Evaluating the Applicability of RiSIM for AI-Based River Plastic Monitoring to an urban river in Indonesia

Monitoring floating plastic transport in rivers is essential for quantifying plastic flux and guiding pollution mitigation strategies. While traditional approaches relied on manual collection or visual observation, recent advancements have increasingly adopted image-based methods that integrate deep learning with remote sensing technologies, including satellite imagery, UAVs, and fixed-point cameras. Although visual observation remains widely used for global-scale assessment, it is constrained by high labor demands, observer subjectivity, and safety risks during flood events.

To address these limitations, Kataoka et al. (2025) developed RiSIM (River Surface Image Monitoring software), which utilizes fixed river cameras and deep learning models for plastic detection, classification, and object tracking for floating debris. RiSIM demonstrated high reliability in Japanese river systems (r = 0.91 for quantity; r = 0.80 for mass). However, its performance has so far been evaluated exclusively within Japan.

As described in Kataoka et al. (2025), a cloud-based monitoring platform, PRIMOS, has been released to facilitate the application of RiSIM. Operating through a standard web browser with server-side computation, PRIMOS eliminates technical barriers such as local environment setup and high-performance hardware requirements. By integrating the fine-tuning capabilities examined in this study, the platform aims to support researchers and monitoring projects in conducting plastic transport analyses across diverse river systems worldwide.

This study evaluates the global applicability of RiSIM using nadir-view video data collected from Saluran Cideng (a tributary of Kali Cideng) in Jakarta, Indonesia—a first step toward assessing its transferability to Southeast Asian rivers. Using the PRIMOS platform, we evaluate the detection performance of the AI model integrated into RiSIM at Saluran Cideng. Furthermore, we examine methodology for fine-tuning and retraining to enhance the system's applicability to the local environment. The broader applicability of the framework and practical considerations for deployment will be discussed.

The monitoring data for this RiSIM evaluation was collected under the “Project on Inventory Development Methodology for a Plastic Leakage into the Environment, Including the Marine Environment”, commissioned by the Ministry of the Environment, Japan.