P' - 0.0766 0 0.95
A4.2 = 0.759 - 0.094D + 0.154K 0.80
B4 = 5.943 + 43.16K 0.83 • 79
The explanatory variables involved in each regression equation are interesting to analyze:
Model 1: .A 1 is negatively correlated with A, which seems to indicate that smaller watersheds are better transformers of water than larger ones: this can probably result from transmission losses, which occur in the dry stream beds. Transmission losses can play an important role in semi-arid environments. The positive correlation with P' can probably be explained by the violence of summer storms in mediterranean climates: these storms produce more runoff during the year.
.B 1 is negatively correlated with D and positively with K and S o .
The negative correlation with D seems logical, since the drainage density express a watershed's ability to release water; the precipitation threshold necessary to create runoff can be lower when D is high. The positive correlation with K indicates that in elongated watersheds, a greater part of the precipitation will be lost, because a longer distance to travel with
increased possibility for transmission losses. The positive correlation with S o
is surprising, and it could be an artefact.
Model 2: .A 2 is positively correlated with P', which is the only
explanatory variable selected from stepwise regression analysis. This is
interesting, since (1 - P') represents the ratio of the wet period precipitation 80 to the annual precipitation. A2 is negatively correlated to (1 - P'), which indicates that when the rainfall is concentrated during the wet season, the efficiency of water conversion is lower (probably because of lower intensity storms).
.B2 is a function of the same three morphometric variables selected to describe B 1 , and the correlations can be explained the same way as for 13 1 .
Model 3: Due to the insufficient number of watersheds analyzed for B3 and
. 1. A3 . 2, a regression equation was developed for only A3
with .A3 . 1 is positively correlated with P' and So, negatively correlated
P and A. The negative correlation with A indicates that smaller watersheds are likely to have a very low response below the threshold B3. The positive correlation with So can be explained by the fact that a high channel slope allows a better "evacuation" of runoff to the outlet, and offers less opportunity for transmission losses. The positive correlation with P' seems
logical, since it indicates that more precipitation during the dry period will
increase the response slope due to the violent response following summer thunderstorms. The negative correlation with P is more difficult to explain. It
could probably be explained by a more violent response of small watersheds
in dryland environments. 81
Model 4: .A4,1 is a function of P' and So . This shows that the watershed response to relatively small storms (smaller than B 4 ) is influenced by the amount of summer precipitation and the channel slope, as explained previously. The positive correlation with P' is logical if we consider that summer storms are rare but of high intensity in mediterranean climates. In the wet season, low-intensity small storms yield no runoff, probably because of infiltration on hillslopes and transmission losses in the channels beds. The negative correlation with S o is difficult to explain.
.A4 2 is only a function of morphometric coefficients (D and K).
The negative correlation with D could indicate that watersheds with a dense drainage network react less violently to high rainfall, which seems to be contradictory. This problem may have its cause in the way drainage density is computed: D was measured on topographic maps. However, different countries don't map the same way stream networks, especially when they are not perennial. For this reason, we cannot have much confidence in this variable. The positive correlation with K means that elongated watersheds trend to react less violently, probably because of a longer travelling time for the runoff water, and therefore more opportunities for transmission losses.
.13 4 appears to be a function of basin morphometry only. The
positive correlation with K indicates that the more violent response appears
latter for elongated watersheds. Again, transmission losses can probably 82 explain this observation.
4.4.4 Validation of Model 1 using an additional watershed
The predictive power of the regression equations shown in Table 4.10 need to be verified using additional mediterranean shrubland watersheds.
However, given the low number of watersheds available, we could not use them for validation. Therefore, an attempt was made to use information found in the literature, which had been found to be too incomplete for the initial analysis. Placer County Watershed A in Northern California was used for this purpose. However, the information provided by Lewis (1968) only allowed the computation of the coefficients for Model 1.
Table 4.10 - Computed and estimated values of Model l's coefficient for the Placer County Watershed A
Parameter Computed Value Estimated Value
A l 0.765 0.551
B 1 468.2 405.5
To estimate the predictive power of Model 1 with the estimated coefficients, the Nash and Sutcliffe coefficient (Nash and Sutcliffe, 1970) was computed and found equal to 0.86. This is a good result, but further testing is needed for fully evaluating the predictive power of the model. 83
4.4.5 Conclusions relative to Hypothesis 1 .
The values of 3 of the descriptive models' parameters were found to be different; and it was shown that an important part of these differences could be explained by the morphologic characteristics of the watersheds.
Among the climatic variables, the one representing the relative wetness of the dry season was the most frequently encountered in regression equations.
For this reason, it appears that we must reject Hypothesis 1 at the present stage. If there are similarities among the hydrological processes in all the mediterranean shrublands, the differences induced by the morphometry and the characteristics of the dry season in the different mediterranean-type climates are larger in amplitude.
However, this does not mean that we didn't reach our goal to use the ecology as a guide for hydrological regionalization. Significant regression
equations were developed for the four descriptive models' parameters. This
shows that there are probably common controlling factors on the small
mediterranean shrubland watersheds studied. This is an important
information for the understanding of the hydrological functionning of these
watersheds.
The description of the study watersheds, and the simple models setup
to describe hydrological behavior could be used in further comparative 84 hydrological studies, that would focus on different ecosystems. Such studies would generate more evidence to test if the controlling factors identified in this study are characteristic of small mediterranean shrubland watersheds. 85
CHAPTER FIVE:
TESTING THE SECOND HYPOTHESIS
The four simple hydrological models selected in part 4.2 were used to compare the hydrological behavior of Arizona and California chaparral. Data was obtained from the Three Bar experimental watersheds in Arizona, and the San Dimas Experimental Forest watersheds in California.
5.1 Assessment of botanical similarities between interior and coastal
chaparral
There is a major climatic difference between the climatic environments
of interior and coastal chaparral. In Arizona, where interior chaparral is
found, the precipitation pattern is distinctly bimodal, and significant amounts
of rain fall during the summer monsoon. In California, the climate istypically
mediterranean. However, the general aspect and the plant families of the
two chaparrals are so similar, that they have been given the same name.
There are numerous botanical similarities between them. Pase and Brown
(1982) state that "many of the species are disjunctive counterparts of those
in Californian chaparral (...), a number of other are undoubtly conspecific."
The watershed Three Bar D in Arizona is dominated by shrub live oak
(Quercus turbinella), and the SDEF watersheds is dominated by chamise 86
(Adenostoma fasciculatum).
5.2 Comparison of the hydrological behavior of chaparral in Arizona and
California
For this comparison, we first checked if the data followed the same relationships that we defined earlier for the mediterranean-type shrublands.
Significant regression relationships were found for Models 1, 2 and 4. There was no significant linear relationship between precipitation and streamflow during the "dry" season (the definition of the dry season was kept the same for coastal and interior chaparral, "warm" season would be more adequate, since interior chaparral is characterized by a summer rainy season).
However, there was no significant linear relationship either in model 3 for the SDEF watersheds. A R 2 was computed to check the fit of the data with a null value of runoff during the "dry" season. A R 2 of 0.68, which is higher than the R 2 obtained for the SDEF watersheds, was obtained. Therefore, it was decided to accept model 3 for the Three Bar D watershed and consider
A3 . 1 = 0 (B3 and A 3.2 undefined).
The four model parameters were computed. A simultaneous F-test was done to assess if TBD's parameters were significantly different from the mean of the four SDEF watersheds. Parameters values are given in table
5.1. 87
Table 5.1: Parameters values of the models 1 to 4 for the interior and coastal chaparral watersheds. •
Parameter Value computed for Mean of the TBD computed values for the SDEF watersheds
A 1 0.365 0.494
B i 521.4 598.5
A 2 0.426 0.427
B2 547.9 542.3
A31 0 0
A32 undefined undefined
B 3 undefined undefined
A 4 1 0.006 0.011
A42 0.092 0.996 (median)
B4 50.1 114.6
Test results showed that except for Model 3 (where the parameters
had the same values), the TBD watershed was always significantly different
from the SDEF watersheds. However, in order to assess the potential effect
of morphometric variables in this differences, the regression equations
presented in Table 4.10 were used to predict TBD's parameters values. This
values are presented in table 5.2. 88
Table 5.2 - Predicted parameters values for the interior and coastal chaparral watersheds.
Parameter Predicted Ratio Mean predicted value for [(y-ye)/y]2 value for the SDEF TBD watersheds
A I -0.157 2.05 0.346
B, 412.4 0.04 602.6
A2 1.072 2.30 0.428
B 2 379.8 0.09 532.9
A31 0.875 co 0.000
A4 . 1 0.082 160.4 0.010
A4 . 2 0.092 9.49 0.910 (Median)
B 4 215.3 10.87 111.2
The comparison of predicted versus computed values shows, a very good agreement for the SDEF watersheds. This is not surprising, since these watersheds have been used in the calibration of the equations.
To appreciate the fit of the TBD watershed to the equations, the ratio
HY - Yestimated)/Y/2 was computed, y being successively equal to all the parameters. This ratio was found to be greater than two in 75% percent of the cases. It is particularly interesting to notice the poor fit observed for A 31 : even if the computed values were identical, the use of predictive equations shows that the mechanisms of flow during the dry season are different in interior and coastal chaparral. This was an expected result, given that the 89 main difference between both shrublands is the amount of precipitation falling during the summer.
5.3 Conclusions relative to Hypothesis 2
It seems obvious that Hypothesis 2 has to be rejected, given the poor fit of the TBD watershed with the equations describing mediterranean-type shrublands' hydrological behavior. However, a certain reserve is needed, for the following reasons:
(1) we have compared in the precedent section a group of watersheds
(SDEF) that had been utilized in the calibration of the regression equations, with a watershed (TBD) that was not. This improves automatically the performance of the SDEF watersheds. It would have been preferable to use for this comparison other coastal chaparral watersheds, that were not part of the calibration group.
(2) only one interior chaparral watershed was used in this comparison, with no real guaranty of its representativeness.
Therefore, it would be interesting to test this hypothesis again, using
a larger number of interior chaparral watersheds, and additional coastal
chaparral watersheds to validate the equations' predictions. 90
CHAPTER SIX:
SUMMARY, CONCLUSIONS AND RECOMMENDATIONS
6.1 Summary of the procedures used and conclusions on the two hypotheses tested
The major objective of this study was to investigate the linkage between watershed hydrology and ecology and to verify whether it was possible to combine these two scientific domains into a new concept that we could call "hydro-eco-system." Two hypotheses were formulated to help analyze this linkage. The first hypothesis stated that all mediterranean-type shrublands have a common hydrological behavior. The second hypothesis stated that a mediterranean and a non-mediterranean chaparral shrubland can have the same hydrological behavior.
The first hypothesis was rejected, because it appeared that the
hydrological descriptors had significantly different values. However, the
differences in model parameter values were essentially due to the
watersheds' morphometric characteristics. It seems that ecological
information can still be used as a basis for regionalization, since meaningful
explanatory regressions were obtained for the models' parameters.
The second hypothesis was rejected, but with reservations because
only few watersheds had been used. It appeared that the explanatory 91 variables, chosen by stepwise regression analysis, could not explain the parameter values of the non-mediterranean chaparral watershed.
6.2 Conclusions relative to the selected identifiers
One of the important contributions of this study is the assessment that mediterranean shrubland watersheds can be described by four simple rainfall-runoff relationships, which allow an analysis of watersheds' dynamic hydrological behavior. These four relations were:
(1) relationship between annual precipitation and annual streamflow.
(2) relationship between precipitation and streamflow occurring during
the wet (and cool) season.
(3) relationship between precipitation and streamflow occurring during
the dry (and warm) season.
(4) relationship between rainfall and stormf low for individual storm
events.
The definition of relationships (2) and (3) was linked to the characteristics of the mediterranean climate, whereas the warm season is very dry, and precipitations only occur during the cooler part of the year.
When using these relationships for other comparative studies, it would be
more appropriate to rename these seasons "warm" and "cool" seasons,
however their definition could remain the same. 92
This study proved that it is possible to use dynamic descriptors to analyze a particular vegetation type's hydrology. It offers an opportunity of comparison with different vegetation type.
6.3 Limits of the study
The limiting aspects of this study were:
(1) a small number of mediterranean shrubland watersheds, that did
not allow the formation of two distinct groups; one for the calibration
of the regression equations and one other for validation of the
equations.
(2) all mediterranean-type shrublands were not represented in the
study.
(3) an insufficient number of non-mediterranean chaparral shrubland
watersheds.
6.4 Recommendations
Further work is needed in various domains:
(1) To overcome the limitations of the present study, data should be
obtained from mediterranean-type shrubland watersheds in Chile and
Australia, and more watersheds should be added from other regions.
For example, more watersheds from the San Dimas Experimental 93
Forest in California could be used, both in the calibration and
validation phases. A more complete watershed sample and an
estimate of the predictive value of models 1 to 4 should allow their
use for regionalization studies and rural engineering purposes.
(2) The simple models available for description of hydrological
behavior should be improved and increased in number. New models, in
particular, should try to take into account variable source areas and
not only watershed-scale rainfall and runoff processes. The concept of
variable source area has been proposed for forested watersheds
(Hewlett, 1982); and it could probably apply to deeply-rooted
shrubland vegetation systems.
The approach adopted in this study seems to offer interesting opportunities to implement further comparative hydrology studies. Two major application fields could be:
(a) Other vegetation types: a similar methodology could be used to
analyze the hydrological behavior of other ecosystems and compare
them to the mediterranean-type shrublands. Such a comparison should
allow an improved understanding of the hydrological processes
involved in each vegetation type, and it would help us tell if the 94 hydrological behavior that has been described for mediterranean-type shrublands is unique.
(b) Watershed Management: our approach also offers a good opportunity to study the effects of management practices such as periodic burning or conversion of brushland to grassland on watershed hydrology. The four simple models presented seem appropriate to put in evidence effects of management on water yield and watersheds' hydrological behavior. The methods of Comparative Hydrology can be extremely valuable tools in Watershed Management. 95
APPENDIX A:
Addresses of the institutions that provided the data
used in this study. 96
Réal Collobrier Research Watersheds:
The watershed research program is administered by
Dr Jacques Lavabre
Centre National du Machinisme Agricole, du Génie Rural, des Eaux et des Forêts (CEMAGREF) Groupement d'Aix-en-Provence Division Ouvrages Hydrauliques et Equipements pour l'Irrigation B.P. 31 Le Tholonet 13612 Aix-en-Provence Cedex 1 FRANCE
Telephone : (33) 42 66 93 10 Facsimilé : (33) 42 66 88 65
Jonkershoek and Zachariashoek Experimental Catchments:
Data were provided by
David Scott
Jonkershoek Forestry Research Centre Private Bag X5011 Stellenbosch 7600 SOUTH AFRICA
Telephone : (27) 2231 99 771 Facsimilé : (27) 2231 98 394 97
San Dimas Experimental Forest Watersheds:
Data were provided by
Dr Philip Riggan
USDA Forest Service Forest Fire Laboratory 4955 Canyon Crest drive Riverside CA 92507 U.S.A.
Telephone : (1) 714 276 65 23
Three Bar Experimental Watersheds
Data were provided by
Dr Malchus Baker
USDA Forest Service Rocky Mountain Forest and Range Experiment Station Arizona State University Tempe AZ 85287-1304 U.S.A.
Telephone : (1) 602 379 43 65 98
L'Avic and La Teula watersheds :
Data were provided by
Dr J. Pitiol
Centre de Recerca Ec6logica i Aplicacions Forestals (CREAF) Universitat Autônoma de Barcelona 08193 Bellaterra, Barcelona SPAIN
Telephone : (34) 581 13 12 Facsimilé : (34) 581 20 03 99
APPENDIX B:
Topographic maps of the watersheds
used in this study 100
34°
1310 1220
1400
580 0 Stream gaging station
400 f # t 'A 400 0 1 km
Figure B.1 : Langrivier Experimental Catchment (South Africa) 101
18 30 19°
Study site 1:=3
0 Cape Town 34°
0 10 20 30 4-4-11--4
0 Stream gaging station
Ls,, Rain gage
I I 0 1 km
Figure B.2 : Bakkerskloof Experimental Catchment (South Africa) 102
Figure B.3 : Réal Collobrier Research Watershed # 5
0 Stream gaging station
A Rain gage 103
Figure B.4 : Real Collobrier Research Watershed # 6
0 Stream gaging station
46, Rain gage Figure B.5 : Réal Collobrier Research Watersheds # 7 & 8
0 Stream gaging station
A Rain gage Dimas Experimental Figure B.6 : Bell Canyon and its subwatersheds in the San Forest (California)
0 Stream Gaging Station
A Rain Gage Figure B.7 - Three bar D Experimental Watershed (Arizona)
0 Stream Gaging Station
A Rain Gage 107
APPENDIX C:
Plots of the four descriptive relationships used for watershed behavior characterization
108 1200
• 1000 - •
BOO - • •
600 - • • • • • 400 - •
200 - • •
o 16100 1 ioo 2000 o 200 400 600 900 10'00 1200 1400 Precipitalion (,m)
Figure C.1 - Annual Precipitation vs Annual Runoff for the Watershed RC 5
1000 • 900 - • • BOO •
700 •
600 •
500 • •• •
400 it. • • • 300 - • 200 - • • • 100 - •
o BOO 1000 1200 1400 1600 1800 0 200 400 600 Precionation (rntn)
Figure C.2 - Annual Precipitation vs Annual Runoff for the Watershed RC 6 109
1000
900 -
700
600 - •
500 -
400 -
300 -
200 -
I. 100 -
0 14b0 MOO o 260 41100 160 160 10'00 12b0 9,1114941bn (mm) Figure C.3 - Annual Precipitation vs Annual Runoff for the Watershed RC 7
1400
1200
1000
I 600
400 -
200 -
0 - 1400 1600 11100 2000 o 200 400 SOO BOO 1000 1200 Precipltdbn (mm)
Figure C.4 - Annual Precipitation vs Annual Runoff for the Watershed RC 8 11 0 500
450 -
400 - • 350 - •
,.., 300 - 1 • ... 250 -
i 200 -
150 - • 100 -
50 -
• . 0 1 • • o 260 400 elk sôo 1000 12100 1400 Prodpitdion (mm) Figure C.5 - Annual Precipitation vs Annual Runoff for the Watershed SDEF 81
500
430 - •
400 -S
350 -
, 300 - • 1 • ) 250 -
200 - •
150 • 100
50 - • 1200 1400 200 400 BOO BOO 1000 o Pracipltdion '(mm)
Figure C.6 - Annual Precipitation vs Annual Runoff for the Watershed SDEF 82 111
500
450 -
400 - . 350 -
300 -
1 . 250 -
J 200 - 4.
150 - • 100 - • . 50 - • • lir • .• o 1260 1400 o 260 • 400 160 oôo 1000 Procipitalfon (nren)
Watershed SDEF 83 Figure C.7 - Annual Precipitation vs Annual Runoff for the
500
450 -
400 • 350
.... 300 - 1 ... 250 - •
J 200
150 • 100 -
S 50 • • • • 6 A• o 1000 12100 1400 260 400 scio sio o Procipliation (,m) vs Annual Runoff for the Watershed SDEF 84 Figure C.8 - Annual Precipitation 112 2000 •
1500
1600 - • • • 1400 - • 1200 - •
1
... 1000 -
600 "
600 -
400 -
200 -
o o 500 1000 1560 2060 2500 Pro:WW1/4m (rwn)
Figure C.9 - Annual Precipitation vs Annual Runoff for the Langrivier Watershed
1400 •
1200 -
1000 -
• • • •• • • • • • • 400 - • • o. • 200 -
o o 260 400 600 I100 1000 1260 1460 1660 11'00 2000 Pr adpitalion (rren) Figure C.10 - Annual Precipitation vs Annual Runoff for the Bakkerskloof Watershed 113
400 •
350 -
300 - • • 250 - • 1 1 200 - • 150 - •
100 - • • 50 - • • • e• • • • * • o o 260 aio silo ock 1 060 1260 1400 Prodpitanon !will)
for the Three Bar D Figure C.11 - Annual Precipitation vs Annual Runoff Watershed 114 1000 • 900 -
BOO - • . 700 - ,...
I 600 -
I 500 - 3 • • 1 400 - 1 • • • 300 - •
200 - •• . 100 - ••
0 0 260 44;0 660 1160 l&00 1260 14;30 16.00 Ws. Period Pretalion.(mm)
Figure C.12 - Wet Period Precipitation vs Wet Period Runoff for the Watershed RC 5
1000
900
BOO • e • • 700 •
600
500 e • • • 400 e
300 • e
200 - • 100 - • .
D 1600 0 260 400 SOO BOO 1 000 1260 14i00 Wet Period Precipitation (rnm)
Figure C.13 - Wet Period Precipitation vs Wet Period Runoff for the Watershed RC 6 115 1000
900 -
700 - ... 1 • 600 - •
J 500 - • ] 1 400 - .3i • 300 -
200 -
100 - • •
o 12b0 1400 o 260 460 silo slim i000 'first Poled Prod981111on (rnm)
Figure C.14 - Wet Period Precipitation vs Wet Period Runoff for the Watershed RC 7 1200 •
1000 • • • • • • • • • ••• I • s • .• 3 400 • •
200
I I I I I I 1800 0 1000 1200 1400 1600 200 400 600 BOO o Wet Poriod ProclOatIon (mm) Figure C.15 - Wet Period Precipitation vs Wet Period Runoff for the Watershed RC 8 116 500
450 -
400 - • 350 - • ,, 300
250 - • • P- 200 -
150_
100 - •
50 - • o o 260 400 660 ado lobo 12b0 1400 wit Period PredoiloKon (rtyn)
Figure C.16 - Wet Period Precipitation vs Wet Period Runoff for the Watershed SDEF 81
500
450 -
400 - •
350 -
•-•300
Pc 250 - • • 200 - • 150 -
100 - • • 50 - • • • •• o o 200 400 600 100 1000 1200 1400 Wet Period Pradpitalieri (vren)
Figure C.17 - Wet Period Precipitation vs Wet Period Runoff for the Watershed SDEF 82 117 500
450 -
400 -
350 - . I •-• 300 -
J 250 - I b 4k- 200 - s
150 -
100 - . • 50 - , • • o r•, ...-- * 0 , • o 200 400 sôo mio 1000 1200' 1400 war PrIod Pradpfteke (nwn) Figure C.18 - Wet Period Precipitation vs Wet Period Runoff for the Watershed SDEF 83
500
450 - • 400 -
350
1 • •-• 300
Il 250 I b • 4. 200 s •
150 • 100 - • • 50 - •
0 o 200 400 600 BOO 1000 1200 1400 We Pariod Pracietalien (rnm)
Figure C.19 - Wet Period Precipitation vs Wet Period Runoff for the Watershed SDEF 84 118 .
1600 - . .
1400 -
,.., 1200 - 1 4. 3 1000 - 2 I
-.. 600 -
400 -
200 -
o 0 200 400 660 1160 10'00 1200' 1400 16100 18100 2000 We Period Pracipadien (,ren)
Figure C.20 - Wet Period Precipitation vs Wet Period Runoff for the Langrivier Watershed 1400 •
1200 -
1000-
..., . 1 • k y 1 600 - z %.
400 - • Si. 41. . 200 -
0 o 200 400 660 860 1000 1200 14100 1600 1600 WO Poriod Pradoltation (own) Figure C.21 - Wet Period Precipitation vs Wet Period Runoff for the Bakkerskloof Watershed 119 250 •
200 - • --,
] •
-; 3t • 50 - • • 0 • • • * • • o 1000 1200 o 200 460 goo ado wit 444 Parbd PradoNoKen (won)
for the Three Bar Figure C.22 - Wet Period Precipitation vs Wet Period Runoff D Watershed 120 400
350 -
300 -
250 -
200 - ] sti 150- • 100 -
50 -
• o • • 0 440 o 160 260 360 sio 600 Dry Mod Predpitallon (rrett) Figure C.23- Dry Period Precipitation vs Dry Period Runoff for the Watershed RC 5 250
200
1 100
50
• • .• • •.
160 260 360 400 560 600 Dry Prod Precipitation (mm)
Figure C.24 - Dry Period Precipitation vs Dry Period Runoff for the Watershed RC 6 121 250
200 -
1 50 -
50 - •
o o 160 260 560 400 sôo 600 Ory Period Pricipitalbn (n en)
Figure C.25 - Dry Period Precipitation vs Dry Period Runoff for the Watershed RC 7 500
450 -
400 -
350 -
1 300 -
al 250 -
200
150 -
100-j
50 - • • • • • • • • • 0 o 100 200 300 400 500 600 Ory Perbd Precipitation (mm) Figure C.26 - Dry Period Precipitation vs Dry Period Runoff for the Watershed RC 8 122
100
90 -
!I 0 -
70 -
30 -
4. 20 -
4. 10 - 4. • o II • S 40 260 250 o 5 10 I lia 413 Dry ROW Pradpholko (wn) Figure C.27 Dry Period Precipitation vs Dry Period Runoff for the Watershed SDEF 81 100
90
BO
70
L0
1 50 1 1 40 g 30
. 20
.
10 . .
. 0 *. • * , •f • r` 0 2; 40 66 BO 160 1 io 1 io 160 Dry Period Prse190060A (mm) Figure C.28 - Dry Period Precipitation vs Dry Period Runoff for the Watershed SDEF 82 123
100
90 -
90 -
70 -
60 -
/ 50 -
40 -
30 -
20 -
10 - • . , • *. * • ea 0 t • 140 160 20 40 sb BO 100 120 Dry Period Precipitation (min)
Figure C.29 Dry Period Precipitation vs Dry Period Runoff for the Watershed SDEF 83 100
90 -
110 -
70 r.
60
I 50
40
• • • • • • • 100 120 140 160 0 20 40 60 10 Cry Perkci Pricipitotbn (mm) Figure C.30 - Dry Period Precipitation vs Dry Period Runoff for the Watershed SDEF 84 124 500
450 -
400 -
350 -
300 - • e 250- • • 1 • 200 - 6 • • 150 -
100 -
50 -
ido 260 3do 460 560 660 760 000 Dry Period Preelpflolion (rrim) Figure C.31 Dry Period Precipitation vs Dry Period Runoff for the Langrivier Watershed 250
200 -
1 150-
50- • • •• 4. • •
00 160 260 360 460 50'0 600 D7 Perbd Prodpitotion (mm) Figure C.32 - Dry Period Precipitation vs Dry Period Runoff for the Bakkerskloof Watershed 125 100
90 -
BO -
*
• *
20 - • • 10 - • 4. • * • * I P o 350 400 o 50 160 150 200 250 360 "Dry" (warm) Period Precipitation (mm)
for the Three Bar D Figure C.33 - Dry Period Precipitation vs Dry Period Runoff Watershed 126
•
-g. -4 • 4 • • • • • • 0 AO • • ••• • • • •• • • • • s. • + • °I° • • i ° • • 4% • 20 40 60 SO 100 120 140 140 1110 200 RaIrdall ( wn)
200
110
160
140
120 . ...-.... • 1 100
4. --..-....--1 SO • -...... --4- 60 • • -• 40 • • • .b. 20 ..-.40. . ' î 40101; qv. • • .°
101100 . 0 0 20 40 60 SO 100 120 140 160 100 200 Rakial (rren)
Watershed RC 5 Figure C.34 - Storm Precipitation vs Storm Runoff for the 127 110
160-•
140-
120 -
I 100
110
60 •-• • 40
. • • 20 • • .4+ • .4 • : • • • • • 1 • • o: • • •• --•ams•0116n461104,401"'t ..' •• 'br 1110 20 40 60 110 100 120 140 160 Rakifa1(from)
140
160-
140..
40 - Li
Figure C.35 - Storm Precipitation vs Storm Runoff for the Watershed RC 6 128 160
140 ...... -.-... I I .-..-- 120 4
100
...n...-n-n-•••1 0
60 • • . • • • 40 .6 • • • -4. -4 • 20 . • • • : 4.• .•S t • • • • • ors, 41.1 • .. • . • totAC 4.74 ft * 0 :A.-1.1E061UL'. 0 20 40 60 SO 100 120 o (svs)
160 .•: i i 1! 140 I, i• ;i I 220 I i iI 100 i 1 I 1 • 110 I I i I 60 I I 1 : •. • i 40 ---* • . •• . : : 20 i I oop • • 416 i 1 I I ! I --Amsawomior."°°41117. , , : 0 100 120 140 160 0 20 40 60 SO isktd (rm)
for the Watershed RC 7 Figure C.36 - Storm Precipitation vs Storm Runoff 129 280
160
I 140 r- I
120 : I
I 100 • SO f •
• ------,- 60 _.... --Ilt • • • • • 40 * mot *. • —• f • • • . 0 1 • .. • • • n • • • • • 20 ----4,••--la * • • • 40 • . • • 40 • • • . • • _ _...... ANAP"..... t.t. • 46:1 • 1, .2. . 0 0 20 40 50 80 100 120 140 150 180 Rdnfal (ftim)
180
160 ,
140
120 •ir
. . 200 . -r • SO
or
---.. qt •------, 50 * •
40 •
20
0 0 20 40 GO BO 100 120 140 160 280 Rdnfal (rren)
Figure C.37 - Storm Precipitation vs Storm Runoff for the Watershed RC 8 130 300
4- I- 250
200
t 150
100 0 ...- -...... 50
• . . .• -- I • 250 300 50 100 150 200 Wei ( wn)
0-
O 300
CC
CO 250
CO *a. CO 200 a- CC
•
100 150 200 250 300 0 50 si added (rivn)
Figure C.38- Storm Precipitation vs Storm Runoff for the Watershed SDEF 81
•• ▪ ▪
131 110 •-. CO . - 160 -
46
▪ 140 -
— 120 Co
CO C c 100 - • CO CC .4.-- - 00 A 60 0
• 40 CO 20 4 D • Ab. 4 4.1: • ...dit • • 0 160 150 0 20 40 60 SO 100 120 140 Adria! (min)
"8 110
CC 160
CO 140 - = CO .= 4 4 CO 120 CC E 100.
• 44 , 1 CO 110 4 2
^1. • 60 . CO
40 CO CC 4 20 - CO •-• D 140 160 110 20 40 60 110 100 120 Li Ralriot (rnm) Watershed SDEF 82 Figure C.39 - Storm Precipitation vs Storm Runoff for the 132 1 0
140
120
I 100 - • 80
--1 60
40 -
20 - • • ••• • "" 0 140 160 1110 0 20 40 • 0 SO 100 1;0 Waal (ffen)
180
160
140
a 4.
60 -4
40
4 -6 20 • • 160 180 20 40 60 BO 100 120 140 (rnm) •
Figure C.40 - Storm Precipitation vs Storm Runoff for the Watershed SDEF 83 133
ISO
160
140-
120-
100- •
110 -
60
• h-• 40-
20 - 110 • • • • • o MI - 150 4 120 140 160 o 20 40 60 $O 100
110
160
140
120-
100 - •
SO -
60-
--f 40-
20- • • • r• • •. * s.- 120 140 160 ISO 20 40 60 S O 100
Figure C.41 - Storm Precipitation vs Storm Runoff for the Watershed SDEF 84 134
56 16o 150 200 2i0 360 350 Rainfall (mm)
+ + + -**++
100 150 200 250 360 350 .c; Rainfall (mm) Figure C.42 - Storm Precipitation vs Storm Runoff for the Langrivier Watershed 135
250 .....it.' 03 a
o 200 C 0 CC
C 03 = 1150 co ...E .Z — CO 1 CC E ts 4 To. p 100 ...... 4. C .c-a o .c0 50 4-, CO C • V) • • D • 0 CCI o 50 100 150 200 250 11001)1 (mn) .....V. co a.
r 250 o C M CC 13 C CO 200 = 03 —"E CO CC i 150 -cp CI) .0 G) •-• 1 CO • 2 1p 100 1:3 C ci2 v cn -Nt C 50 CD CC CO 4 • C 4* 'FA ---...... • •flp • ••• * D 0 o 50 100 150 200 250 il Malnial (rrrn) Figure C.43 Storm Precipitation vs Storm Runoff for the Three Bar D Watershed 136
LITERATURE CITED
Barbour, M. G. and R. A. Minnich. 1990. The myth of chaparral convergence. Israel Journal of Botany. 39:453-463.
Beven, K. 1987. Towards a new paradigm in hydrology. In: Water for the Future: Hydrology in Perspective (Proceedings of the Rome Symposium, April 1987. pp. 393-403.
Biswas, A. K. 1970. History of Hydrology. North Holland. Amsterdam. 336 p.
Bosch, J. M. and J. D. Hewlett. 1982. A review of catchments experiments to determine the effect of vegetation changes on water yield and evapotranspiration. Journal of Hydrology. 55:3-23.
CEMAGREF, 1990. Les bassins versants du Réal Collobrier. Centre National du Machinisme Agricole, du Génie Rural, des Eaux et des Forêts. Aix- en Provence, France. 23 p.
Chapman, T. 1989. Techniques for inter-regional comparison. In: M. Falkenmark and T. Chapman eds. Comparative Hydrology: an ecological approach to land and water resources. Unesco, Paris. pp. 131-144.
Chow, V. 1964. Runoff. In: Chow V. ed. Handbook of applied hydrology. McGraw-Hill, New York. pp 14-1 - 14-55.
Crawford, J. M. Jr. 1962. Soils of the San Dimas experimental forest. USDA Forest Service Pacific Southwest Forest and Range Experiment Station miscellaneous paper No 76. Berkeley, California. 20 p.
Di Castri, F. 1981. Mediterranean-type shrublands of the world. In: Di Castri, F., D. W. Goodall and R. L. Specht editors. Mediterranean-type shrublands. Elsevier, Amsterdam. pp 1-52.
Di Castri, F. and H. A. Mooney. 1973. Mediterranean type ecosystems: origin and structure. Springer-Verlag, Heidelberg. 405 p.
Diskin, M. H. 1970. Definition and uses of the linear regression model. Water Resources Research. 6:1668-1673. 137
Falkenmark, M. 1989. Introduction. In: M. Falkenmark and T. Chapman eds. Comparative Hydrology: an ecological approach to land and water resources. Unesco, Paris. pp. 1-8.
Fogel, M. M. and L. Duckstein. 1971. Prediction of convective storm runoff in semi-arid regions. In: IASH-UNESCO symposium on the results of research on representative and experimental basins. IASH-UNESCO, Wellington, New Zealand. pp 465-478.
Hawkins, R. H. 1992. A taxonomy of small watershed rainfall-runoff. Hydrology and water resources in Arizona and the Southwest: 21.
Hewlett, J. D. 1982. Principles of Forest Hydrology. The University of Georgia Press, Athens. 183 p.
Hibbert, A. R. 1969. Forest treatment effects on water yield. In: W. E. Sopper and H. W. Lull eds. International Symposium on Forest Hydrology. Pergamon, Oxford. 813 p.
Hibbert, A. R. 1971. Increases in streamflow after converting chaparral to grass. Water Resources Research. 7: 71-80.
Hill, L. W. 1963. The San Dimas experimental forest. USDA Forest Service Pacific Southwest Forest and Range Experiment Station. Berkeley, California. 26 p.
Johnson, A. W. 1973. Historical view of the concept of ecosystem convergence. In: Di Castri, F. and H. A. Mooney. 1973. Mediterranean type ecosystems: origin and structure. Springer-Verlag, Heidelberg. pp 1-7.
Kim, K. H. 1989. Classification of environmental hydrologic behaviors in northeastern United States. Unpublished Master's thesis. University of Arizona. School of Renewable Natural Resources. 212 p.
Lavabre, J. 1989. Research drainage basin of the Real Collobrier (Mediterranean France): flow modelization and regional analysis. In: Friends in Hydrology, IAHS publication No 187. International Association of Hydrological Sciences. Washington D.C. pp 423-434. 138
Lindley, A. J., J. M. Bosch and D. B. van Wyk. 1988. Changes in water yield after fire in fynbos catchments. Water SA. 14: 7-12.
Linsley, R. K., M. A. Kohler and J. L. H. Paulhus. 1975. Hydrology for engineers. Mc Graw Hill, New York. 482 p.
Lewis, D. C. 1968. Annual hydrologic response to watershed conversion from oak woodland to annual grassland. Water Resources Research. 4: 59-72.
Naha!, I. 1981. The mediterranean climate from a biological viewpoint. In Di Castri, F., D. W. Goodall and R. L. Specht editors. Mediterranean-type shrublands. Elsevier, Amsterdam. pp 63-86.
Nash, J. E., and J. V. Sutcliffe. 1970. River flow forecasting through conceptual models. Part 1: a discussion of principles. Journal of Hydrology, 10: 282-290.
Pase, C. P. and D. E. Brown. 1982. Interior chaparral. Desert Plants. 4: 95- 99.
Piriol, J., M. J. Lledo and A. Escarré. 1991. Hydrological balance of two Mediterranean forested catchments (Prades, northeast Spain). Hydrological Sciences Journal, 36: 2. pp 95-107.
Rice, R. M. 1969. Storm runoff from chaparral watersheds. Ph.D. Dissertation. University microfilms, Ann Arbor, Michigan. 152 p.
Seyhan, E. 1982. A universal application of the multivariate statistical techniques to hydrology. In: V. P. Singh editor. Statistical analysis of rainfall and runoff. Water resources publication, Littleton, Colorado.
Shachori, A. Y. and A. Michaeli. 1965. Water yields of forest, maquis and grass covers in semi-arid regions: a literature review. In: F. E. Eckardt editor. Methodology of plant eco-physiology. Proceedings of the Montpellier symposium. UNESCO, Paris.
Specht, R. L. 1969. A comparison of the sclerophyllous vegetation characteristic of mediterranean type climates in France, California and Southern Australia. Part 1: Structure, morphology and succession. Australian Journal of Botany. 17: 277-292. 139
UNESCO, 1977. Map of the world distribution of the arid regions. Explanatory note. MAB technical notes, 7. UNESCO, Paris. 54 p.
Van der Zel, D. W. and F. J. Kruger. 1975. Results of the multiple Catchment experiments at the Jonkershoek research station, South Africa. 2. Influence of protection of fynbos on stream discharge in Langrivier. Forestry in South Africa. 16: 13-18.
Van Wilgen, B. W. and F. J. Kruger. 1985. The physiography and fynbos vegetation communities of the Zachariashoek catchments, south- western Cape province. South African Journal of Botany. 51: 379- 399.
Van Wyk, D. B. 1987. Some effects of afforestation on streamflow in the western Cape province, South Africa. Water SA. 13: 31-36.
Zavoianu, I. 1985. Morphometry of drainage basins. Elsevier, Amsterdam. 238 p.