jaza (2025) Wetland Areas Trend and Examining Effective Factors with Machine Learning

Identification

• Journal: Kirkuk University Journal For Agricultural Sciences
• Year: 2025
• Date: 2025-12-22
• Authors: jaza
• DOI: 10.58928/ku25.16430

Research Groups

• Surveying Department, Darbandikhan Technical Institute, Sulaimani Polytechnic University, Sulaimani, KGR, IRAQ
• Civil Engineering and Architecture Faculty, Shahid Chamran University of Ahvaz, Ahvaz, IRAN

Short Summary

This study analyzed long-term water level trends in Hammar Marsh, Iraq (2000-2025) using remote sensing and machine learning to identify key environmental drivers. It revealed an overall increasing trend in water levels, with the Palmer Drought Severity Index (PDSI) and soil moisture identified as the dominant controlling factors.

Objective

• To examine the long-term trends (2000-2025) of water levels in Hammar Marsh, Iraq, determining if levels are rising or falling over time.
• To identify and quantify the leading environmental drivers influencing these water level trends using advanced machine learning techniques.

Study Configuration

• Spatial Scale: Hammar Marsh, Iraq, one of the Mesopotamian marshes (historically up to 4,500 square kilometers).
• Temporal Scale: 2000 to 2025 (26 years), with monthly data analysis.

Methodology and Data

• Models used:
  • Mann–Kendall test (for monotonic trend detection)
  • Sen's slope estimator (for trend rate quantification)
  • Stepwise Variance Inflation Factor (VIF) analysis (for multicollinearity reduction)
  • Random Forest (RF) model (for simulating wetland water level behavior, performance evaluation, and variable importance estimation)
• Data sources:
  • Platform: Google Earth Engine (GEE)
  • Satellite Imagery:
    • MODIS MOD09A1 (Normalized Difference Water Index - NDWI, for water surface area)
    • MODIS MOD13Q1 (Normalized Difference Vegetation Index - NDVI)
    • MODIS MOD11A2 (Land Surface Temperature - LST)
    • MCD19A2 (Aerosol Optical Depth - AOD)
  • Reanalysis/Climate Datasets:
    • CHIRPS (Precipitation - P)
    • TERRACLIMATE dataset (Actual Evapotranspiration - AET, Reference Evapotranspiration - PET, Wind Speed - WS, Vapor Pressure - VAP, Soil Moisture - SM, Palmer Drought Severity Index - PDSI, Runoff - R)
  • Temporal Resolution: All datasets resampled to monthly frequency.
  • Spatial Resolution: Varies by sensor (e.g., 250 meters, 500 meters, 1000 meters, 4638 meters, 5566 meters).

Main Results

• An overall increasing trend in water levels was observed in Hammar Marsh from 2000 to 2025.
• Seasonal variations showed the strongest positive trends in Fall (Kendall's Tau = 0.413, p = 0.0034; Sen's slope = 924 hectares) and Summer (Kendall's Tau = 0.373, p = 0.0085; Sen's slope = 1,222 hectares). Winter and early spring exhibited weaker or less significant increases.
• Monthly analysis indicated the highest positive Kendall's Tau values in September (0.46), June (0.42), and July (0.393), with the steepest monthly increases in March (1,698 hectares), April (1,694 hectares), and February (1,556 hectares).
• Stepwise VIF analysis led to the removal of Actual Evapotranspiration (AET) and Reference Evapotranspiration (PET) due to high multicollinearity; all remaining variables had VIF values below 10.
• The Random Forest model demonstrated strong predictive performance with a test set R² of 0.690 and an RMSE of 0.154.
• Variable importance analysis identified the Palmer Drought Severity Index (PDSI, importance = 0.302) and soil moisture (SM, importance = 0.250) as the dominant controlling factors of water level change.
• Vegetation cover (NDVI, importance = 0.110) and land surface temperature (LST, importance = 0.102) were also important drivers.
• Other variables, including vapor pressure (VAP, 0.074), precipitation (P, 0.061), aerosol optical depth (AOD, 0.041), wind speed (WS, 0.040), and runoff (RO, 0.021), had secondary but measurable contributions.

Contributions

• Provides a comprehensive, long-term (26-year) assessment of water level dynamics and their environmental drivers for Hammar Marsh, a critical Ramsar site in Iraq.
• Demonstrates an effective integration of multi-temporal remote sensing (Google Earth Engine), non-parametric trend analysis, and machine learning (Random Forest) for robust wetland monitoring and understanding complex ecohydrological processes.
• Quantifies the relative importance of various hydro-climatic and ecological factors, specifically identifying PDSI and soil moisture as primary drivers, which is crucial for targeted conservation and management strategies.
• Offers valuable insights for sustainable management and conservation efforts in Hammar Marsh and similar wetland ecosystems, particularly emphasizing drought mitigation and soil moisture monitoring.

Funding

Not explicitly mentioned in the paper.

Citation

@article{jaza2025Wetland,
  author = {jaza},
  title = {Wetland Areas Trend and Examining Effective Factors with Machine Learning},
  journal = {Kirkuk University Journal For Agricultural Sciences},
  year = {2025},
  doi = {10.58928/ku25.16430},
  url = {https://doi.org/10.58928/ku25.16430}
}