Aslan et al. (2025) Hybrid wavelet–ANN modelling for LAI forecasting under climatic variability: comparative case studies from the mediterranean basin

Identification

• Journal: Acta Geophysica
• Year: 2025
• Date: 2025-12-17
• Authors: Zafer Aslan, Buket İşler, Gamze M. Müftüğolu, Enrico Feoli
• DOI: 10.1007/s11600-025-01761-9

Research Groups

• Department of Computer Engineering, Faculty of Engineering, Istanbul Aydın University, Istanbul, Türkiye
• Department of Software Engineering, Faculty of Engineering, Istanbul Topkapı University, Istanbul, Türkiye
• Department of Software Engineering, Faculty of Engineering, Istanbul Aydın University, Istanbul, Türkiye
• Department of Life Sciences, University of Trieste, Trieste, Italy
• Abdus Salam International Centre for Theoretical Physics (ICTP), Trieste, Italy (acknowledged for support)

Short Summary

This study developed and evaluated a hybrid Wavelet–Artificial Neural Network (W-ANN) model for forecasting Leaf Area Index (LAI) in the Mediterranean Basin under climatic variability. The W-ANN model significantly outperformed a conventional Artificial Neural Network (ANN), demonstrating 15–85% higher accuracy and revealing divergent LAI trends across four contrasting urban locations.

Objective

• To propose and evaluate a hybrid Wavelet–Artificial Neural Network (W-ANN) model that integrates wavelet decomposition with ANN to improve Leaf Area Index (LAI) forecasting accuracy.
• To benchmark the W-ANN model against a conventional ANN approach using harmonized climate and vegetation data across different Mediterranean locations.
• To assess the predictive performance of the W-ANN model in capturing LAI dynamics under complex, multi-scale, and nonlinear climatic influences.

Study Configuration

• Spatial Scale: Four urban locations in the Mediterranean Basin: Antalya and Istanbul (Kandilli) in Türkiye, and Enna (Sicily) and Trieste in Italy. These sites represent a spectrum of topographic and climatic contexts, including coastal lowlands, continental interiors, and maritime–continental transitional zones.
• Temporal Scale:
  • Data period: Monthly MODIS-based LAI data from 2007 to 2023. Monthly meteorological data from 2007 to 2023 (Antalya, Kandilli, Trieste) and 2010 to 2023 (Enna).
  • Forecasting period: Monthly LAI forecasts for 2024–2030.

Methodology and Data

• Models used:
  • Hybrid Wavelet–Artificial Neural Network (W-ANN) model: Integrates Discrete Wavelet Transform (DWT) using Daubechies wavelet family (for input decomposition) with a feedforward Artificial Neural Network (ANN).
  • Conventional Artificial Neural Network (ANN) model: Feedforward ANN applied directly to raw data.
  • Wavelet Transform: Discrete Wavelet Transform (DWT) for input variables (Daubechies wavelet, 3 decomposition levels: d1, d2, d3); Continuous Wavelet Transform (CWT) for LAI time series analysis (Mexican Hat wavelet).
  • ANN architecture: Feedforward neural network with a single hidden layer (8 neurons for conventional ANN; 3, 6, 9, or 12 neurons, optimized per site, for W-ANN). ReLU activation in hidden layer, linear activation in output layer.
  • Optimizers: Levenberg–Marquardt (initial optimization), Adam (training).
• Data sources:
  • Leaf Area Index (LAI): Moderate Resolution Imaging Spectroradiometer (MODIS) Terra product (MOD15A2H), 500 m spatial resolution, 8-day temporal resolution, aggregated to monthly values. Data from NASA LP DAAC.
  • Meteorological data: Ground-based observations from the nearest official weather stations, aggregated to monthly values. Variables include total monthly precipitation (mm) and monthly near-surface air temperature (°C).

Main Results

• The hybrid W-ANN model consistently outperformed the conventional ANN model in LAI forecasting.
  • W-ANN achieved 15–85% higher accuracy compared to ANN.
  • Mean Squared Error (MSE) values for W-ANN on test datasets ranged from 0.01 to 0.04.
  • The W-ANN model improved the average Pearson correlation coefficient (R) between observed and predicted LAI from approximately 0.60 to 0.90. Notably, for Enna, ANN yielded R ≈ 0.11, while W-ANN achieved R ≈ 0.91.
• Wavelet decomposition revealed that high-frequency (d1, monthly) and medium-frequency (d2, seasonal to annual) precipitation components were strongly correlated with LAI fluctuations (R > 0.50), indicating vegetation's sensitivity to intra-annual rainfall distribution. Low-frequency (d3, multi-year) components showed negligible influence (R < 0.22).
• Scenario simulations for 2024–2030 projected divergent LAI trends:
  • A declining trend in LAI was observed for Antalya and Enna, suggesting potential decreases in vegetation productivity due to increased drought frequency and reduced precipitation.
  • An increasing or stable trend in LAI was projected for Istanbul (Kandilli) and Trieste, possibly driven by milder winters, earlier greening, and local micro-environmental factors.
• Site-specific LAI characteristics: Kandilli (Istanbul) exhibited the highest average LAI (~1.82), while Antalya recorded the lowest (~0.77). Trieste showed the highest LAI variability (coefficient of variation ≈ 0.65), and Antalya the lowest (coefficient of variation ≈ 0.20).

Contributions

• Applied a novel hybrid W–ANN framework for LAI modeling across four diverse Mediterranean regions, addressing a gap in the spatial application of such models.
• Constructed a harmonized dataset comprising MODIS-based LAI (2007–2023) and co-located meteorological observations (precipitation and temperature).
• Evaluated model performance using both historical validation and scenario-based projections for the period 2024–2030.
• Provided operational insights for developing vegetation-based decision support systems in semi-arid and climate-sensitive regions, informing land use planning, irrigation scheduling, erosion control, and climate adaptation strategies.

Funding

• Abdus Salam International Centre for Theoretical Physics (ICTP) under the Simons Associate-ship Foundation.

Citation

@article{Aslan2025Hybrid,
  author = {Aslan, Zafer and İşler, Buket and Müftüğolu, Gamze M. and Feoli, Enrico},
  title = {Hybrid wavelet–ANN modelling for LAI forecasting under climatic variability: comparative case studies from the mediterranean basin},
  journal = {Acta Geophysica},
  year = {2025},
  doi = {10.1007/s11600-025-01761-9},
  url = {https://doi.org/10.1007/s11600-025-01761-9}
}