Improving LAI retrieval in complex mountains with implicit mutual terrain irradiance

Leaf area index (LAI), a key indicator of vegetation canopy structure, is vital to regional ecological assessment and environmental monitoring. However, rugged terrain can induce the radiation distortion recorded by senses, thus resulting in substantial uncertainties in LAI retrievals. Existing canopy reflectance models often ignore the mutual terrain irradiance within soil-canopy layers, thereby limiting the completeness of the physical mechanism. In this study, an LAI retrieval method with physical completeness was proposed tailored for mountainous regions. Radiation interactions within diverse forest scenes were simulated using a discrete anisotropic radiative transfer model to account for the mutual terrain irradiance. Machine learning algorithms were integrated to enhance computational efficiency. The proposed method was validated through typical retrieval algorithms, analyses of topography and canopy effects, and comparisons of LAI retrievals with field observations. The results suggested that the proposed LAI retrieval method, which use the random forest algorithm, performed optimally under various terrain conditions (R 2 = 0.99, RMSE= 0.15, MAE = 0.13). Moreover, the proposed method significantly outperformed models that neglect mutual terrain irradiance (by 0.29, 0.17, and 0.10 in terms of R 2 , RMSE, and MAE, respectively). The method also exhibited strong robustness across varying terrains and canopy structural conditions, accurately capturing the terrain-induced variations in LAI distribution (R 2 = 0.88, RMSE = 0.33, MAE = 0.28) in a real mountainous region. Therefore, considering that the mutual terrain irradiance is crucial in surface radiative transfer modelling, the proposed LAI retrieval method can provide a reliable tool for biophysical parameter extraction over rugged terrains.