📄 earth_science

Assessing Precipitation Memory and Operational Controls on National-Scale Reservoir Storage and Hydropower Dynamics in Türkiye

This study employs a data-efficient framework combining STL, an optimized precipitation-memory index, and VARX modeling to demonstrate that while hydro-climatic forcing drives water availability fluctuations in Türkiye's national reservoirs, institutional operational management ultimately governs how these signals persist and propagate through the water-energy system over time.

Original authors: Bülent SELEK

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Bülent SELEK

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Giant Water Bank

Imagine Türkiye's reservoir system as a massive, national "water bank." This bank has two main jobs:

Storing water (like money in a savings account) for when it's dry.
Generating electricity (like spending that money to pay bills) when the turbines spin.

The author, Bülent Selek, wanted to understand how this "bank" behaves. Specifically, he asked: How much does the weather (rain) control the water level, and how much do human decisions (like when to release water for power) control it?

The Problem: Why Rain Doesn't Always Mean Full Tanks

Usually, we think: "It rained yesterday, so the tank should be full today." But in reality, the relationship is messy.

The Lag: Rain doesn't instantly fill a dam. It has to travel through soil, rivers, and underground aquifers first. This is like pouring water into a sponge; it takes time to soak through.
The Human Factor: Even if there is plenty of water, humans might hold back the release to save for later, or they might release it all at once to make electricity.

The paper argues that to understand the system, we can't just look at "rain today." We have to look at the history of rain over the last few months.

The Solution: The "Rain Memory" Tool

The author created a special mathematical tool called "Precipitation Memory."

The Analogy: Think of this like a weighted average of your recent mood. If you had a great day today, but a terrible day yesterday, your overall mood isn't just "great"—it's a mix.
How it works: The study found that the "memory" of rain in Türkiye lasts about 12 months, but it's heavily weighted toward the most recent 3 months.
- Rain from today counts for about 55%.
- Rain from last month counts for about 25%.
- Rain from 3 months ago counts for about 10%.
- Rain from a year ago barely matters anymore.

This tool helps filter out the "noise" of the seasons (like knowing it's always wet in winter and dry in summer) to see the real cause-and-effect relationships.

The Experiment: Shaking the System

The author used a computer model (VARX) to simulate what happens when the "Rain Memory" changes.

1. The "Shock" Test (Impulse Response)
Imagine someone suddenly adds a huge bucket of water to the system (a heavy rain event).

The Reservoir (The Tank): The water level goes up immediately and stays high for a long time. It's like a heavy backpack; once you put it on, you carry it for a while.
The Hydropower (The Generator): The electricity output spikes immediately but then drops back down quickly. It's like a sprinter who bursts out of the blocks fast but then slows down to a jog. The system uses the extra water to make power right now, then settles back to normal.

2. The "Who's in Charge?" Test (Variance Decomposition)
The study asked: "When we try to predict the future, what matters more: the weather or the human operators?"

The Result: Surprisingly, human operations (the internal rules of the power plant) are the biggest factor in predicting future electricity output (about 75%).
The Weather's Role: The weather (rain memory) and the current water level only explain about 25% of the changes.
The Takeaway: The system is so well-managed that it smooths out the weather. The humans are the "captains" steering the ship, while the weather is just the wind.

3. The "Drought" Simulation
The author simulated what happens if it rains 30% less than usual for two years straight.

The Result: The water level doesn't just drop once; it keeps dropping slowly over time. By month 24, the storage is about 6.8% lower than normal.
The Analogy: It's like a leaky bucket. If you stop pouring water in, the leak (or the usage) slowly empties the bucket over time. The system tries to buffer the drought, but eventually, the "savings account" runs low.

The Main Conclusion

The paper concludes that Türkiye's water system is a tightly coupled team of nature and humans.

Nature (Rain) provides the raw material and starts the process.
Humans (Operators) decide how long that water stays in the system and how fast it is used.

Even though the weather is the source of the water, the behavior of the system (how long the water stays, how much power is made) is mostly controlled by the rules and decisions of the people managing the dams.

Why This Matters (According to the Paper)

This study proves that we can't just look at "how much it rained today" to predict water shortages. We have to look at the history of rain (the memory) and understand that human management acts as a giant buffer. If the rain stops for a long time, that buffer eventually runs out, leading to a slow but steady decline in water storage.

In short: The system is smart and resilient, but if the "rain memory" stays negative for too long, even the best managers can't keep the tank full forever.

Technical Summary: Assessing Precipitation Memory and Operational Controls on National-Scale Reservoir Storage and Hydropower Dynamics in Türkiye

Problem Statement
Multi-purpose reservoir systems operate at the intersection of hydro-climatic variability and human decision-making. In semi-arid and Mediterranean regions like Türkiye, precipitation is the primary meteorological forcing, yet observed reservoir storage dynamics often exhibit weak contemporaneous correlations with monthly precipitation. This decoupling arises because storage evolution is not a direct response to immediate inputs but rather the cumulative outcome of delayed catchment processes (e.g., soil moisture redistribution, routing) intertwined with complex operational regulations (e.g., hydropower dispatch, irrigation demands).

A central methodological challenge in analyzing this coupling is distinguishing true climate forcing from shared seasonality and trend components, which can lead to spurious correlations if not explicitly separated. Furthermore, precipitation exerts "memory" effects across multiple timescales, yet these are often represented implicitly via fixed lag structures or autoregressive terms, limiting physical interpretability. There is a lack of integrated frameworks that simultaneously (i) explicitly model precipitation memory, (ii) separate structural time-series components, and (iii) quantify directional dynamic dependencies among precipitation, storage, and hydropower at a national scale.

Methodology
The study employs a data-efficient, transferable framework utilizing nationally aggregated monthly data for Türkiye (precipitation, active reservoir storage ratio, and hydropower capacity factor). The analytical workflow consists of four integrated stages:

Seasonal–Trend Decomposition (STL): Raw time series are decomposed into trend, seasonal, and stationary residual components using Loess. All subsequent analyses utilize the stationary residuals to isolate genuine dynamic interactions from common seasonal cycles.
Optimized Precipitation-Memory Index ( $PM_t$ ): Instead of arbitrary lag selections, precipitation memory is modeled using an exponentially weighted distributed lag formulation:
$PM_t = \sum_{k=0}^{K} w_k P_{t-k}$
where weights $w_k$ follow an exponential decay function controlled by a memory-decay parameter ( $\tau$ ). This structure is optimized by minimizing the Akaike Information Criterion (AIC) and cross-validation errors.
Vector Autoregressive Model with Exogenous Inputs (VARX): A multivariate model is constructed where the endogenous variables are reservoir storage residuals ( $S_t$ ) and hydropower capacity factor residuals ( $CF_t$ ). The optimized precipitation-memory index ( $PM_t$ ) serves as the exogenous climatic forcing. The model captures dynamic feedbacks without strong a priori assumptions regarding interaction direction.
Dynamic Diagnostics and Simulation:
- Impulse Response Functions (IRF): Analyze the temporal propagation of shocks from precipitation memory to storage and hydropower.
- Forecast Error Variance Decomposition (FEVD): Quantify the relative contribution of climatic forcing, storage variability, and endogenous hydropower dynamics to forecast uncertainty.
- Drought Scenario Simulation: Forward-looking simulations impose sustained precipitation deficits (10%, 20%, 30%) to evaluate system resilience and the accumulation of storage anomalies over a 24-month horizon.

Key Results

Precipitation Memory Structure: The optimization identified a 12-month temporal lag window with a decay parameter of $\tau = 1.25$ . Despite the extended window, the influence is heavily concentrated at short lags: approximately 55% of the weight is contemporaneous, and ~91% occurs within the first three months.
Dynamic Responses:
- Storage: Reservoir storage responds immediately (at $h=0$ ) to precipitation-memory shocks, with the response remaining positive but gradually attenuating over 12 months.
- Hydropower: The capacity factor also shows an immediate positive response but exhibits a significantly faster rate of attenuation, returning toward baseline conditions within two months. This suggests operational adjustments rapidly utilize improved water availability for generation.
Variance Decomposition (FEVD): While precipitation memory contributes a stable, non-trivial share (~14–15%) to hydropower forecast uncertainty, the system is dominated by endogenous dynamics. Approximately 74–75% of the forecast error variance in the hydropower capacity factor is attributable to its own innovations, indicating strong internal persistence driven by operational schedules and infrastructure constraints.
Drought Propagation: Sustained precipitation deficits generate progressive, non-linear storage anomalies. Under a severe 30% precipitation deficit scenario, the simulated storage drawdown reaches approximately 6.8 percentage points relative to baseline conditions after 24 months. This demonstrates that while operational buffering mitigates short-term volatility, it cannot fully offset prolonged climatic stress.

Significance and Claims
The paper claims that national-scale reservoir dynamics are governed by a tightly coupled interaction between external hydro-climatic forcing and internal operational regulation. The study demonstrates that weak contemporaneous precipitation–storage correlations can coexist with strong precipitation-driven influence once catchment-scale precipitation memory is explicitly modeled.

Methodologically, the work provides a transferable framework that integrates structural time-series decomposition, optimized memory estimation, and dynamic multivariate inference. This approach cleanly disentangles hydro-climatic impulses from operational modulation without the exhaustive data requirements of physically distributed models.

Practically, the findings offer diagnostic insights for adaptive water-energy governance. The results indicate that hydro-climatic variability provides the primary source of water-availability fluctuations, while institutional regulation controls how these signals persist and propagate. The proposed framework serves as a tool for assessing reservoir resilience and understanding water–energy interactions under increasing hydro-climatic variability, particularly in Mediterranean-type systems where drought risks are intensifying. The study concludes that macro-scale reservoir systems must be conceptualized as coupled human–water systems, where physical processes and management decisions co-evolve over time.