Review of Yield Comparison Options for Planning Irrigation Reservoirs in Ungauged Watersheds

Irrigation reservoir design is often associated with ungauged watersheds. Irrigation system planners have the option of selecting either traditional institutional guidelines or more recently calibrated and verified hydrologic models to compute watershed yield. A design watershed yield is the key for sustainable irrigation reservoir development. Design values not only should lead to reliable estimates but also require the incorporation of safety factors for both rainfall uncertainty and modelling uncertainties. Empirical design yield calculation options in guidelines require explicit rationalization to determine appropriate design levels. Use of popular watershed models , developed and calibrated with other datasets, too requires the guidance to convert the estimations as design watershed yield with the use of safety factors. Therefore, both options need to clearly express the reliability of yield estimations and the associated safety factors for a sustainable practice of water management. Comparison of available alternatives in the light of methods and incorporation of safety is a necessity to select the appropriate option for water resources design in ungauged watersheds. Accordingly, the objective of the present work is to carry out a critical review of available methods to recommend a selection for estimation of streamflow yield from ungauged watersheds. A comprehensive literature review is carried out considering available text books, guidelines and other literature. The study looked at the runoff estimation methods, strengths and difficulties, incorporation of safety factors etc., when deriving design watershed yield from watershed yield. The review of 110 related documents revealed that, although there is literature on models for the estimation of watershed yield, there is a gap in the applicability at various locations and in the conversion to design watershed yield. Irrigation Department Sri Lanka has reported the most detailed direct yield model with threshold considerations. Unit Hydrograph model has many text book references while HEC HMS is the most commonly cited process-based model for runoff estimations. The review shows that, amidst the drawbacks in the present documentation, the best available option for a practicing engineer to compute design watershed yield is to use a similar approach that is indicated in the guidelines of Irrigation Department of Sri Lanka. There is an urgent necessity to verify guideline recommended design watershed yield estimation methods by using observed data and take measures to incorporate explicit safety factors.


Introduction
Irrigation systems that contribute towards food security are considered as the backbone of many nations. They provide strength to a large farming community to survive amidst difficult living conditions. Reliable estimation of streamflow from watersheds is a key factor when carrying out planning and design of irrigation systems [1]. World over, streamflow gauging for distributed hydrologic modelling is a rarity. Therefore, irrigation infrastructure development is mostly on ungauged watersheds. In case of ungauged watersheds within a gauged watershed or otherwise, reliable streamflow extrapolation methods are required for rational engineering design. Even in the case of gauged watersheds, streamflow estimations are mentioned as unreliable due to poor monitoring of streams and lack of sufficient records to enable estimations [1]. Irrigation systems are large water related infrastructure investments that cover vast extents of land.
Since water infrastructure investments depend on the reliability of water yields, estimation of streamflow magnitudes and patterns in ungauged watersheds is one major concern with regards to the sustainability of such systems. Water is considered as the most valuable resource because of its natural scarcity and due to higher demand arising from population increase, hence reliable streamflow estimation to a greater accuracy is the main challenge faced by ungauged watershed Design yield for practical applications in ungauged watersheds needs to identify a watershed model that will ensure the accuracy of estimations while incorporating appropriate safety factors.
In case of watershed streamflow estimations, on one hand there are time tested and coarse resolution empirical models, while on the other hand there are most recent generalized process based hydrologic models which can provide distributed outputs at finer temporal and spatial resolutions.
Presently available guidelines show a clear necessity to provide adequate details for comparative evaluations in order to make a selection. Hence, a practicing water engineer, planner or a watershed manager is faced with the dilemma of whether to use a hydrological process based mathematical model or to follow a more straightforward time-tested empirical model. Therefore, even at locations where models can be calibrated and verified, there is a need for the explicit identification of safety factors. The desire of practicing engineers and managers to use modelling advances for their work makes it an urgent need to establish guidelines when design watershed yields are computed.
Accordingly, the objective of the present work is to critically review the available guidelines and research publications to recommend the most suitable watershed yield estimation option for the design of water infrastructure such as sustainable irrigation reservoirs at ungauged locations.

Watershed Yield and Uncertainties
Water yield form watersheds is the main factor for rational water use decisions.
Hence watershed yield is the most important input for reservoir systems that treat water as a resource. Dependable yield is useful and important in water management systems especially during droughts [2]. In literature, water yield is referred to as watershed yield, watershed runoff, streamflow, watershed outflow, catchment inflow to reservoirs etc. Watershed yield is the amount of streamflow accumulated over a time period and at a particular geographic location. Design watershed yield corresponds to the streamflow from a watershed which is considered as the inflow for engineering infrastructure designs. In order to ensure safe designs, it is necessary to accommodate safety factors to account for real world uncertainties. Streamflow assessments for water use are mainly associated with medium and low flows [1].
Accordingly, planning and engineering for water use is concerned with the complete streamflow hydrograph over a representative period of several years [3].
The evaluations and recommendations for water use are commonly based on either monthly or seasonal temporal scales [4], [5], [6]. Guideline of the Irrigation Department of Sri Lanka [7], the references related to India [8], and many others for water planning in irrigation and associated development are for monthly or coarser time resolutions [9], [10], [11].
Watershed yield estimations face two main types of uncertainties. One is the uncertainty associated with respect to the determination of watershed rainfall. The other is the uncertainty associated with respect to the model chosen to represent the watershed heterogeneity and associated hydrologic and hydraulic processes [12], [13], [14]. Issues such as inadequate representation of hydrologic processes, lack of guidance on parameter values, optimization deficiencies coupled with the mathematical algorithms etc., cause uncertainties that need to be considered when converting model outputs as design watershed yield [15], [16]. There are two kinds of uncertainty with respect to rainfall. One is the watershed averaged rainfall computed with point measurements and the other due to occurrence of rainfall.
In hydrologic computations, uncertainty of rainfall is dealt by considering the probability of occurrence. The same for a selected model is taken care of by considering the accuracy of model estimations. However, in the conventional design of water resources engineering projects, only uncertainties of rainfall occurrence are considered by selecting a large return period and artificially considering this design level as a safety factor [3], [17], [18]. Uncertainties and effects due to rainfall distribution over a catchment has been discussed by many with respect to watershed response [19]; the need to evaluate rainfall distributions considering inter-event dry periods for the reliability of runoff predictions had been identified [20]; the importance of design rainfall for hydrologic and hydraulic design calculations [21] are stated in literature but the lack of clear statements regarding design catchment yield could be noted.
According to Indian standards, yield estimation is done based on either previous rainfall and streamflow records, using rainfall runoff coaxial relations or with unit hydrograph method [8]. In Nepal guideline, ungauged catchment runoff is estimated by two methods as (1) WECS regional regression for long term mean and low flows and (2) MIP design manual method for mean 80% flows. According to WECS method, mean monsoonal precipitation iso-lines that had been developed for the entire country and monthly flow calculation method is given by using a monthly coefficient. According to MIP method, Country is divided into seven regions & mean monthly flow and 80% reliable flow charts had been developed. 80% monthly flow value is taken for the design criterion of irrigation schemes to ascertain the reliability of full supply. The concept of "hydrologically similar catchments" is used for getting runoff values when there is lack of data [9].
In literature, there are many regression type models such as box-jenkins time series regression, auto regressive moving average, WECS regional regression method and USGS empirical regression method for watershed runoff estimation [2], [9], [13], [26]. Other than regression models, there are many empirical equations developed for regions [45]. Inglis regional formula for annual streamflow which had been developed for hilly and plain areas in India [42], [46] is also mentioned in Irrigation guideline in Pakistan [10]. The Khosla's formula for monthly or annual streamflow estimation had successful applications in India and USA [46]. Parker's formula for annual streamflow has versions for British isles, USA and Germany [42]. Irrigation Department (ID) guideline of Sri Lanka [7] facilitates a direct method for monthly watershed yield estimation based on seasonal iso-yield curves. According to this guideline, the net catchment area multiplied by the specific seasonal yield gives the seasonal yield for Yala and Maha seasons. The specific seasonal yields are obtained by iso-yield curves that had been published in 1984. There are two safety considerations in the method as, (1) using 75% probable rainfall and (2) using seasonal yield limitations for the obtained yield. Considering the need to account for climatic changes, land use pattern changes and the change to irrigation practices, several researchers showed the need to verify the design parameters of present Irrigation Department guideline [47], [48]. In ID guideline method, the apportioning of yield to each month and the use of flow thresholds lead to a safe watershed yield.
Storm water drainage manual of Hong Kong [22], while recommending the use of many statistical and deterministic models, highlights the details of common deterministic models such as rational method, time area method, UH method and reservoir routing models.

Australian Rainfall Runoff guidelines [49], [50]
provide methodologies on estimation of rainfall, flood estimation and surface runoff estimation. Areal reduction factor, that can be considered as a safety factor for rainfall is provided in both editions. Though simple and design event hydrograph estimation methods are provided in old edition [49], Ensemble and Monte Carlo methods are used in new guideline, [50]. Uncertainty in design rainfall estimation and design flood estimation is discussed in both guidelines.
In many guidelines for hydrology, water resources, irrigation, watershed management etc.
There are many streamflow estimation computer models which facilitate detailed modelling of many hydrological processes in a watershed at high spatial and temporal resolutions. There are brief text book references to popular models such as HEC-HMS, SWMM, SWAT, TR20, TR55, HSPF ( [18], [23], [24], [40], [55]). These and many other popular computers based watershed models provide individual detailed technical documentation on the incorporation and use of hydrologic rationalizations in model computations.
UH is a popular recommendation in the guidance materials on watershed runoff estimation. Snyder's Synthetic Unit hydrograph [56], the Soil Conservation Service (SCS) Unit Hydrograph [57] and Clark's unit hydrograph method, Gary's method etc., are commonly selected streamflow computation methods [35], [58], [59], [60]. Derivation of a composite hydrograph by using a series of unit hydrographs enables the generation of streamflow times series. Chow et al. [3] provide details on the potential of using UH for the generation of hydrographs and the use of Kinematic Wave method for routing along stream channels to capture the hydrograph at a desired location. There are a wide variety of research that had taken place on UH applications, parameter optimization, parameter identification with physical characteristics, parameter transferability etc., [54], [60], [61], [62], [63], [64], [65], [66], [67].

Design Yield/ Yield Thresholds
In literature, there is a clear deficiency of methods to compute the design yield from a watershed. Many point out the need for design yield computations to carefully examine the dependability of mathematical models and associated simplifications with regards to the reliability of streamflow estimations [3], [4], [5], [6], [81], [82], [83].
Troin et al. [15], in their work, stating that hydrologic model structure contributes most to uncertainty on the projected streamflow, indicate the importance to determine appropriate safety factors when determining design watershed yields. Bourdin & Stull [13], Olsson and Lindström [14], and Mays [84], highlighting the uncertainty in both the meteorological input and the hydrologic model, discuss the importance of an appropriate final value for the hydrological forecast. Turner & Jeffrey [85] recognized the importance of clearly expressing methods to overcome the modelling uncertainties in order to make appropriate planning decisions associated with the development of water resources. Konukcu [81] recognized the increased reservoir costs and the environmental hazards in agricultural area due to over prediction of water yield by a method used by Turkish General Directorate of Rural Services.
The need for judgement and experience, the inclination to err on the side of over estimation, and the incorporation of conservative factors [22], [86], [87], [88]  There are practical situations that had considered only a percentage of dependable annual yield to cater to the randomness of watershed yield estimated either by using empirical or mathematical models [41]. A direct method to determine water availability at reservoir sites endorsed by the National Water Development Agency (NWDA) of the Ministry of Irrigation/ Water Resources, India has a design consideration which uses the 75% probable natural streamflow value in the multiplication factor that caters to inflow uncertainties [89], [90]. Irrigation guideline in Nepal uses 80% dependable yield for planning purposes [9].
The direct method of estimation from historical yield values recommended by the Irrigation Department guidelines of Sri Lanka uses 75% probable rainfall to apportion seasonal values to monthly values and then incorporates two thresholds as percentages to avoid extremely high or low design yield values. As an example, Welsh Water Resources Planning guideline [91], which is one of the documentations that emphasises the need of a rational risk factor for the hydrological yield when determining the Deployable Output (DO), does not lead a practicing person with technical guidance for computations.

General
The present work, in its attempt to establish the state-of-the-art of design yield computation methods, reviewed 15 related guidelines, 10 books, 18 monographs/theses and 56 peer reviewed journal publications. In this list of documents, the balance of coverage between flood and yield estimation references is shown by Figure 1.

Figure 1 -Coverage of Assessment Types
Literature cited many runoff estimation tools. Watershed streamflow estimation models frequently cited in guidelines are restricted to a few models such as empirical models, rational method, Unit Hydrograph method, SCS Curve number method and HEC HMS model ( Figure  2). The other category refers to methods which have not been used by a significant number of references.

Figure 2 -Common Runoff Estimation Tools
It is noteworthy that guideline references are mainly on empirical and UH type models while the process type models are yet to be established as guideline material.

Yield Models and Issues
Due to uncertainty of parameters, interdependency of parameters, dependency of parameters on the model and objective functions [92], unavailability of data for calibration [93], expensiveness of models, availability of higher number of parameters [94] etc., it is difficult for a practicing engineer to use many of the available watershed yield models for most practical situations.
The empirical methods in guidelines not only seem to use probable rainfall in design inputs but also consider thresholds for streamflow estimations. These thresholds appear as implicit safety factors.
Except for a few water resources design guidelines that consider adjustment factors, others reflect a lack in the explicit indication of safety factors [22], [30], [88]. Though the need for incorporating a rational risk factor for water yield is stressed in guidelines/manuals [23], [95], [96] there is no guidance at least with an appropriate range of risk factors.

5.2.1
Empirical Methods Empirical methods and regional methods are specific to particular localities. In empirical models, the mathematical formulae must be revalidated for better accuracy.
Though empirical methods are better practiced in specific geographical regions, the main advantage is because it is a direct method and has been used over a long period of time. In some guidelines, rainfall and streamflow uncertainties have been considered in these methods [9], [22]. Some have embedded safety factors in an implicit manner [7], [9]. Use of empirical relationships is common in the south Asian region. Nepal guidelines convert rainfall iso-lines as monthly yield using coefficients while Sri Lankan guidelines use seasonal isoyield curves which are converted to monthly yield using rainfall pattern.

Irrigation Guideline -Sri Lanka
Among the direct methods in literature, the ID empirical model [7] is a well-balanced, and rational option for design watershed yield computations. This has been used to compute design watershed yield for many major, medium and minor surface water harvesting schemes in Sri Lanka. In this guideline, development of iso-yield curves and 75% probable rainfall values had been with data between 1943-1977 period [97].
Hence updating is required to accommodate climate change and land cover change factors that may otherwise lead to errors in the yield estimations. Seasonal maps also lead to issues when climate change effects occur, not only between seasons but also between months. Yield maps have been drawn at a very coarse resolution raising concerns about interpolated values.
Yield thresholds do not point to publications that can be considered as verifications.

UH Method
Though unit hydrograph method is mostly used for direct runoff computations, total runoff hydrograph computation is carried out by adding the baseflow estimates to the direct runoff [3]. When using this method, soil water balance, Hershfield nomograph, USDA-SCS method [98] and Φ-index method [3] are some effective rainfall calculation methods. However, Hershfield method is not suited for rice growing lowlands and USDA-SCS method has shown under predicted results for South Indian watersheds [99]. India, Japan, Vietnam and Burma use different empirical methods for effective rainfall calculation [100] that mainly consider the effective rainfall for plant growth, and not for watershed yield. Φ-Index method is a simple method for effective rainfall calculation that assumes a uniform loss rate throughout the year. Baseflow estimation can be done using various techniques. Separation of observed hydrographs, frequency analysis, recession method [101], [102], [103] storage yield analysis [103], [104], [105] and baseflow index method (BFI) [101], [104], [106], [107], [108] are available methods. However, BFI method is the easiest and simplest method available for ungauged catchments. There is a difficulty when applying BFI for shorter time durations such as monthly or daily. Regional parameters for Sri Lanka watersheds have guidance material with limited research work which can be used as verification [7], [66], [37], [109]. The large number of assumptions is a critical factor that requires consideration when predicting ungauged watershed yield using UH method.

5.2.4
HEC-HMS Model HEC-HMS is a process-based model that can be used for streamflow estimation under different conditions and spatiotemporal scales. This model has many conceptual options for each hydrologic process. Hence, process selection is complex, model consists of many parameters, requires an above average hydrological knowledge for development, and demands a long data series for calibration and verification. In case of ungauged watersheds, the model requires calibration and verification or applicable approximations. Model applications in ungauged catchments are questionable. Educated selection of donor catchments will be beneficial for ungauged catchment modelling [110]. Literature does not provide generalised guidance on model parameters. However, in case of model components, the most popular loss transform and baseflow models are deficit and constant [76], [77], [80], SCS unit hydrograph or Snyder's unit hydrograph model [76], [77], [79], [80], and recession method [33], [76], [77], [78], [79].

Safety Factors
Engineering designs for infrastructure construction always embed safety factors to cater to the uncertainties. Margins of safety are usually conceptualised by considering the associated uncertainties. Though rainfall and watershed process conceptualisations are known to possess uncertainties, none of the literature provided explicit guidance for a design engineer to perform infrastructure designs by using the outputs from hydrologic models. This factor leads to a drawback when a designer has to decide whether to use a streamflow from a traditional model or a recent model.
Though the literature shows uncertainties related to rainfall and model, it is difficult to find safety factors to be used for irrigation infrastructure design from literature.

Qualitative Evaluation
The qualitative evaluation based on the findings in the work is shown in Table 1. The selected key factors for the selection of a method which were based on the literature survey are, a) possibility of monthly estimates b) availability of rainfall safety factors c) availability of a safety factor for streamflow d) application simplicity of the model e) availability of in country guide or research publications as guidance, and f) the data demanded for a meaningful application.
The climatic and topographic variations pertaining to a particular country or a watershed are features embedded either in the data such as rainfall, evaporation etc., or in the sophistication of the model structure. Hence these were not considered as key factors. Weightages were given considering high (H), medium (M), low (L) categorization for each key factor based on literature. Table 2 explains the consideration given for the aligning of H, M and L for each key factor. The qualitative ranking was then converted to a numerical indicator by using a scale of 3, 2 and 1 for High, Medium and Low respectively.
According to the results, Sri Lankan ID yield calculation method gets first priority (weightage 17), and guideline recommended methods clearly show a higher priority than other hydrological models to be used in practical situations.