Despite the growing need for formalized frameworks to model socio-ecological systems (SES), no existing approach adheres to the Object Management Group (OMG) standards for structuring and simulating their complexity. Agro-sylvo-pastoral systems exemplify SES where micro-level agent behaviors shape macro-level dynamics, while policy interventions influence local decision-making. However, existing modeling approaches lack a standardized structure to integrate artificial intelligence (AI) algorithms within agents, limiting their ability to adapt to macro-level policy changes and optimize decisions dynamically. The absence of a formal metamodel further restricts interoperability, reusability, and systematic policy evaluation in SES. In this study, we present Ecological System Modeling Language (EcoSysML), a novel metamodel for SES designed in accordance with OMG standards to formally define their structure and dynamics. A key innovation of EcoSysML is the LearningModel, which enables the seamless integration of AI-driven decision-making within agents, allowing them to learn, adapt, and optimize their strategies over time. By leveraging fundamental concepts (Actor, Activity, Resource, ExternalProcess, and PolicyProduct), the metamodel provides a coherent and structured representation of SES, explicitly defining how agents interact with their environment, manage resources, and respond to policy interventions. By bridging agent-based modeling with artificial intelligence, EcoSysML establishes a scalable, interoperable foundation for SES simulation and AI-driven policy design.