Digital Economy Simulation of Small Farmers' Income Using Multi-Source Data and Deep Learning

Abstract

In the evolving digital economy, accurately simulating and forecasting the income of smallholder farmers is critical for formulating effective agricultural policies, enabling targeted financial services, and ultimately ensuring rural livelihood security. However, this task remains challenging due to the complex, dynamic, and localized nature of agricultural systems, which are influenced by a multitude of factors including climate variability, market price fluctuations, and socio-economic conditions. Traditional economic models often rely on aggregated, low-frequency data, failing to capture these fine-grained, non-linear interactions. To bridge this gap, this paper proposes a novel Digital Economy Simulation Framework for Farmer Income (DESFI). The DESFI framework leverages multi-source data integration, including satellite imagery, daily market prices, meteorological data, and farmer transaction records, processed through a dedicated deep learning architecture. At its core is a Spatio-Temporal Fusion Network (STF-Net), which combines a Convolutional Neural Network (CNN) for extracting spatial features from satellite data with a Long Short-Term Memory (LSTM) network for modeling temporal dependencies in price and weather sequences. A key innovation is the introduction of an Income Influence Score (IIS), a metric derived from explainable AI (XAI) techniques that quantifies the contribution of each driving factor to individual farmers' income fluctuations. We validate DESFI using a large-scale, real-world dataset from a major agricultural region in China over three years. Our results demonstrate that DESFI significantly outperforms traditional econometric and machine learning baselines, achieving an R2 of 0.89 in income simulation. In a case study on policy simulation, DESFI accurately projected that a combined policy of optimized subsidy allocation and micro-insurance rollout could increase average income by 15.2% under adverse climate scenarios, providing actionable insights for policymakers.