Testing PrecipGen Across Four Sites

Posted by  Jason Lillywhite

This blog post presents the results of testing PrecipGen across four diverse locations to evaluate its performance:

• Australia Plains, AU
• Grand Junction Walker Field, CO, USA
• Dublin, EI
• Orlando, FL, USA

The findings demonstrate that PrecipGen successfully represents both short- and long-term daily precipitation totals at individual locations, validating its ability to replicate observed climatic behavior across varied environments.

To read more about the PrecipGen model and how to use it, please refer to the GoldSim Library where you can read about it and download PrecipGen PAR and PrecipGen for your own use: Precipitation Simulator (PrecipGen) – GoldSim Help Center

Site Data

Obtaining long-term, high-quality precipitation data is essential for testing and validating models like PrecipGen. For this initial study, I selected four locations with approximately 120 years of daily precipitation records:

• Australia Plains, AU
• Grand Junction Walker Field, CO, USA
• Dublin, EI
• Orlando, FL, USA

While these datasets provided robust historical coverage, they contained sporadic missing records. To address these gaps, I used stochastic imputation with PrecipGen to generate plausible precipitation values based on surrounding data and time of the year. This step ensured that the input datasets were both complete and consistent, allowing for meaningful comparisons between observed and simulated precipitation patterns.

The dataset for all sites tested consists of daily total precipitation measurements spanning January 1, 1900, to January 1, 2022. While the records were extensive, minor gaps were identified and filled by replicating adjacent years to maintain consistency in the time series. Data source: Data Search | National Centers for Environmental Information (NCEI)

After preparing the data, I used PrecipGen PAR to calculate the input parameters required to run PrecipGen for each site. Proper configuration of the model was critical to ensure accurate simulations. For each location, PrecipGen PAR generated key outputs, including seasonal parameters, long-term random walk settings, and correlation coefficients, all tailored to the specific historical data of each site.

Explanation of Model Performance

PrecipGen was tested using data from four sites around the world to evaluate its accuracy in simulating daily precipitation patterns over the long term. The testing process required careful data collection and preprocessing, including filling minor gaps in historical records through stochastic imputation with PrecipGen. This ensured consistent and complete datasets for each location.

For this initial analysis, I focused on a set of parameters that seem reasonable for comparing simulated results to historical data:

• Monthly Total Precipitation: I randomly sampled the historical time series and calculated the average monthly totals across all sampled years. To compare, I ran 1 calendar year simulations for multiple realizations, matching the number of realizations to the number of years in the historical record.
• Annual Total Precipitation Distribution: Simulated using a 1-year duration across multiple realizations.
• Peak Day Precipitation Distribution: Simulated using a 1-year duration across multiple realizations.
• Running Averages (e.g., 5-year and 10-year): This metric required a long-term simulation to capture trends and variability over extended periods. For this analysis, I used a simulation period spanning 2025 to 2175, allowing the model to represent long-term dynamics effectively.

To calculate input parameters for PrecipGen, I applied a 2-year sliding window over the historical data, generating rolling estimates of the following key parameters:

• SUM: Total precipitation for each window.
• PWW: Probability of a wet day following another wet day.
• PWD: Probability of a wet day following a dry day.
• ALPHA: Gamma distribution shape parameter.
• BETA: Gamma distribution scale parameter.
• Volatility: This measures the degree of variation in precipitation over time, indicating how much precipitation levels fluctuate.
• Reversion Rate: This parameter describes the speed at which precipitation levels return to their long-term mean after a deviation.
• Correlation Coefficients for parameter pairs PWW-PWD, PWW-ALPHA: These coefficients measure the strength and direction of the relationship between pairs of parameters, such as the probability of a wet day following a dry day (PWD) and the gamma distribution shape parameter (ALPHA).

These parameters formed the foundation for simulating both short- and long-term precipitation dynamics.

Overall, the results of the tests are promising and indicate that PrecipGen can simulate daily precipitation patterns over the long term. Monthly totals, annual distributions, and extreme events are captured with reasonable accuracy. While the 10-year running average and long-term realizations compare well with historical data, further analysis is ongoing to refine and evaluate the accuracy of these simulations.

The greater peaks and valleys observed in the simulated annual precipitation totals compared to historical records are a natural outcome of the random walk mechanism used in the model. This behavior reflects the accumulation of variability over time and is consistent with stochastic modeling principles. Far from being a limitation, these fluctuations are valuable for exploring a wide range of plausible future scenarios, including rare extremes that may not be represented in historical data. This variability allows users to assess risks and uncertainties, providing critical insights for planning and decision-making in contexts where understanding potential extremes is essential.

Detailed Results for Each Location

In the following sections, I will dive into the details of each location, discussing the specific results and observations for Australia Plains, Grand Junction Walker Field, Dublin, and Orlando.

Australia Plains AU

Australia Plains is a rural locality situated in South Australia, known for its semi-arid climate and variable precipitation patterns. The region experiences seasonal rainfall primarily influenced by subtropical weather systems. Understanding long-term precipitation trends in this area is crucial for managing water resources and supporting agricultural practices in the region.

Monthly Total Precipitation

Annual Total Precipitation Distribution

Peak Day Precipitation Distribution

10-Year Running Average: Percentiles

Sample Realization

To illustrate the model’s long-term simulation capability, I compared a single realization of annual precipitation totals to the historical record. This realization extends into the future and wraps around to include historical patterns.

• Annual Totals: The simulated annual totals over the long term reflect historical variability.

• 5-Year Running Average: A sample realization of the 5-year running average demonstrates how the model represents medium-term variability.

Grand Junction Walker Field, CO

Grand Junction, located on Colorado's Western Slope, lies at the confluence of the Gunnison and Colorado Rivers. The area is characterized by a semi-arid climate with hot summers, mild winters, and relatively low annual precipitation. Precipitation in the region is influenced by both orographic effects from nearby mountain ranges and seasonal monsoonal patterns. Understanding precipitation variability in Grand Junction is essential for water resource management, agriculture, and maintaining ecological balance in this arid environment. 

Monthly Total Precipitation

Annual Total Precipitation Distribution

Peak Day Precipitation Distribution

Long-Term Simulation Results

10-Year Running Average

---

Sample Realization

A single realization of annual precipitation totals provides insight into the model’s long-term simulation dynamics. This realization extends into the future, incorporating historical trends and variability:

• Annual Totals: The simulated annual totals maintain consistency with the historical record over the long term.

 

• 5-Year Running Average: This sample realization captures medium-term variability, demonstrating the model’s ability to simulate smoothed trends over shorter windows.

 

Dublin EI

Background
Dublin, the capital city of Ireland, experiences a temperate maritime climate characterized by relatively mild winters, cool summers, and consistent precipitation throughout the year. Rainfall is influenced by the North Atlantic Drift and westerly weather systems, with variability driven by seasonal patterns and long-term climatic oscillations. Understanding precipitation trends in this region is important for urban water management, flood prevention, and agricultural planning.

Unique from the other sites, I found that the log-normal distribution was a better fit for the daily rain simulation based on annual and peak daily statistics. This is an easy thing to change within PrecipGen. Just go into the Markov Container and edit the Stochastic element, “Intensity_Stoch”.

Monthly Total Precipitation

Annual Total Precipitation Distribution

Peak Day Precipitation Distribution

The distribution of peak daily precipitation rates aligns closely with historical observations, suggesting the use of log-normal distribution instead of gamma might be a good idea.

Long-Term Simulation Results

10-Year Running Average

Sample Realization

A single realization of annual precipitation totals provides insight into the model’s long-term simulation dynamics. This realization extends into the future, incorporating historical trends and variability:

• Annual Totals: The simulated annual totals maintain consistency with the historical record over the long term.

 

• 5-Year Running Average: This sample realization captures medium-term variability, demonstrating the model’s ability to simulate smoothed trends over shorter windows.

Orlando FL

Orlando, located in central Florida, experiences a humid subtropical climate with distinct wet and dry seasons. The region is influenced by tropical weather systems, including summer thunderstorms and the occasional hurricane, which contribute to its high annual precipitation. Understanding precipitation patterns in Orlando is critical for urban planning, flood mitigation, and maintaining the area's extensive network of lakes and wetlands.

Monthly Total Precipitation

Annual Total Precipitation Distribution

Peak Day Precipitation Distribution

Long-Term Simulation Results

10-Year Running Average

Sample Realization

A single realization of annual precipitation totals provides insight into the model’s long-term simulation dynamics. This realization extends into the future, incorporating historical trends and variability:

• Annual Totals: The simulated annual totals maintain consistency with the historical record over the long term.

 

• 5-Year Running Average: This sample realization captures medium-term variability, demonstrating the model’s ability to simulate smoothed trends over shorter windows.

 

Conclusion

Overall, the seasonal patterns and long-term dynamics simulated by PrecipGen have been replicated to a degree that produces reasonably well-matching results when compared to sample historical data. The model's ability to align simulated outcomes with observed precipitation metrics—such as monthly totals, annual distributions, and running averages—demonstrates its potential as a tool for exploring both short-term variability and long-term trends in precipitation behavior. Further refinement of validation methods will continue to enhance confidence in these results.