A downscaled soil moisture agricultural drought index (dSMADI) through spatiotemporal fusion

Overview

This repository accompanies the study “Assessing how spatiotemporal fusion enables an agricultural drought index to represent ecosystem evaporative stress dynamics at continent-to-field scales” (Yu et al., 2026). It contains the Australian paper workflow and a newer experiment that tests dSMADI transferability at flux-tower sites in Switzerland, the United States, and China.

The repository is organized as a basic research profile. It is a paper companion with staged experiments, external scientific data, and Gadi-specific production workflows; it is not an installable software package or a supported public API.

Workflow status

Workflow	Status	Scope
Australian experiments	Reference-only	Published/reference processing scripts for continental, local, OzFlux, ECOSTRESS, and component-attribution analyses. Many scripts retain institutional paths and require data that are not redistributed.
AU flux-tower ESI pipeline	Runnable with external data	Tested Python subworkflow for deriving tower ET/PET stress metrics from OzFlux and FLUXNET archives.
Global transferability	Runnable with external data on Gadi	Maintained download, preparation, fusion, dSMADI, evaluation, plotting, audit, and PBS workflow for CH-Oe2, US-Mo3, and CN-Dmn.
Figure reproduction	Runnable with external processed products	Separate AU and global figure-script collections. Generated figures are not tracked by default.

Repository structure

dSMADI/
├── 0_ancillary/
│   └── au/                         Small Australian site metadata and study mask
├── 1_smadi_design/                 dSMADI formulation and implementation pointers
├── 2_experimental_scripts/
│   ├── au/                         AU reference workflow and flux-tower validation
│   └── global_transferability/     Maintained three-site Gadi workflow and tests
├── 3_figure_scripts/
│   ├── au/                         Australian paper figures
│   └── global_transferability/     Global transferability figures and map inputs
├── docs/
│   ├── history/                    Historical repository-scaffold documents
│   └── migration/                  Old-to-new path mapping and provenance notes
├── figures/                        Curated-figure boundary; generated drafts are ignored
├── preprint/                       ESS Open Archive manuscript PDF
├── CITATION.cff                    GitHub citation metadata
├── LICENSE                         MIT license
└── README.md                       Public repository map

Scientific workflow

The experiments combine monthly climatology-normalized soil moisture, land-surface temperature, and lagged vegetation condition. Spatiotemporal fusion supplies the high-resolution LST and NDVI inputs used to calculate dSMADI. The Australian workflow evaluates the index from continental to field scales; the global workflow tests whether the same construction transfers to three non-Australian flux sites. See the formulation note for equations and implementation links.

Quick validation

Run commands from the repository root. On Gadi, parsing and path inspection are lightweight login-node checks; run the test suites and scientific processing inside a PBS allocation.

# Python syntax
python -m compileall -q \
  2_experimental_scripts/global_transferability \
  3_figure_scripts/global_transferability

# R syntax
find 2_experimental_scripts/global_transferability \
     3_figure_scripts/global_transferability \
     -name '*.R' -print0 | \
  xargs -0 -n 1 /apps/R/4.5.0/bin/Rscript -e \
  'args <- commandArgs(TRUE); parse(file = args[[1]])' --args

# PBS and shell syntax
find 2_experimental_scripts -type f \
  \( -name '*.pbs' -o -name '*.sh' \) -print0 | \
  xargs -0 -n 1 bash -n

The full global test commands and Gadi prerequisites are documented in the global workflow README.

Data and outputs

Large inputs and generated rasters are intentionally external to Git.

Location	Role	Durability
0_ancillary/au/	Small redistributable repository inputs	Tracked
/scratch/os22/yy4778/agri_drought/global	Raw downloads, prepared rasters, gap-filled/fused products, models, indices, and inventories	Purgeable; useful for restart and audit, not an archive
/g/data/ym05/agri_drought/global_transferability	Final global tables, provenance, audits, and durable results	Persistent project storage
/g/data/ym05/agri_drought/figures/global_transferability	Generated global figures	Persistent project storage

The existing scratch processing tree is useful because the global scripts are restartable and validate existing products before skipping them. Do not copy that tree into this repository or treat it as the only durable record. Preserve final outputs and provenance under /g/data.

Limitations

• AU scripts preserve the original scientific sequence but are not normalized into a single portable command. Absolute paths and historical HPC assumptions remain visible as provenance.
• The global workflow is configured for the present NCI Gadi checkout and external Earth Engine, FLUXNET, ERA5-Land, MODIS, VIIRS, SMAP, Landsat, emissivity, and ubESTARFM resources.
• The repository does not redistribute the large source datasets or generated paper products.
• Figure scripts require processed products and, in some cases, external map layers and Arial fonts.

See the repository-layout migration note for source mappings and intentionally omitted generated material.

Preprint and citation

The tracked manuscript is preprint/Yu-2026-ESSOAr-Preprint-15001424v1.pdf. Cite the preprint as:

@article{Yu2026_dSMADI_ESSOAr,
  author  = {Yi Yu and Tim R. McVicar and Thomas G. Van Niel and Luigi J. Renzullo and Siyuan Tian and Brendan P. Malone and Jian Peng and Patrick Filippi and Thomas F. A. Bishop},
  title   = {Assessing how spatiotemporal fusion enables an agricultural drought index to represent ecosystem evaporative stress dynamics at continent-to-field scales},
  journal = {ESS Open Archive},
  year    = {2026},
  doi     = {10.22541/essoar.15001424/v1}
}