Integration of a Landsat Time-Series of NBR and Hydrological Modeling to Assess <i>Pinus pinaster</i> Aiton. Forest Defoliation in South-Eastern Spain

Abstract

Climate change is increasing the vulnerability of Mediterranean coniferous plantations. Here, we integrate a Landsat time series with a physically-based distributed hydrological model (Watershed Integrated Management in Mediterranean Environments—WiMMed) to examine spatially-explicit relationships between the mortality processes of Pinus pinaster plantations and the hydrological regime, using different spectral indices of vegetation and machine learning algorithms. The Normalized Burn Ratio (NBR) and Moisture Stress Index (MSI) show the highest correlations with defoliation rates. Random Forest was the most accurate model (R2 = 0.79; RMSE = 0.059), showing a high model performance and prediction. Support vector machines and neural networks also demonstrated a high performance (R2 > 0.7). The main hydrological variables selected by the model to explain defoliation were potential evapotranspiration, winter precipitation and maximum summer temperature (lower Out-of-bag error). These results show the importance of hydrological variables involved in evaporation processes, and on the change in the spatial distribution of seasonal rainfall upon the defoliation processes of P. pinaster. These results underpin the importance of integrating temporal remote sensing data and hydrological models to analyze the drivers of forest defoliation and mortality processes in the Mediterranean climate.

Keywords

• forest disturbance
• <i>pinus</i> plantations
• landsat time-series data
• hydrological model
• machine learning
• defoliation mapping