Hussain et al. (2026) Multi-model drought index for Pakistan's croplands: A data fusion framework and comparative performance analysis

Identification

• Journal: Physics and Chemistry of the Earth Parts A/B/C
• Year: 2026
• Date: 2026-06-09
• Authors: Akash Hussain, HAN Wenting, Muhammad Saleem, Mubashir Ali Ghaffar, Mohsin Ijaz, Wajahat Waseem, Mahmood Hemat, Atiyyah Rafaqat
• DOI: 10.1016/j.pce.2026.104592

Research Groups

• College of Water Resources and Architectural Engineering, Northwest A&F University, China
• Institute of Water-Saving Agriculture in Arid Areas of China, Northwest A&F University, China
• College of Mechanical and Electronic Engineering, Northwest A&F University, China
• Key Laboratory of Agricultural Soil and Water Engineering in Arid and Semiarid Areas of the Ministry of Education, Northwest A&F University, China
• College of Natural Resources and Environment, Northwest A&F University, China

Short Summary

The study develops a Multi-Model Drought Index (MMDI) using a data fusion framework to monitor agricultural drought in Pakistan, finding that a deep learning approach (EEDNN) significantly outperforms traditional and machine learning methods.

Objective

• To develop a robust agricultural drought monitoring framework (MMDI) that integrates multi-source remote sensing and reanalysis data to overcome the limitations of single-variable drought indices in Pakistan's croplands.

Study Configuration

• Spatial Scale: Croplands of Punjab, Sindh, and Khyber Pakhtunkhwa (KPK) provinces, Pakistan.
• Temporal Scale: 23 years (2000–2023).

Methodology and Data

• Models used: Entity Embedding Deep Neural Network (EEDNN), Light Gradient Boosting Machine (LGBM), and Analytical Hierarchy Process (AHP).
• Data sources: MODIS, CHIRPS, and ERA5-Land datasets; validated against the TerraClimate Palmer Drought Severity Index (PDSI).

Main Results

• Model Performance: $\text{MMDI}{\text{EEDNN}}$ demonstrated the highest accuracy with a ROC-AUC of 0.97, followed by $\text{MMDI}{\text{LGBM}}$ (0.83) and $\text{MMDI}_{\text{AHP}}$ (0.75).
• Regional Trends (Sindh): Statistically significant increasing drought trend ($Z < -1.96, p < 0.05$); highest drought frequency (>10 episodes) and intensity (0.33–0.40) in eastern and southern regions.
• Regional Trends (Punjab): Experienced the longest drought durations (>20 months) with moderate to high severity (8.25–9.45) in south-central areas.
• Regional Trends (KPK): Lowest drought occurrence, with only 3–4 episodes and shorter durations (1.2–5.8 months).

Contributions

• Proposes a novel data fusion framework (MMDI) that integrates deep learning, machine learning, and multi-criteria decision-making to improve the precision of agricultural drought monitoring.
• Provides a detailed spatiotemporal analysis of drought frequency, intensity, and duration across Pakistan's primary agricultural provinces, offering actionable data for food security and climate adaptation strategies.

Funding

• Not specified in the provided text.

Citation

@article{Hussain2026Multimodel,
  author = {Hussain, Akash and Wenting, HAN and Saleem, Muhammad and Ghaffar, Mubashir  Ali and Ijaz, Mohsin and Waseem, Wajahat and Hemat, Mahmood and Rafaqat, Atiyyah},
  title = {Multi-model drought index for Pakistan's croplands: A data fusion framework and comparative performance analysis},
  journal = {Physics and Chemistry of the Earth Parts A/B/C},
  year = {2026},
  doi = {10.1016/j.pce.2026.104592},
  url = {https://doi.org/10.1016/j.pce.2026.104592}
}