Organizations: University of Pittsburgh, Pittsburgh, PA, USA · University of Wisconsin-Madison, Madison, WI, USA · University of Alabama, Tuscaloosa, AL, USA · University of Maryland, College Park, MD, USA
This paper proposes a new method for crop yield prediction, which is essential for developing management strategies, informing insurance assessments, and ensuring long-term food security. Although existing data-driven approaches have shown promise in this domain, their performance often degrades when applied across large geographic regions and long time periods. This limitation arises from two key challenges: (1) difficulty in jointly capturing short-term and long-term temporal patterns, and (2) inability to effectively accommodate spatial data variability in agricultural systems. Ignoring these issues often leads to unreliable predictions for specific regions or years, which ultimately affects policy decisions and resource allocation. In this paper, we propose a new predictive framework to address these challenges. First, we introduce a new backbone model architecture that captures both short-term daily-scale crop growth dynamics and long-term dependencies across years. To further improve generalization across diverse spatial regions, we augment this model with a retrieval-based adaptation strategy. Recognizing the substantial yield variation across years, we design a novel retrieval-and-refinement pipeline that adjusts retrieved samples by removing cross-year bias not explained by input features. Our experiments on real-world county-level corn yield data over 630 counties in the US demonstrate that our method consistently outperforms different types of baselines. The results also verify the effectiveness of the retrieval-based augmentation method in improving model robustness under spatial heterogeneity.
Figures & tables
Figure 1. (a) County-level prediction errors (prediction – ground truth) from a single global model trained across all counties. (b) Spatial variability of soil sand content (TSAND) across all counties.
Figure 2. The overall framework of LYRA-RaTAR.
Method
2016
2017
2018
2019
2020
LSTM
36.37 ± 0.90
35.02 ± 1.18
38.82 ± 3.72
32.72 ± 3.59
31.40 ± 1.41
TCN
30.93 ± 6.09
35.44 ± 2.37
36.71 ± 3.30
30.81 ± 0.85
30.69 ± 0.80
Transformer
43.10 ± 1.53
48.79 ± 3.55
53.80 ± 6.44
45.20 ± 1.71
39.55 ± 2.61
Pyraformer
57.53 ± 0.56
56.12 ± 1.72
68.32 ± 7.61
50.43 ± 0.04
41.90 ± 1.14
iTransformer
60.94 ± 1.46
58.87 ± 3.93
65.05 ± 6.40
50.53 ± 0.10
42.93 ± 1.61
TS-RAG
45.54 ± 6.25
41.48 ± 2.09
46.38 ± 3.37
36.80 ± 3.12
34.39 ± 1.50
Table 1. The prediction root mean squared errors (RMSE) by different approaches in each testing year. We report the mean and standard deviation over 3 runs with random model initialization. The best results are highlighted in bold.
Figure 3. Spatial patterns of prediction errors for LSTM, TCN, and LYRA-RaTAR evaluated on the 2016 and 2017 test years.
Method
2016
2017
2018
2019
2020
LYRA-RaTAR
28.46
30.04
34.92
27.21
27.83
w/o refine
29.22
30.18
35.72
27.67
28.46
LYRA
29.97
31.15
37.37
30.84
29.88
GRU-ATT
34.38
37.25
39.97
34.52
31.90
Table 2. Performance (in RMSE) for different variants of the proposed method.
Figure 4. Performance (in RMSE) of different sample integration methods (fine-tuning vs. context augmentation) in testing years from 2016 to 2020.
Figure 5. Comparison of three approaches for retrieving relevant samples: residual-based retrieval, neighboring-based retrieval, and embedding-based retrieval. For each approach, we show the results for the fine-tuned LYRA model using both the original retrieved samples (w/o refine) and the refined retrieved samples (w/refine) in testing year 2016.
Figure 6. Performance variation of the proposed method in the testing year 2016 with different numbers of retrieved samples, shown for two model versions: with refinement (w/ refine) and without refinement (w/o refine).
Figure 7. Performance of the proposed LYRA-RaTAR and LYRA (i.e., the same backbone model w/o retrieval) in 2016 with different numbers of years in the look-back window.
Figure 8. Attention weights over historical years in the look-back window for different target years. The curves show the attention weights for 20% randomly sampled counties.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 9. Study region of 630 counties across 14 U.S. states. A map showing 630 counties across 14 U.S. states, highlighting the study region.
Variable
Description
Unit
Weather forcing
RADN
Downward solar radiation
MJ m -2 d -1
TMAX_AIR
Daily maximum temperature
∘ C
TDIF_AIR
Diurnal temperature range
∘ C
HMAX_AIR
Daily maximum humidity
kPa
HDIF_AIR
Diurnal humidity range
kPa
Appendix
Table 3. Descriptions and units of the 19 input features used for county-level corn yield prediction.
Accurate crop yield prediction is crucial for sustainable agriculture and global food security. While existing methods are predominantly developed for single-crop prediction, they often struggle to generalize across diverse crop types, without addressing the unique crop phenological responses that are dynamically modulated by complex weather patterns. In this paper, we propose PhenoYieldNet, a multi-crop yield prediction framework that learns crop-specific phenology by explicitly modeling their responses with temporal drivers. Specifically, we develop a crop-aware temporal decoder consisting of a Crop Phenology Bank (CPB) and a Crop Phenology Attention (CPA) module. The CPB integrates a set of learnable embeddings, which leverage a query to guide the CPA module to learn the most relevant phenology patterns for the specific crop. And the CPA module explicitly captures multi-scale trend and variation components to construct temporal contexts, enabling the model to dynamically adjust the attention across different phenological stages. To learn robust and generalizable features for multi-crop prediction, the encoder is initialized with a pre-trained foundation model, and further adapted via a self-supervised Temporal Contrastive Adaptation strategy to align with agricultural temporal dynamics. Extensive experiments conducted on multi-crop datasets indicate that our proposed method significantly outperforms state-of-the-art methods, exhibiting strong generalization capabilities across different regions and crops.
Yu Luo, Xiaogang Zhu, Shan Zeng +4
School of Computer Science, The University of Sydney, NSW 2006, Australia · School of Computer Science and Information Technology, Adelaide University, SA 5005, Australia · College of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, China +2
Crop yield prediction is one of the most important challenge, which is crucial to world food security and policy-making decisions. The conventional forecasting techniques are limited in their accuracy with reference to the fact that they utilize static data sources that do not reflect the dynamic and intricate relationships that exist between the variables of the environment over time [5,13]. This paper presents Attention-Based Multi-Modal Deep Learning Framework (ABMMDLF), which is suggested to be used in high-accuracy spatio-temporal crop yield prediction. The model we use combines multi-year satellite imagery, high-resolution time-series of meteorological data and initial soil properties as opposed to the traditional models which use only one of the aforementioned factors [12, 21]. The main architecture involves the use of Convolutional Neural Networks (CNN) to extract spatial features and a Temporal Attention Mechanism to adaptively weight important phenological periods targeted by the algorithm to change over time and condition on spatial features of images and video sequences. As can be experimentally seen, the proposed research work provides an R^2 score of 0.89, which is far better than the baseline models do.
Gopal Krishna Shyam, Ila Chandrakar
Presidency University, Bengaluru, India · University of Europe for Applied Sciences, Berlin
Reliable agricultural yield statistics are typically reported at coarse administrative scales, whereas modern geospatial machine learning methods require spatially explicit, pixel level supervision. This mismatch has limited the development of large-scale benchmarks for crop yield learning using multimodal Earth observation data. A reproducible benchmark, AgroBench, is presented for transforming publicly available U.S. county level crop yield statistics into weakly supervised pixel-level crop time series. Each crop pixel time series is paired with a county-level yield value as a weak supervisory signal rather than a directly measured pixel-level yield label. Our geospatial data generation pipeline integrates USDA crop yield statistics with crop-specific land cover masks, Sentinel 2 multispectral imagery, Sentinel-1 synthetic aperture radar observations, climatic variables, and terrain information to produce temporally aligned multimodal sequences describing individual crop pixels throughout the growing season. The resulting benchmark contains over 13 million observations from 788,654 unique crop pixels spanning 5,107 county year combinations across eight growing seasons (2017 to 2024) for five major U.S. crops. To facilitate standardized evaluation, we establish a crop yield prediction benchmark using a Leave-One-Year-Out evaluation protocol and provide baseline results using representative machine learning models. By releasing the complete data generation pipeline, benchmark dataset, and evaluation protocol, AgroBench provides a reproducible foundation for future research in weakly supervised learning, multimodal remote sensing, spatiotemporal modeling, and geospatial foundation models for agriculture.