A genetic algorithm-optimized ik-NN approach for forest biophysical parameter estimation using bi-seasonal ALOS-2 PALSAR-2 data

This study investigates the seasonal effects of fully polarimetric L-band SAR data on forest parameter estimation in the Hyrcanian mixed deciduous forests, a UNESCO World Heritage site. Bi-temporal ALOS-2 PALSAR-2 data acquired in April (dry season, leaf-on) and January (wet season, leaf-off) were analyzed together with field measurements from 79 forest inventory plots in temperate Hyrcanian forests. To improve estimation accuracy, we propose a non-parametric improved k-nearest neighbour (ik-NN) method, optimized through a Genetic Algorithm (GA) for feature weighting. The results indicate that forest parameters are estimated more accurately using dry-season PALSAR-2 data than wet-season data. The GA-optimized ik-NN method consistently outperformed traditional linear regression models, demonstrating its ability to handle the complex, non-linear relationships between SAR backscatter and forest biophysical variables. Using dry-season data, the relative RMSE percentage ranged from 22% for mean diameter to 46% for Above-Ground Biomass (AGB) estimation. The correlation analyses indicated smaller sensitivity of HV-Winter scene than HV-Spring scene to forest variables, while VV-Winter and VV-Spring backscatters were at the same level. The results could be explained in terms of the different dielectric properties of the forest in spring and winter, both that of the forest floor and trees, while the ik-NN framework provides a robust computational tool for multi-seasonal SAR analysis.