CDSim: A Synthetic Climate Data Simulation Framework for Temperature and Rainfall Modelling in R

Introduction

Climate and weather datasets underpin a wide range of scientific and applied research, including environmental modelling, hydrological forecasting, agricultural planning, ecological management, and climate risk assessment [1, 2, 3]. Long-term, continuous records of temperature and precipitation are particularly critical for validating climate models, assessing climate variability and change, and supporting data-driven approaches such as statistical analysis and machine learning-based prediction systems [4, 5]. In educational settings, these datasets also serve as essential resources for teaching climate data processing, modelling workflows, and reproducible research practices.

Despite the growing availability of global and regional climate products from sources such as ERA5, NASA POWER, NOAA, and satellite-based rainfall datasets, practical barriers to their use remain [6, 7]. Many regions—particularly in developing countries—continue to experience incomplete station records, data gaps, or restricted access to high-quality observational data [8, 9]. In addition, existing datasets often require substantial preprocessing, involve complex file formats, or present licensing and data-sharing constraints that limit their reuse in open research and teaching environments. Spatial and temporal resolution mismatches further reduce their suitability for localised studies or classroom applications.

These challenges are especially pronounced for students and early-career researchers, who frequently lack access to ready-to-use climate datasets for learning analytical pipelines, testing statistical or machine learning methods, or developing reproducible workflows. As a result, methodological experimentation and teaching often rely on fragmented or ad hoc data sources.

Synthetic climate datasets offer a pragmatic and increasingly accepted alternative in such contexts [10]. When transparently generated and evaluated against known climatological behaviour, synthetic time series can support experimentation, benchmarking, education, and method development without reproducing proprietary observations or violating data-sharing restrictions [11, 12]. They enable controlled simulations, facilitate reproducibility, and lower barriers to entry for climate data analysis.

In response to these needs, this study introduces CDSim, an R-based simulation framework designed to generate synthetic yet climatologically realistic monthly time series data for key climate variables, including temperature and rainfall, across multiple automated or user-defined stations. CDSim provides a flexible, lightweight, and reproducible environment for generating climate data tailored to specific temporal spans and spatial contexts. By supporting experimentation, workflow testing, and educational use, CDSim aims to enhance climate data accessibility and methodological transparency, particularly in data-constrained regions.

Methods and model design

The CDSim framework generates synthetic monthly climate time series by combining deterministic seasonal structure with stochastic variability. For each station s and month t, the mean temperature follows a harmonic seasonal signal,

where μ(s) and A(s) are station-specific baseline and amplitude parameters, and σ = 0.5. Minimum and maximum temperatures are generated as correlated series around this mean,

ensuring realistic diurnal separation while preserving seasonal coherence.

Monthly rainfall totals are simulated using a gamma distribution,

where the scale parameter θ(t,s) varies seasonally to represent monsoon intensity and station-level variability, while the shape parameter k controls precipitation intermittency and skewness. This formulation follows established stochastic weather generator principles [11, 12].

All stochastic components are fully reproducible through controlled random seed initialisation. The framework does not replicate observed time series or extreme events but prioritises realistic statistical behaviour and long-term structure for methodological testing, education, and benchmarking.

Implementation and architecture

CDSim is implemented as an R package within the R statistical computing environment [13], providing a modular and reproducible framework for generating synthetic monthly climate time series.

The core simulation function accepts station metadata (latitude, longitude, and station identifier), a user-defined temporal range, and an optional random seed, and returns station-specific series for temperature and rainfall. Seasonal signals and stochastic components are computed internally and combined into structured station-level time series. Outputs are provided in both comma-separated values (CSV) and Network Common Data Format (NetCDF), ensuring compatibility with standard climate analysis, modelling, and data-sharing workflows. Table 1 summarises the key CDSim functions and their primary usage.

CDSim follows a modular architecture consisting of four primary components: (i) station metadata generation, (ii) climate simulation engine, (iii) visualisation utilities, and (iv) export interfaces. The simulation engine integrates deterministic seasonal signal generation with stochastic sampling modules. Station metadata are passed to the simulation layer, which constructs temperature and rainfall series using vectorised computations. The resulting datasets are stored in structured data frames before being routed to plotting or export functions.

The architecture separates data generation from input/output operations, enabling flexible reuse and future extension. Visualisation is implemented independently using the ggplot2 package [14], while NetCDF export relies on the ncdf4 [15] dependency.

Computationally, CDSim is lightweight and memory-efficient for typical research and teaching scenarios. Memory usage scales linearly with the number of stations and simulated years. Empirical testing indicates that simulating 50 stations over 50 years of monthly data remains well within typical desktop memory limits (under approximately 100 MB of RAM). Processing time is minimal due to vectorised operations, with typical simulations completing within seconds for moderate station counts (<100 stations over 50 years). The package is not optimised for high-resolution daily simulations or very large-scale spatial grids, which may require additional memory management strategies.

Quality control

To ensure reliability and reproducibility, CDSim incorporates multiple levels of quality control aligned with CRAN submission standards and open research software practices.

Validation and evaluation

The CDSim outputs were evaluated to ensure realistic and coherent climate behaviour. Seasonal consistency was assessed by confirming that simulated temperature and rainfall series reproduce expected annual cycles and timing. Key statistical properties, including central tendency, variability, and distributional shape, were examined to verify climatological plausibility, with temperature showing smooth seasonal variability and rainfall exhibiting positively skewed behaviour.

Simulated series were further compared against observational envelopes derived from Ghana Meteorological Agency data and published literature on Ghanaian climate and the West African monsoon. Although the synthetic data do not replicate observed values, they remain within realistic climatological ranges.

Limitations are acknowledged: the dataset does not represent actual historical events or extremes and is not intended for operational or policy use. Instead, the evaluation focuses on statistical realism and long-term structure, supporting methodological testing, education, and reproducible research.

Application example

In this section, we demonstrate how the CDSim package can be used to simulate climate data. Figure 1 provides an overview of the CDSim workflow pipeline, while Figure 2 presents an example implementation carried out in the RStudio environment.

Figure 1

Figure 2

Station metadata are first defined, followed by simulation of temperature and rainfall time series using deterministic seasonal signals and stochastic variability. The resulting datasets are exported in CSV and NetCDF formats and can be visualised using built-in plotting utilities, which are implemented using the ggplot2 package [14].

Figures 3, 4, 5 present the simulated temporal patterns of maximum temperature, minimum temperature, and rainfall for the Accra station, generated by the CDSim package and visualised using the built-in plotting functions as part of the demonstration workflow.

Figure 3

Figure 4

Figure 5

Conclusion and future development

CDSim offers a lightweight and accessible framework for generating realistic synthetic climate datasets, addressing common challenges related to data availability, licensing restrictions, and reproducibility in climate research and education. By combining simple seasonal structure with stochastic variability, the framework enables users to prototype analytical workflows, test modelling approaches, and develop teaching materials in a controlled and transparent environment.

Future development of CDSim will focus on extending its realism and flexibility. Planned enhancements include explicit spatial correlation modelling using approaches such as Gaussian processes, improved representation of extremes and rare events, and integration with machine learning-based climate generators, including variational autoencoders and related architectures. These extensions will further broaden the applicability of CDSim for methodological research while maintaining its emphasis on reproducibility and open access.

Operating system

Platform independent

Programming language

R Version 4.5.1 or higher

Additional system requirements

2 GB RAM, 5 GB HDD or SSD

Dependencies

ggplot2 and ncdf4

List of contributors

Isaac Osei, Acheampong Baafi-Adomako, and Sivaparvathi Dusari

Software location

Archive (e.g., institutional repository, general repository) (required—please see instructions on journal website for depositing archive copy of software in a suitable repository)

• Name: Comprehensive R Archive Network (CRAN)

• Persistent identifier: 10.32614/CRAN.package.CDSim

• Licence: MIT

• Publisher: Isaac Osei

• Version published: 0.1.1

• Date published: 15/12/25

Code repository (e.g., SourceForge, GitHub etc.) (required)

• Name: GitHub

• Identifier: https://github.com/ikemillar/CDSim

• Licence: MIT

• Date published: 25/11/25

Language

English