Neural network-based modelling for forest biomass assessment

Abstract

Forest biomass is an important parameter for assessing the status of forest ecosystems. In the present study, forest biomass was assessed by integrating remotely-sensed satellite data and field inventory data using an artificial neural network (ANN) technique in Barkot forest, Uttarakhand, India. Spectral and texture variables were derived from Resourcesat-1 (RS1) LISS-III (Linear Imaging Self-Scanning Sensor) data of April 24, 2013. ANN was used for finding the relation of spectral and texture variables to field-measured biomass. The top 10 variables, namely shortwave infrared (SWIR) band reflectance, near infrared (NIR) band reflectance, normalized difference vegetation index (NDVI), difference vegetation index (DVI), green band contrast, green band variance, SWIR band contrast, NIR band dissimilarity, SWIR band second angular moment, and red band mean, were selected for generating a multiple linear regression model to predict the biomass. The predicted biomass showed a good relationship (R2 = 0.75 and root mean square error (RMSE) = 85.32 Mg ha−1) with field-measured biomass. The model was validated yielding R2 = 0.74 and RMSE = 93.41 Mg ha−1. The results showed that RS1 LISS-III satellite data have good capability to estimate forest biomass, and the ANN technique can be used to enhance the scope of biomass estimation with a minimum number of spectral and texture variables.

Keywords

• forest biomass
• resourcesat-1 liss-iii
• spectral variables
• texture variables
• artificial neural network (ann)