Data for manuscript "Emulator-based calibration of a dynamic grassland model using recurrent neural networks and Hamiltonian Monte Carlo" by Aakula et al.

Data and python code for the manuscript "Emulator-based calibration of a dynamic grassland model using recurrent neural networks and Hamiltonian Monte Carlo", for performing emulator hyperparameter optimization and training. Python file optimize_LSTM_emulator.py can be used either for training an LSTM emulator with predefined hyperparameters or to optimize hyperparameters from a given hyperparameter space. The training data for each fold is included in the files of shape training_data_fold_{}.parquet. The data is obtained from model simulations, including model inputs (meteorological forcings obtained from ERA5 data), model parameters (sampled from distributions defined in the manuscript) and model (BASGRA) outputs. Text file examples.txt gives instructions and examples on running the script.

本数据集配套论文手稿《基于循环神经网络与哈密顿蒙特卡洛的动态草地模型模拟器校准》（Emulator-based calibration of a dynamic grassland model using recurrent neural networks and Hamiltonian Monte Carlo），用于开展模拟器超参数优化与模型训练工作。Python脚本"optimize_LSTM_emulator.py"既可用于以预设超参数训练长短期记忆网络（Long Short-Term Memory, LSTM）模拟器，也可基于给定超参数空间完成超参数优化任务。各折训练数据存储于命名格式为"training_data_fold_{}.parquet"的Parquet文件中。该数据集源自模型模拟结果，包含模型输入项（取自ERA5的气象强迫数据）、模型参数（从手稿中定义的分布中采样得到）以及BASGRA模型的输出结果。文本文件"examples.txt"提供了该脚本的运行说明与使用示例。