⏳Intro to Time Series Unit 16 Review

Hydrological time series analysis applies the core tools of time series modeling to water-related data: streamflow, water levels, precipitation, and similar measurements. Understanding these patterns is essential for predicting water availability, managing flood risk, and planning infrastructure. Climate change and land use shifts are actively reshaping hydrological systems, making trend detection and scenario modeling more important than ever.

Analysis of Streamflow Data

Before you can model or forecast anything, you need clean data and a solid understanding of what's in it. Hydrological time series analysis typically follows three stages: preprocessing, exploratory analysis, and pattern identification.

Preprocessing ensures your data is reliable before any modeling begins:

Handle missing values and outliers using interpolation or statistical methods (mean imputation, regression imputation, or more advanced gap-filling techniques)
Apply standardization or normalization so data from different stations or measurement scales can be compared (z-score standardization, min-max scaling)

Exploratory data analysis helps you understand the structure of your data visually and numerically:

Inspect time series plots (line plots, scatter plots) to spot patterns, anomalies, and apparent trends
Calculate summary statistics to characterize the distribution: mean, median, standard deviation, skewness, and kurtosis

Pattern identification uses formal time series tools to uncover structure:

Autocorrelation (ACF) and partial autocorrelation (PACF) functions reveal dependencies and repeating patterns in the data, such as seasonality or persistence from one time step to the next
Spectral analysis detects periodic components and their dominant frequencies. For streamflow data, you'll often find strong annual cycles and sometimes sub-annual (e.g., monthly) patterns
Trend analysis determines whether water levels or streamflow are systematically increasing or decreasing over time

Analysis of streamflow data, HESS - Exploring the role of hydrological pathways in modulating multi-annual climate ...

Trends in Hydrological Data

Most hydrological time series contain multiple overlapping signals. Decomposition separates these into interpretable components so you can study each one individually.

Additive vs. multiplicative decomposition:

Additive model: $Y_t = T_t + S_t + R_t$ . Use this when the seasonal fluctuations stay roughly constant in magnitude over time.
Multiplicative model: $Y_t = T_t \times S_t \times R_t$ . Use this when seasonal swings grow or shrink proportionally with the level of the series.

In both cases:

$T_t$ is the trend component, capturing long-term direction (gradual increase or decrease in streamflow, for example)
$S_t$ is the seasonal component, capturing recurring patterns within a fixed period (annual snowmelt peaks, monsoon cycles)
$R_t$ is the residual component, the leftover fluctuations after trend and seasonality are removed. Analyzing residuals helps you judge whether your decomposition captured the important structure.

Estimating trends can be done several ways:

Linear or nonlinear regression to fit a trend line (linear, quadratic, or other functional forms)
Moving averages and smoothing techniques (simple moving average, exponential smoothing) to filter out noise and reveal the underlying direction

SARIMA models handle both seasonal and non-seasonal patterns in a single framework. The notation is SARIMA( $p,d,q$ )( $P,D,Q$ ) $_m$ , where:

$p, d, q$ are the non-seasonal autoregressive order, differencing order, and moving average order
$P, D, Q$ are the seasonal counterparts of those same three parameters
$m$ is the number of periods per season (e.g., 12 for monthly data with an annual cycle, 4 for quarterly data)

The non-seasonal parameters capture short-term dependencies, while the seasonal parameters capture patterns that repeat every $m$ time steps. For example, a SARIMA model on monthly streamflow data would use $m = 12$ to account for the annual cycle.

Forecasting of Water Availability

Forecasting water availability means choosing a model, fitting it to historical data, and then evaluating how well it predicts unseen observations. Several model families are commonly used.

ARIMA models are the standard statistical approach. The ARIMA( $p,d,q$ ) notation means:

$p$ : number of lagged values of the variable used as predictors (autoregressive terms)
$d$ : number of times the series is differenced to achieve stationarity
$q$ : number of lagged forecast errors used as predictors (moving average terms)

Building an ARIMA model follows these steps:

Check for stationarity (using plots and tests like the Augmented Dickey-Fuller test). If the series isn't stationary, difference it until it is. The number of differences becomes $d$ .
Examine the ACF and PACF of the differenced series to identify candidate values for $p$ and $q$ .
Fit candidate models and compare them using information criteria (AIC, BIC). Lower values indicate a better balance of fit and parsimony.
Estimate parameters via maximum likelihood or least squares.
Run diagnostic checks on the residuals: they should resemble white noise (no remaining autocorrelation, approximately normal distribution).

Exponential smoothing methods offer a simpler alternative, chosen based on the data's characteristics:

Simple exponential smoothing for data with no trend or seasonality
Holt's linear trend method for data with a trend but no seasonality
Holt-Winters' seasonal method for data with both trend and seasonality

Machine learning approaches can capture nonlinear relationships that statistical models may miss:

Artificial neural networks (ANNs) learn complex patterns from the data
Support vector machines (SVMs) perform regression-based forecasting
Random forests, an ensemble method, can improve accuracy and robustness by averaging many decision trees

Model evaluation and selection is critical regardless of which approach you use:

Split data using cross-validation strategies suited to time series (rolling window validation is preferred over random k-fold, since time order matters)
Calculate error metrics to quantify forecast accuracy:
- RMSE (root mean squared error) penalizes large errors heavily
- MAE (mean absolute error) gives equal weight to all errors
- MAPE (mean absolute percentage error) expresses error as a percentage, making it easier to interpret across different scales
Compare models using AIC and BIC, which penalize model complexity to guard against overfitting

Impact Assessment and Climate Change

Climate Impact on Hydrological Processes

Climate change and land use change both alter how water moves through a landscape. A key challenge is figuring out which factor is responsible for observed changes, and how they interact.

Detecting climate change impacts:

Trend analysis of temperature and precipitation records reveals long-term shifts (warming trends, altered rainfall patterns)
Changes in hydrological regimes show up as shifts in timing or magnitude of streamflow. Earlier snowmelt, more frequent floods, and longer dry spells are common examples.

Assessing land use change impacts:

Urbanization and deforestation increase surface runoff and reduce infiltration, which lowers groundwater recharge
Agricultural practices like irrigation and water abstraction can reduce streamflow and increase nutrient loads in waterways

Attributing changes to specific causes requires statistical tools that can separate overlapping influences:

Multiple regression and principal component analysis help isolate the contributions of climate vs. land use
Multivariate time series techniques capture relationships between variables:
- Cointegration and error correction models identify long-run equilibrium relationships (e.g., between water demand and supply, or groundwater levels and pumping rates) along with short-term dynamics
- Granger causality tests assess whether changes in one variable systematically precede changes in another (e.g., does land use change predict subsequent changes in streamflow?)

Scenario analysis and projections use models to explore possible futures:

Climate model outputs (under scenarios like RCP4.5 or RCP8.5) serve as inputs to hydrological models, producing projections of future water availability and flood risk
Different land use scenarios (afforestation, sustainable agriculture) can be modeled to estimate their hydrological effects

Adaptation and mitigation strategies translate analysis into action:

Water resource management practices such as conservation programs and demand management help cope with shifting conditions
Land use planning measures like zoning regulations and riparian buffer zones can reduce the negative hydrological effects of development

⏳Intro to Time Series Unit 16 Review