Selective fishing is increasingly required to mitigate bycatch and overfishing. Many studies have developed selective fishing gears, and some have been implemented at sea. In addition to gear design, fish behaviour plays a critical role in selectivity because fish behaviour and fishing gear dynamics are closely interrelated during operation. While most previous light-based approaches have simply focused on attractive or evasive responses, our longer-term goal is to construct desired three-dimensional swimming paths by leveraging light-evoked behavioural responses as a basis for practical trajectory steering. Here, as a first step, we quantify and parameterise these responses and estimate governing parameters using experiments, 3D modelling and data assimilation. We extend this line of work by embedding quantified responses into predictive three-dimensional schooling models. First, we conducted tank experiments under several light conditions to quantify behavioural responses to light stimuli and to collect trajectory data for parameter estimation via data assimilation. The observations suggested that the effectiveness of light-induced responses may depend on species and behavioural characteristics. Second, we developed a three-dimensional simulation model of schooling that explicitly incorporates behavioural responses to light stimuli. Third, we estimated unknown parameters governing individual behaviour by assimilating observed trajectories into the model. Non-light-related parameters were first estimated using trajectories obtained under non-light conditions to validate the model; subsequently, parameter describing responses to light stimuli was estimated using trajectories recorded under illuminated conditions. This model framework and the estimated results provide insights into feasibility and strategies for controlling collective fish behaviour towards selective fishing.