Computational models of landscape tree markets

Problem

A landscape tree is grown for years before it is sold, and it is sold into a market that almost nobody can see whole. Prices move with construction cycles, public planting budgets, regional supply, and weather, while the transactions that reveal them are scattered across nurseries, brokers, and site contractors. Growers decide what to plant on instinct and hearsay. Buyers pay whatever the nearest quote says.

Approach

We treat the market as a measurable system. For each species, the relevant quantities are transaction prices, sales volumes, cultivated area, production volumes, and weather. Given a consistent record of these, machine-learning models can estimate where prices and distribution volumes are heading and, just as importantly, show how uncertain that estimate is.

The registered patent KR 10-2278664 describes an apparatus and method that does exactly this. It collects per-species price, sales, area, production, and weather data and predicts landscape tree prices and distribution volumes with deep-learning or machine-learning models, so that trees already under management can be moved more efficiently.

Status

The patented method was developed in an industry setting in 2021 and is registered with the Korean Intellectual Property Office. At Alpha Lab we keep it as an open research question rather than a product: which of these signals actually carry forecasting power, over what horizon, and how thin a market can become before prediction stops being meaningful.