Reconstructing pre-fire vegetation condition in the wildland urban interface (WUI) using artificial neural network.

Abstract

Wildfires occurring near and within cities are a potential threat to the population's life and health and can cause significant economic damage by destroying infrastructure and private property. Due to the relatively small area of these wildlands, the accuracy of fire risk-assessment plays a significant role in fire management. Introducing the experience of real events can improve accuracy. But this approach is limited by a lack of knowledge of pre-fire conditions, mainly vegetation characteristics as related to their definition as a fuel parameter because of their high temporal variation. To solve this problem, an Artificial Neural Network (ANN) was designed to reconstruct the spectral characteristics of the vegetation just before the fire with spatial resolution 0.5-2 m from the Landsat image. To test the effectiveness of the proposed methods, the approach has been examined on urban vegetation sites and applied to restore spectral information of the actual vegetation patch before it was burned in 2016 in Haifa, Israel. The results show that the reconstructed RGB image allows for mapping the location of green vegetation with high spatial accuracy. However, spectral data in the visible range have some limitations when it comes to identifying differences between soil and dry plants. The reconstructed image was used to sharp the original data from Landsat. Normalized Difference Vegetation Index maps were produced from the resulting high-resolution multispectral image. The output maps allow to determine the location of vegetation and estimate the level of its dryness on the urban wildland landscape. The proposed method aims to estimate vegetation dryness and, as a result, identify the fuel characteristics at the time of the fire. It has the potential of using for evaluation and improve the weights of the input parameters for the fire-risk assessment and fire-behavior modeling on a specific area.