New AI tool combines GTAP and APSIM to assess shocks in agricultural supply chains

In July 2026, researchers published a preprint on arXiv of a new AI tool for assessing the resilience of agricultural supply chains. The system combines the economic model GTAP and the biophysical APSIM into a single interface with natural language queries, designed for policymakers and market participants.

Why do agricultural supply chains need integrated analysis?

Food supply chains are vulnerable on two fronts at once: biophysical and economic. A drought in one region reduces crop yield — that is a biophysical fact. But through global markets, the same shock reaches importing countries as rising prices and supply shortages — already economic consequences.

Until now, these two layers were analyzed in different tools by different specialists. Agronomists and climatologists worked with biophysical simulators, economists with general equilibrium models. To obtain a joint assessment, it was necessary to run both systems manually, coordinate results, and possess expertise in both domains — a rare combination even for research organizations.

Food security is one of the priority global challenges. Climate shocks and trade instability in recent years have made export restrictions and tariff barriers familiar instruments of government policy. In a crisis — a sudden drought or trade embargo — slow manual analysis procedures become a critical constraint.

How do GTAP and APSIM work under AI management?

GTAP (Global Trade Analysis Project) is one of the key global economic models. International organizations, governments, and universities apply it to assess the consequences of trade policies: tariffs, subsidies, sanctions, and export restrictions. The model accounts for inter-industry and inter-country linkages and is considered the standard for analyzing shocks at the global scale.

APSIM (Agricultural Production Systems Simulator) is a biophysical platform that models crop growth, soil dynamics, and crop response to climate change. Developed in Australia, it is widely used by agronomists for scenario-based crop yield forecasting.

The key innovation is an AI layer that coordinates both systems and accepts questions in plain language. A user formulates a scenario, for example: "What will happen to global wheat prices if Australia is hit by a two-year drought?" — the tool independently translates it into parameters for both models, runs the calculation, and interprets the joint result.

• Natural language query — no programming required
• Coordinated execution of GTAP and APSIM under one scenario
• Integrated output: biophysical consequences plus economic effects by country
• Target audience — policymakers and analysts without dual specialization

What this means

The work illustrates a sought-after application scenario for AI in science: not replacing experts, but removing barriers between specialized tools. GTAP and APSIM have existed for decades and are well-validated. AI acts here as an orchestrator: accepts the question, distributes calculations between two systems, and synthesizes the answer in a convenient format.

Such an approach — AI as a bridge between domain-specific models — is especially valuable where well-validated specialized tools already exist but lack the infrastructure for joint use. Crop yield forecasting and trade impact assessment have long existed as separate disciplines; running them jointly from a single interface is a non-trivial step.

If the tool passes independent validation and finds application in actual analytical practice, it could accelerate decision-making on food security — a task that today requires days of joint work by two teams of specialists.