A hidden<scp>M</scp>arkov model combined with climate indices for multidecadal streamflow simulation

Bracken, C.; Rajagopalan, Balaji; Zagona, Edith

doi:10.1002/2014wr015567

Cited by 49 publications

(44 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…With a regime‐shifting model, the distribution of an observable hydrological variable (e.g., streamflow) shifts in concert with an unobserved (or “hidden”) regime state variable. The latter is assumed to be a discrete, stochastic process, such as a Markov chain, that tends to transition on an interannual basis [ Thyer and Kuczera , ; Sveinsson et al ., ; Akintug and Rasmussen , ; Prairie et al ., ; Gelati et al ., ; Bracken et al ., ]. The regime‐shifting model offers several desirable features, including compatibility with actual climate drivers that fluctuate inter‐annually, accurate reflection of the lack of knowledge of the factors that underpin observed hydrology, and compatibility with the useful and theoretically defensible assumption of stationarity [ Koutsoyiannis and Montanari , ; Serinaldi and Kilsby , ].…”

Section: Introductionmentioning

confidence: 99%

Regime‐shifting streamflow processes: Implications for water supply reservoir operations

Turner

Galelli

2016

Water Resources Research

View full text Add to dashboard Cite

This paper examines the extent to which regime-like behavior in streamflow time series impacts reservoir operating policy performance. We begin by incorporating a regime state variable into a well-established stochastic dynamic programming model. We then simulate and compare optimized release policies-with and without the regime state variable-to understand how regime shifts affect operating performance in terms of meeting water delivery targets. Our optimization approach uses a Hidden Markov Model to partition the streamflow time series into a small number of separate regime states. The streamflow persistence structures associated with each state define separate month-to-month streamflow transition probability matrices for computing penalty cost expectations within the optimization procedure. The algorithm generates a four-dimensional array of release decisions conditioned on the within-year time period, reservoir storage state, inflow class, and underlying regime state. Our computational experiment is executed on 99 distinct, hypothetical water supply reservoirs fashioned from the Australian Bureau of Meteorology's Hydrologic Reference Stations. Results show that regime-like behavior is a major cause of suboptimal operations in water supply reservoirs; conventional techniques for optimal policy design may misguide the operator, particularly in regions susceptible to multiyear drought. Stationary streamflow models that allow for regime-like behavior can be incorporated into traditional stochastic optimization models to enhance the flexibility of operations. Key Points:Regime shifts in streamflow impact reservoir operating performance A regime state can be incorporated into stochastic reservoir optimization Allowing retrospectively for regime shifts improves simulated operating performance Vogel, 2015]. In this study, we examine the limitations of short-memory streamflow models when applied to reservoir operating problems. We focus on the use of such models for informing water release decisions through a classical stochastic dynamic programming (SDP) method, and seek to quantify the potential gains in operating performance that could be proffered by allowing for regime-like behavior in reservoir inflow. PUBLICATIONS The Role of Hydrological Information in Reservoir OperationThe question of how to best incorporate hydrological information into reservoir operations has been the focus of many studies. When constructing an optimization model, one normally aims to draw on useful hydrological information to train the policy, whist also acknowledging that this information will need to be available to the operator when deciding how much water to release. The seminal studies of Gal [1979] and Maidment and Chow [1981] employed the previous period's inflow, which is useful if the inflow time series has significant autocorrelation. Alternatively, if one assumes a reservoir operator can adjust the release as the decision period unfolds-which may be appropriate for monthly operations-the current period inflow can be used as th...

show abstract

Section: Introductionmentioning

confidence: 99%

Regime‐shifting streamflow processes: Implications for water supply reservoir operations

Turner

Galelli

2016

Water Resources Research

View full text Add to dashboard Cite

show abstract

Section: Methodsmentioning

confidence: 99%

“…The HMM uses Markov chain assumption, in which the probability of the occurrence of next state depends only on the current state and not on the rainfall state that preceded it in the state space (Zucchini and Guttorp, 1991). In sum, the HMM accounts for spatial dependence in the data and capture the spatio-temporal pattern of rainfall intensity and its probability of occurrence over a network of stations or grid points (Robertson et al, 2006;Greene et al, 2008;Bracken et al, 2014;Pal et al, 2015). The HMM tool developed by the International Research Institute (IRI) was employed in this study to identify rainfall states that could capture the spatio-temporal variability within the NEIMR season over the study region (IRI, 2007).…”

Section: Methodsmentioning

confidence: 99%

Characterizing pentad rainfall variations during the North‐East Indian monsoon season over the southeastern peninsular India

Shanmugasundaram

Lee

Srinivasan³

2018

Intl Journal of Climatology

View full text Add to dashboard Cite

North‐East Indian Monsoon rainfall (NEIMR) during October–December is of immense socio‐economic importance to the agriculture‐dependent population in the southeastern peninsular India. NEIMR is subject to extreme year‐to‐year and intra‐seasonal variability that needs to be understood to enhance climate resilience. In this study, we employed hidden Markov model to characterize the spatio‐temporal variations of NEIMR at pentad time step and its probability of occurrence during 1982–2014. The results indicated the dominant presence of three rainfall states during the NEIMR season, which were the wet (State‐1), coastal wet (State‐2), and dry (State‐3) states. Seasonal total NEIMR was significantly and positively correlated with the frequency of State‐1, whereas it was negatively correlated with that of State‐3, indicating a crucial role of the rainfall states in determining water requirements in the southeastern peninsular India. The rainfall states were associated with distinctive atmospheric circulation and surface temperature conditions, particularly the wet (State‐1) and dry (State‐3) conditions. Wet conditions were characterized by enhanced cyclonic activities and increased moisture convergence at 850 hPa over the southeastern peninsular India and its neighbouring oceanic regions (Bay of Bengal and Indian Ocean). In contrast, dry conditions were associated with anticyclonic circulation and reduced moisture convergence at 850 hPa. The plausible physical mechanisms behind the wet (dry) condition could be that anomalous warmer (cooler) land temperature above 20°N induced lower (higher) sea level pressure anomalies and drove anomalous southwesterly (northeasterly) surface winds over the NEIMR region. These anomalous surface winds and the associated lower level cyclonic (anticyclonic) circulations could enhance (suppress) moisture transport from the convergence regions over the Bay of Bengal and northern Indian Ocean into the southern peninsular India. This study revealed the pentad variability of NEIMR with the classified three rainfall states and identified the key atmospheric circulation and surface temperature conditions linked to these rainfall states.

show abstract

“…[] and Bracken et al . []. Lower Colorado Basin flows are also modulated by these climate forcings [ Thomas , ].…”

Section: Introductionmentioning

confidence: 99%

“…Recently , Bracken et al . [] showed that regime‐like behavior of the Colorado River flows are forced by AMO and PDO, and used a nonhomogeneous hidden Markov model to simulate the flow properties using these two climate forcings.…”

Section: Introductionmentioning

confidence: 99%

Wavelet‐based time series bootstrap model for multidecadal streamflow simulation using climate indicators

Erkyihun

Rajagopalan

Zagona

et al. 2016

Water Resources Research

View full text Add to dashboard Cite

A model to generate stochastic streamflow projections conditioned on quasi‐oscillatory climate indices such as Pacific Decadal Oscillation (PDO) and Atlantic Multi‐decadal Oscillation (AMO) is presented. Recognizing that each climate index has underlying band‐limited components that contribute most of the energy of the signals, we first pursue a wavelet decomposition of the signals to identify and reconstruct these features from annually resolved historical data and proxy based paleoreconstructions of each climate index covering the period from 1650 to 2012. A K‐Nearest Neighbor block bootstrap approach is then developed to simulate the total signal of each of these climate index series while preserving its time‐frequency structure and marginal distributions. Finally, given the simulated climate signal time series, a K‐Nearest Neighbor bootstrap is used to simulate annual streamflow series conditional on the joint state space defined by the simulated climate index for each year. We demonstrate this method by applying it to simulation of streamflow at Lees Ferry gauge on the Colorado River using indices of two large scale climate forcings: Pacific Decadal Oscillation (PDO) and Atlantic Multi‐decadal Oscillation (AMO), which are known to modulate the Colorado River Basin (CRB) hydrology at multidecadal time scales. Skill in stochastic simulation of multidecadal projections of flow using this approach is demonstrated.

show abstract

A hiddenMarkov model combined with climate indices for multidecadal streamflow simulation

Cited by 49 publications

References 61 publications

Regime‐shifting streamflow processes: Implications for water supply reservoir operations

Regime‐shifting streamflow processes: Implications for water supply reservoir operations

Characterizing pentad rainfall variations during the North‐East Indian monsoon season over the southeastern peninsular India

Wavelet‐based time series bootstrap model for multidecadal streamflow simulation using climate indicators

Contact Info

Product

Resources

About