Welcome to Smart Agriculture 中文

Smart Agriculture

   

Time-Series Prediction of Crop Growth and Yield: A Review of Machine Learning and Deep Learning Methods

LI Zhenbo1,2,4,5,6,7(), KONG Xiangming1,2, PENG Zhen1,2, LIU Yeqiang1,2, LI Xiaomeng3, SUN Haoxiang1,2, ZHANG Yuxuan8   

  1. 1. College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China
    2. National Innovation Center for Digital Fishery, Ministry of Agriculture and Rural Affairs, Beijing 100083, China
    3. College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310058, China
    4. Key Laboratory of Agricultural Informatization Standardization, Ministry of Agriculture and Rural Affairs, Beijing 100083, China
    5. Beijing Engineering and Technology Research Center for Internet of Things in Agriculture, Beijing 100083, China
    6. Key Laboratory of Smart Farming Technology for Agricultural Animals, Ministry of Agriculture and Rural Affairs, Beijing 100083, China
    7. National Digital Agricultural Product Circulation (Supply Chain and Logistics) Innovation Sub-Center, Beijing 100083, China
    8. School of Artificial Intelligence, China University of Mining and Technology (Beijing), Beijing 100083, China
  • Received:2026-05-13 Online:2026-10-07
  • Foundation items:国家农业科技重大项目(0304)
  • corresponding author:
    LI Zhenbo, E-mail:

Abstract:

[Significance] Time-series prediction of crop growth and yield based on machine learning and deep learning is an important research topic in smart agriculture, with significant value for cultivar breeding, crop management optimization, yield assessment, and regional agricultural decision-making. Although substantial progress has been made in recent years, existing reviews mainly focus on specific crops, single scales, or individual methodological categories, and a systematic review from a computational perspective is still lacking. To clarify the task boundaries, technical evolution, and major bottlenecks of this field, a total of 70 relevant studies published between 2019 and 2026 were systematically reviewed, and crop time-series prediction is categorized into two core tasks: 1) individual-plant-scale crop growth prediction, mainly represented by facility-cultivated crops or model plants such as tomato, komatsuna, and Arabidopsis thaliana, which focuses on fine-grained modeling of plant growth status, phenotypic traits, and future morphological changes; and 2) population-scale crop yield prediction, mainly represented by field crops such as wheat, rice, maize, and soybean, which aims to predict yield at field or regional scales. [Progress] From the perspective of computer science, the technical evolution of crop growth and yield time-series prediction is analyzed in terms of machine learning and deep learning methods, representative model architectures, data modalities, prediction targets, and their advantages and limitations. The analysis shows that data sources and modeling strategies in this field have become increasingly diversified. Manually measured phenotypic traits, handcrafted vegetation indices, pixel-level annotations, and ground-truth yield measurements remain important foundations for model training, supervision, and validation. At the same time, recent studies increasingly incorporate image sequences, raw spectral bands, multi-source remote sensing data, and deep representation learning. At the individual-plant scale, prediction targets have expanded from numerical phenotypic traits, such as plant height, biomass, and fresh weight, to future images, masks, and morphological sequences. At the population scale, yield prediction studies increasingly integrate unmanned aerial vehicle (UAV) or satellite imagery, spectral data, meteorological records, soil information, and management variables. Overall, this field has expanded from traditional phenotypic measurement and feature engineering toward richer data sources and learned spatial, temporal, and spatiotemporal representations. [Conclusions and Prospects] Current studies still face challenges related to high-quality dataset construction, multi-source heterogeneous data fusion, long-sequence modeling, integration of biological prior knowledge, and cross-crop, cross-year, and cross-region generalization. Future research should further develop standardized multi-crop, multi-region, and multimodal datasets, design efficient strategies for heterogeneous data fusion and long-sequence modeling, incorporate biological and agronomic prior knowledge, and improve model robustness across different crops, years, regions, and cultivation environments.

Key words: time-series prediction, crop growth prediction, crop yield prediction, precision agriculture, machine learning, deep learning

CLC Number: