Estimating uncertainty in flood model outputs using machine learning informed by Monte Carlo analysis - Script and Data version

Necessary data here are used for training and testing the Bayes Neural Network with Bayes by Backprop models - model 1, 2, and 3. There are five big folders/files and two Github files:Model1_classification_proportion folder: Model 1 is used to predict the labelled map identifying where is always, sometimes, and never flooded. This output will be used as an input for the model 2. There are three 3 files corresponding to 5-, 10-, and 20-m resolutions under 7zip format. Please unzip them to see the previous results and to use them.Model2_regression_proportion folder: Model 2 is used to predict the propotion of each pixel being flooded (the pF map) using the output of the model 1 as one of the inputs. There are three 3 files corresponding to 5-, 10-, 20-m resolutions under 7zip format. Please unzip them to see the previous results and to use them. Another file is "Analysis" for formatting the results after predictions.Model3_regression_sd folder: Model 3 is used to predict the standard deviation of maximum flood depth (the sdMWD map). There are three 3 files correspond to 5-, 10-, and 20-m resolutions under 7zip format. Please unzip them to see the previous resutls and to use them. Another file is "Analysis" for formatting the results after predictions.Comparisons_model1_model2.7z: Stores all files used to generate confusion matrices, rmse, and accuracy metrics.Figures_Tables.7zip: Stores all figures an tables used in the publication.Github_FloodUnEn_package.zip: Stores Github source for package named "FloodUnEs" to predict uncertainty using BNNBB models. Please also see here for the online versionGithub_MC_simulation_generation.zip: Stores Github source for generating Monte Carlo simulations. Please also see here for the online version. For the whole dataset that were already generated within this Monte Carlo framework, because it is too large to be uploaded (about 4TB), please contact the author via [e-mail], if necessary.

本数据集所需数据用于训练与测试基于后验贝叶斯方法（Bayes by Backprop）的贝叶斯神经网络（Bayes Neural Network），涵盖模型1、模型2与模型3三类任务。数据集包含5个核心文件夹/文件与2个GitHub代码文件：
1. Model1_classification_proportion文件夹：模型1用于预测标注淹没地图，以识别区域的永久、间歇性及非淹没区域，其输出将作为模型2的输入之一。该文件夹下包含3个分别对应5米、10米、20米分辨率的7zip格式压缩文件，需解压后方可查看历史实验结果并用于后续研究。
2. Model2_regression_proportion文件夹：模型2以模型1的输出作为输入之一，预测每个像素的淹没占比（即pF地图）。该文件夹下包含3个分别对应5米、10米、20米分辨率的7zip格式压缩文件，需解压后方可查看历史实验结果并用于后续研究。另有1个名为"Analysis"的文件，用于对模型预测结果进行格式化处理。
3. Model3_regression_sd文件夹：模型3用于预测最大淹没深度的标准差（即sdMWD地图）。该文件夹下包含3个分别对应5米、10米、20米分辨率的7zip格式压缩文件，需解压后方可查看历史实验结果并用于后续研究。另有1个名为"Analysis"的文件，用于对模型预测结果进行格式化处理。
Comparisons_model1_model2.7z：存储用于生成混淆矩阵、均方根误差（RMSE）及准确率评估指标的全部相关文件。
Figures_Tables.7zip：存储论文发表所需的全部图表文件。
Github_FloodUnEn_package.zip：存储名为"FloodUnEs"的代码包源码，该包可基于贝叶斯后验神经网络模型实现不确定性预测，其在线版本可通过对应GitHub仓库获取。
Github_MC_simulation_generation.zip：存储用于生成蒙特卡洛（Monte Carlo）模拟的代码源码，其在线版本可通过对应GitHub仓库获取。
此外，基于该蒙特卡洛框架生成的完整数据集体积过大（约4TB），无法直接上传，若有相关需求可通过邮箱[e-mail]联系作者获取。