CLIGEN Parameters & Formulation

Understanding CLIGEN Parameters

[!NOTE] This guide summarizes the statistical formulations used by CLIGEN to model precipitation occurrence, daily rainfall depth, and daily temperature.

The USDA CLIGEN (Climate Generator) model was developed by the Agricultural Research Service to generate synthetic daily weather inputs for hydrological, erosion, and water resources models.

Rather than assuming daily precipitation follows a normal distribution, CLIGEN uses a two-state Markov chain for rain occurrence, a skewed distribution for wet-day rainfall depth, and wet/dry conditioning for temperature.


1. Rainfall Occurrence: The 2-State Markov Chain

In nature, weather has "memory." If it rained today, a storm front is overhead, meaning the odds of rain tomorrow are significantly higher than if today was a dry, sunny day.

CLIGEN captures this behavior using two transition probabilities calculated for each calendar month:

  • $P(W|W)$, Probability of a Wet Day given Yesterday was Wet:
    In plain English: "If it rained today, what are the odds it rains again tomorrow?"
    Typical values: Usually ranges between $0.40$ and $0.70$ ($40\% - 70\%$). High values mean storms tend to linger for several days in a row (storm clustering).
  • $P(W|D)$, Probability of a Wet Day given Yesterday was Dry:
    In plain English: "If today was dry and clear, what are the odds a brand-new storm arrives tomorrow?"
    Typical values: Usually much lower, often between $0.10$ and $0.25$ ($10\% - 25\%$).
stateDiagram-v2
    Dry --> Dry: 1 - P(W|D)
    Dry --> Wet: P(W|D) (New Storm Starts)
    Wet --> Wet: P(W|W) (Storm Persists)
    Wet --> Dry: 1 - P(W|W) (Sky Clears)

By calculating separate $P(W|W)$ and $P(W|D)$ values for each of the 12 months, the simulator replicates seasonal dry spells (long streaks of dry days) and seasonal storm clusters.


2. Rainfall Amount: The Pearson Type III Distribution

Once the model decides that a day is "Wet", it must determine: How much rain falls on that day?

Daily rainfall cannot follow a standard bell curve because:

  1. Rainfall can never be negative (you cannot have $-5\text{ mm}$ of rain).
  2. Most rainy days produce small or moderate showers (light rain).
  3. Very rarely, an extreme downpour delivers a huge amount of water in 24 hours.

This creates a heavily skewed distribution with a long right-hand tail. CLIGEN models this using the Pearson Type III (Gamma-family) distribution, defined by three statistical properties for each month:

  • $\text{Mean}$ (Average Daily Wet-Day Depth):
    The average rainfall amount on days when measurable rain occurs (ignoring all dry days).
  • $\text{StdDev}$ (Standard Deviation):
    How widely storm depths vary around the monthly average.
  • $\text{Skew}$ (Skewness Coefficient):
    A measure of asymmetry. Higher skewness means downpours occur that are much larger than the average shower. Capturing historical skewness is what allows hydrologic models to evaluate flood and spillway overflow risks.

3. Daily Temperatures: Wet/Dry Conditioning & Persistence

For each calendar month, the simulator calculates four baseline temperature parameters from the historical record:

  • $\mu_{T,\max}$ (Mean Tmax): Unconditional monthly mean of daily maximum temperature.
  • $\sigma_{T,\max}$ (SD Tmax): Monthly standard deviation of daily maximum temperature.
  • $\mu_{T,\min}$ (Mean Tmin): Monthly mean of daily minimum temperature.
  • $\sigma_{T,\min}$ (SD Tmin): Monthly standard deviation of daily minimum temperature.

How Wet/Dry Conditioning Works

Cloud cover and evaporative cooling reduce daytime heating on rainy days, so wet days have lower maximum temperatures on average than dry days in the same month. To account for this without requiring separate wet and dry input tables, the generator shifts the daily expected maximum temperature ($\mu_{\max,t}$) around the unconditional monthly mean $\mu_{T,\max}$ based on the day's wet/dry status:

  1. Monthly Wet-Day Fraction ($P_w$): From the Markov transition probabilities $P(W|W)$ and $P(W|D)$, the steady-state probability of a wet day in that month is: $$P_w = \frac{P(W|D)}{1 - P(W|W) + P(W|D)}$$

  2. Wet/Dry Mean Shift: Using the wet-day cooling offset $\Delta T$ ($2.0^\circ\text{C}$ or $3.6^\circ\text{F}$), the expected maximum temperature is raised on dry days and lowered on wet days in proportion to $P_w$: $$\mu_{\max,t} = \begin{cases} \mu_{T,\max} + P_w \Delta T & \text{if Day } t \text{ is Dry} \\ \mu_{T,\max} - (1 - P_w) \Delta T & \text{if Day } t \text{ is Wet} \end{cases}$$

Because $(1 - P_w)(+P_w \Delta T) + P_w(-(1 - P_w)\Delta T) = 0$, wet days average $\Delta T$ cooler than dry days while the overall monthly mean across all days remains equal to $\mu_{T,\max}$.

Cross-Correlation and Day-to-Day Persistence

Daily temperatures exhibit both same-day correlation between highs and lows and multi-day persistence across consecutive days:

  1. Cross-Correlation ($r = 0.65$): Two independent standard normal draws $z_1, z_2 \sim \mathcal{N}(0, 1)$ are combined so warmer daytime highs tend to pair with warmer nighttime lows: $$z_{\min} = r z_1 + \sqrt{1 - r^2}\, z_2$$

  2. Lag-1 Autoregression ($\rho = 0.60$): Standardized residuals ($\epsilon_{\max,t}, \epsilon_{\min,t}$) carry memory from the previous day ($t-1$) to reproduce multi-day warm and cold spells: $$\epsilon_{\max,t} = \rho\,\epsilon_{\max,t-1} + \sqrt{1 - \rho^2}\, z_1, \qquad \epsilon_{\min,t} = \rho\,\epsilon_{\min,t-1} + \sqrt{1 - \rho^2}\, z_{\min}$$

  3. Daily Temperature Synthesis: $$T_{\max,t} = \mu_{\max,t} + \sigma_{T,\max}\,\epsilon_{\max,t}, \qquad T_{\min,t} = \mu_{T,\min} + \sigma_{T,\min}\,\epsilon_{\min,t}$$

If a random draw produces $T_{\min,t} > T_{\max,t}$, the two values are swapped so that $T_{\min,t} \le T_{\max,t}$ always holds.


Summary of the 12-Month Matrix

When the Climate Simulator calculates your site's parameters, it produces two 12-row tables (one row for each month, January through December) containing:

  • 5 Precipitation Values per Month: $P(W|W)$, $P(W|D)$, $\text{Mean}$, $\text{StdDev}$, and $\text{Skew}$
  • 4 Temperature Values per Month: $\mu_{T,\max}$ (Mean Tmax), $\sigma_{T,\max}$ (SD Tmax), $\mu_{T,\min}$ (Mean Tmin), and $\sigma_{T,\min}$ (SD Tmin)

These 12 monthly rows define the statistical inputs used by the simulation engine.