一种面向天气驱动的全国作物产量预测的节俭输入深度学习框架：以巴西大豆为例

arXiv cs.LG 论文
crop-yield-forecasting deep-learning weather agriculture transformer mamba agri-tech

摘要

论文提出了一种节俭输入的深度学习框架，仅使用常规天气数据和两个静态情境变量即可进行全国作物产量预测，并在巴西大豆20个生长季的数据上对MLP、CNN、LSTM、Transformer和Mamba等模型进行了基准比较。Transformer取得了最佳精度（RMSE为149 kg ha⁻¹，rRMSE为5.3%），相比农户基线降低了47.6%的误差；SHAP诊断表明，作物年份驱动长期趋势，而天气变量则驱动年际间的波动。

arXiv:2609.38447v1 Announce Type: new Abstract: Reliable, timely crop-yield forecasts are essential for market stability and risk management, yet many approaches rely on costly or hard-to-scale inputs. We present a frugal, transferable, and architecture-agnostic deep learning framework that uses routine weather as the only time-varying input plus two lightweight static context inputs (crop year and an agro-environmental label) to capture long-run change and regional heterogeneity, while supporting multiple sequence encoders under identical data requirements. Using a 20-season Brazilian soybean case study (2001/02-2020/21) with leave-one-year-out cross-validation, we benchmark MLP, CNN, LSTM, CNN-LSTM, a Transformer encoder and the Mamba state-space model against linear ridge regression and a five-year moving-average "farmer" baseline. All deep learning variants outperform ridge, and all sequential encoders surpass the non-sequential MLP. The Transformer achieves the best national accuracy (RMSE 149 kg ha^-1; rRMSE 5.3%; R^2 = 0.784), reducing error by 47.6% relative to the farmer baseline. In-season forecasts improve monotonically from early- to late-season issuance, reaching approximately 50% lower error than the baseline at the latest forecast point. Ablations indicate that the agro-environmental label and spatial instance expansion (multiple grid-node weather sequences per municipality-year) contribute positively without increasing input complexity. SHAP diagnostics suggest crop year explains most of the long-run trajectory, whereas within-season weather and agro-environmental context primarily drive interannual deviations, with moisture/cloud and thermal-demand variables dominating. Overall, the framework is straightforward to deploy across other crops and geographic regions and is naturally compatible with operational weather forecasts for routine monitoring.
查看原文
查看缓存全文

缓存时间: 2026/10/02 09:49 

# An Input-Frugal Deep Learning Framework for Weather-Driven National Crop-Yield Forecasting: A Case Study of Brazilian Soybean
Source: [https://arxiv.org/abs/2609.38447](https://arxiv.org/abs/2609.38447)
Authors:[Fernando Dupin da Cunha Mello](https://arxiv.org/search/cs?searchtype=author&query=da+Cunha+Mello,+F+D)\(1\),[Prashant Kumar](https://arxiv.org/search/cs?searchtype=author&query=Kumar,+P)\(2, 3, 4\),[Erick G\. Sperandio Nascimento](https://arxiv.org/search/cs?searchtype=author&query=Nascimento,+E+G+S)\(1, 2, 3, 4\) \(\(1\) Stricto Sensu Department, SENAI CIMATEC University, Salvador, Bahia, Brazil, \(2\) Global Centre for Clean Air Research \(GCARE\), School of Engineering, Civil and Environmental Engineering, Faculty of Engineering and Physical Sciences, University of Surrey, Guildford GU2 7XH, United Kingdom, \(3\) Institute for Sustainability, University of Surrey, Guildford, GU2 7XH, United Kingdom, \(4\) Surrey Institute for People\-Centred AI, Faculty of Engineering and Physical Sciences, University of Surrey, Guildford GU2 7XH, UK\)

[View PDF](https://arxiv.org/pdf/2609.38447)

> Abstract:Reliable, timely crop\-yield forecasts are essential for market stability and risk management, yet many approaches rely on costly or hard\-to\-scale inputs\. We present a frugal, transferable, and architecture\-agnostic deep learning framework that uses routine weather as the only time\-varying input plus two lightweight static context inputs \(crop year and an agro\-environmental label\) to capture long\-run change and regional heterogeneity, while supporting multiple sequence encoders under identical data requirements\. Using a 20\-season Brazilian soybean case study \(2001/02\-2020/21\) with leave\-one\-year\-out cross\-validation, we benchmark MLP, CNN, LSTM, CNN\-LSTM, a Transformer encoder and the Mamba state\-space model against linear ridge regression and a five\-year moving\-average "farmer" baseline\. All deep learning variants outperform ridge, and all sequential encoders surpass the non\-sequential MLP\. The Transformer achieves the best national accuracy \(RMSE 149 kg ha^\-1; rRMSE 5\.3%; R^2 = 0\.784\), reducing error by 47\.6% relative to the farmer baseline\. In\-season forecasts improve monotonically from early\- to late\-season issuance, reaching approximately 50% lower error than the baseline at the latest forecast point\. Ablations indicate that the agro\-environmental label and spatial instance expansion \(multiple grid\-node weather sequences per municipality\-year\) contribute positively without increasing input complexity\. SHAP diagnostics suggest crop year explains most of the long\-run trajectory, whereas within\-season weather and agro\-environmental context primarily drive interannual deviations, with moisture/cloud and thermal\-demand variables dominating\. Overall, the framework is straightforward to deploy across other crops and geographic regions and is naturally compatible with operational weather forecasts for routine monitoring\.

## Submission history

From: Fernando Mello \[[view email](https://arxiv.org/show-email/5e64f98b/2609.38447)\] **\[v1\]**Tue, 29 Sep 2026 19:40:09 UTC \(2,339 KB\)

相似文章

针对地表预测中非平稳偏差的状态空间遗忘方法

arXiv cs.LG

本文提出了 SSU-LSF，这是首个面向地表预测、基于 Mamba 系列结构化状态空间模型的机器遗忘框架。该方法利用 EKFac 影响函数与投影梯度上升，从 NDVI、LST 以及作物物候预测中移除非平稳混杂偏差，相比完整重训练，成本降低了 8.4 倍。