Forecasting the Rainfall Pattern on Upstream of Hirakud Reservoir Using L-Moment for Accessing the Inflow

Changes in the rainfall pattern are a challenge for filling schedule of reservoir, when it is fulfilling various demands. In monsoon fed reservoirs, the target remains for attaining full reservoir capacity in order to meet various demands during non-monsoon period and the flood control. The planners always eye towards the inflow trend and perspective frequency of rainfall in order to counter the extreme events. In this study, the case of Hirakud reservoir of Mahanadi basin of India is considered as this reservoir meets various demands as well as controls devastating floods. The inflow trend has been detected by using Mann Kendall test. The frequency analysis of monthly rainfall is calculated using L-moment program for finalizing a regional distribution. The falling trend in inflow to reservoir is visualized in the month of July and August. The Wakeby distribution is found suitable for the monthly rainfall of July, September and October, where as in June and August, General Extreme Value (GEV), General Normal (GN) and Pearson Type-III (PT-III) distributions are found suitable. The regional growth factors for the 20, 40, 50 and 100-year return period rainfalls along with inflow to reservoir observed between 1958-2010 are calculated in this study as a referral for reservoir operation policy.


Introduction
Hirakud reservoir as a multipurpose dam resolving many demands of state of How to cite this paper: Gupta, K.K., Kar, A.K., Jena, J. and Jena, D.R. (2017) Forecasting the Rainfall Pattern on Upstream of Hirakud Reservoir Using L-Moment for Odisha.The operation of the reservoir is governed by a specified rule curve and it is supposed to attain its full reservoir level at the end of October 31st (end of monsoon season) in order to meet the demands till start of the monsoon season (end of May).In this regard the rainfall and its distribution of upstream districts play pivotal role in filling schedule of the reservoir.There are numbers of studies in these regards and a few of them are discussed.Reference [1] shows the characteristics of Miyun reservoir of China.It was found that annual inflow, flood peaks decreases in both quantity and proportion and this decreasing extent is higher than that of common runoff.Human activities like high population growth and corresponding land use changes remain the major concern.Forecasting of future water availability to meet conflicting demands is the major task.Under this study [2] seasonality, stochasticity and non-linearity of the inflow were identified.Reference [3] has analysed the effect of climate change for northern reservoirs of Indiana using Self Organisation Map (SOM) to perform trend and cluster analysis.The Mann Kendall and Revised Mann Kendall were used for regionalization.Another study [4] made on a saline reservoir of state Odisha, India which focused on trend analysis of rainfall using Mann Kendall, inflow quantification using ARNO model and performance evaluation of reservoir using WEAP model under changing climatic situations.Reference [5] has studied about trends in water quality and quantity for 11 major reservoirs of the Brazos and Colorado river basins in the southern Great Plains.The components like water quality, major contributing-stream inflow, storage, local precipitation, and basin-wide total water withdrawals were analyzed.The study of [6] has applied the Mann-Kendall (MK) statistical trend test on a wide range to analyze increasing, decreasing or trendless characteristics of precipitation, temperature, inflow to dam reservoirs, release from dam reservoirs, and storage volume in dam reservoir in Thailand from historical operation recorded data.As per [7], reservoir inflow corresponding to varying rainfall pattern under climate change conditions causing the variation of the reservoir inflow.They have proposed a hybrid SD model named Wavelet Support Vector Machine (WSVM) in combination of the multiscale Principal Components Analysis (MSPCA) and nonlinear Support Vector Machine regression model for Sutami Reservoir, Indonesia and found that WSVM shown better result in inflow forecasting.
L-moment approach has been applied to many events in number of cases for finding parent distribution.In case of flood frequency analysis L-moment has been applied in many occasions.However, in different rainfall events application of L-moment is seen in cases like [8], found that the frequency analysis of the largest or the smallest of a sequence of hydrologic events has long been an essential part of the design of hydraulic structure.Reference [9] has used L-moment for regionalization of annual precipitation in northern Central Italy.Reference [10] has applied L-moment for selection of parent distribution to fit maximum monthly rainfall data of 18 sites of the Zayanderhood basin, Iran.The obtained extreme rainfall values can be used for meteorological drought management in the arid zone.Reference [11] has applied L-moment for probability distribution of extreme value 1-day rainfall events.

Study Area
The Hirakud reservoir of Mahanadi basin cover almost 83,400 sq km of catchment area out of total catchment of 141,589 sq km and covering most part of two states like Chhatisgarh and Odisha (Figure 1).
The reservoir Hirakud was commissioned since 1958 as a multipurpose project.A catchment of 83,400 sq km drains into reservoir Hirakud covering 24 districts of four states fully or partly (Figure 2).The reservoir is monsoon fed  As the study is related to inflow of reservoir, districts of its upstream part are only considered.[14] [15] [16]).It is a common practice to use a non-parametric test to detect a trend in a time series.This test, being a function of the ranks of the observations rather than their actual values, is not affected by the actual distribution of the data and is less sensitive to outliers.On the other hand, parametric trend tests, although more powerful, require the data to be normally distributed and are more sensitive to outliers.The Mann-Kendall test is therefore more suitable for detecting trends in hydrological time series, which are usually skewed and may be contaminated with outliers.This test has been extensively used with environmental time series [17].The Mann-Kendall trend test is based on the correlation between the ranks of a time series and their time order.For the statistics S is calculated as Equation ( 1).This statistic represents the number of positive differences minus the number of negative differences for all the differences considered as ( )

Methodology
where, n is the number of total data points, i x and j x are the data values in time series i and ( ) j i > , respectively, and ( ) is the sign function as: ) The variance of Mann-Kendall test is calculated by Equation (3) as where, n is the number of total data points, m is the number of tied groups.The tied group means a simple data having a same value.The i t indicates the number of ties of extent, i.In case of the sample size, 10 n > , the standard nor- mal test statistic s Z is estimated by Equation ( 4) as ( ) The positive values of s Z show increasing trends while negative values represent falling trends.As 5% significance level is taken standard for this study, the null hypothesis of no trend is rejected if 1.96 s Z > .
2) L-moment analysis: Three statistical measures discordancy measure, heterogeneity measure and goodness of fit measure as per L-moment approach are used in regional studies.These measures are explained by [18].It is used to estimate the degree of heterogeneity and to assess whether they might reasonably be treated as homogeneous.Specifically, the heterogeneity measure compares between site variations in sample L-moments for the group of sites with that expected for a group of region.Hosking's Heterogeneity test fits 4 parameter Kappa distributions.A series of 500 simulations ( sim N ) done and L-statistics of ac- tual region is compared with a simulated series.The H-statistics defined as, ( ) For each simulated region, the measures of variability i V (where i V is any of three measures 1 V , 2 V and 3 V ) is calculated.From the simulated data, the mean v µ and standard deviation v σ of the sim N values of i V are deter- mined.Journal of Water Resource and Protection The critical H statistics for a region to be homogeneous is as mentioned below 1 H < Homogeneous (6) Reference [12] has observed that statistics  V and 3 V lack the power to discriminate between homogeneous and hetero- geneous regions but 1 H based on 1 V has much better discriminating power.
So 1 H is treated as a much better indicator of heterogeneity measure.Also, 1 H was found to be a better indicator of heterogeneity in large regions, but has a tendency to give false indication of homogeneity for small regions [19].
The measure

2
H indicates whether at-site and regional estimates will be close to each other.A large value of 2 H indicates whether or not the at-site and regional estimates will be in agreement, whereas a large value of 3 H indicates a large deviation between at-site estimates and observed data.
3) Selection of Regional Distribution: The regional distribution has been adjudged on the basis of Z-statistics as follows: It indicates suitability of a candidate distribution to a data series and is appropriate for evaluating and comparing a distribution.The Z-statistics for the goodness of fit measure as defined by Hosking is ( )

Results and Discussion
From the catchment map (Figure 2) it is found that, there are 23 districts in the upstream catchment of Hirakud reservoir out of them 7 districts comprising 13.10% of upstream catchment belongs to Odisha state.Chhatisgarh has 15 districts occupying 86.54% of the same.
As the upstream catchment analysis is needed for finding the inflow characteristics first of all the influence area of each districts which is influencing the inflow are obtained and mentioned in Table 1.occupy 50% of geographical catchment, rainfall over these districts also to be carefully viewed as these may substantially influence the reservoir inflow.
The time series for month wise inflow to the reservoir is drawn for the period from 1958 to 2010 (Figures 3-7).The average inflows during these periods are in June 0.98, July 5.93, August 9.64, September 6.65 and in October 1.89 MAc ft.
From the time series of June it is seen that, only 4 times an inflow of more than 4 MAc ft have occurred.2. So, the reservoir has to meet these demands as well as to control the flood or to create space for expected flood.
Applying trend analysis using Mann Kendall test, it was found that, no trends have been observed in June, September and October.In July no trend was seen at 1% and 5% significance level but a falling trend at 10%.In August, no trend was seen at 1% Significance level but falling trend at 5% and 10% (Table 3).
For finding the frequency of rainfall first of one homogeneous region was established by applying L-moment approach.The unsuitable districts are removed from the homogeneous regional group as per the discordancy test.The districts having discordancy values more than 3 are discarded.The rainfall values of Bilaspur district has been discarded in all the 5 months.The suitable distributions are selected from the goodness of fit criteria.The distributions and corresponding growth factors are shown in Table 4.Further the inflow to dam is cal-

Conclusion
The rainfall over five districts like Raipur, Durg, Raigarh, Bilaspur, Korba is very sensitive as this occupies almost 50% of the upstream catchment area.The rainfall during month of August is showing a falling trend.As most of the agricultural activities continue during this time a falling trend may hamper.The regional distributions like GEV, GN, PT-III and Wakeby are found suitable for month of June and August whereas Wake by found suitable for the month of July, September and October.The distributions GL, GN and Wakeby hold good for inflow forecasting taking the observed inflow series.The time series, trend analysis, regional fitting distributions and corresponding growth factors derived are useful for further planning and management of the reservoir operation.

1 )
Trend Detection: To identify trend in climatic variables with reference to climate change, the Mann-Kendall test has been employed by a number of researches with temperature, precipitation and stream flow data series ([12] [13] K. K. Gupta et al.DOI: 10.4236/jwarp.2017.9120851339 Journal of Water Resource and Protection

2 H and 3 H
based on measure of

) 4 Z 4 σ
DIST = a particular distribution, 4 DIST Z = L-kurtosis for fitted distribution, = pooled L-kurtosis, 4 B = bias correction, = estimate of sample va- riability of L-kurtosis.The DIST Z value should be close to zero.However, a value between −1.64 and 1.64 is considered to be suitable for a fitting distribution at 10% significance level.While a number of distributions may qualify the goodness-of-fit criteria, the most potential will be one that has minimum DIST Z value.The rainfall values based on different frequencies are calculated from the relationship between t R and m R , where t R is the rainfall value at particular return period t and m R is the mean rainfall.Data Used: The district wise rainfall data from 1958-2010 is used from the India Meteorological Department (IMD) sources and the Inflow to reservoir has been obtained from Government of Odisha [20].

Figure 3 .
Figure 3. Timeseries of inflow to Hirakud Reservoir in the month of June.

Figure 4 .
Figure 4. Timeseries of inflow to Hirakud Reservoir in the month of July.

Figure 5 .
Figure 5. Timeseries of inflow to Hirakud Reservoir in the month of August.

Figure 6 .
Figure 6.Timeseries of inflow to Hirakud Reservoir in the month of September.

Figure 7 .
Figure 7. Timeseries of inflow to Hirakud reservoir in the month of October.

Table 1 .
Influence area of Upstream districts.

Table 4 .
Suitable regional distributions for monthly rainfall with corresponding growth factors.