DrAhmedKheir/Egypt-Maize-APSIM-ML-Downscaling-Pipeline

This script integrates APSIM Next Generation, climate data, soil information, and machine learning to generate high-resolution (0.1°) maize yield maps for Egypt, including uncertainty and climate-change scenarios. 1️⃣ Environment Setup rm(list=ls()) options(stringsAsFactors = FALSE) 🔹 Ensures a clean R session 🔹 Avoids accidental reuse of variables from previous runs 2️⃣ Packages pkgs <- c("apsimx","data.table","dplyr","sf","foreach","doParallel", "ranger","FNN","ggplot2","DBI","RSQLite","rnaturalearth", "jsonlite","httr") Why each package is needed Package Role apsimx Run APSIM NG & edit APSIMX files DBI, RSQLite Read APSIM .db outputs data.table, dplyr Fast data manipulation sf, rnaturalearth Spatial grids & Egypt boundary foreach, doParallel Parallel APSIM runs ranger Random Forest ML FNN k-nearest-neighbor interpolation httr, jsonlite SoilGrids API ggplot2 Visualization 3️⃣ User Settings (Paths & Controls) base_dir <- "D:/APSIMICARDATraining/AnotherFullScript" work_dir <- file.path(base_dir, "EG_Maize_CMIP6_FULL") coarse_dir <- file.path(work_dir, "coarse") Folder Logic EG_Maize_CMIP6_FULL/ ├─ coarse/ │ ├─ EG_0001/ │ │ ├─ MaizeFull.apsimx │ │ ├─ EG_0001.met │ │ └─ MaizeFull.db (created by APSIM) ├─ outputs_master/ └─ _cache/ This structure is critical — APSIM expects the .apsimx and .met files inside each site folder. 4️⃣ Helper Functions Logging msg("Running APSIM…") Adds timestamps so participants can see progress. Numeric Safety to_na() impute_median() Guarantees: No crashes from Inf, NaN ML models never fail due to missing data 5️⃣ Indexing Coarse APSIM Sites site_dirs <- list.dirs(coarse_dir, recursive=FALSE) ✔ Automatically discovers all EG_#### folders ✔ Builds a run index that controls the entire pipeline This design allows: 10 sites or 1,000 sites No hard-coded filenames 6️⃣ APSIM Report Configuration (Key Stability Step) report_vars <- c( "[Clock].Today as Date", "[Maize].Grain.Total.Wt*10 as Yield", "[Maize].Leaf.LAI as LAI", "[Maize].AboveGround.Wt as Biomass" ) Why this works Uses existing Report node No JSON hacking Event = EndOfDay (universally safe) Aliases columns so DB tables are predictable This is the main reas …