Probabilistic NDVI Forecasting from Sparse Satellite Time Series and Weather Covariates
arXiv:2602.17683v1 Announce Type: new Abstract: Accurate short-term forecasting of vegetation dynamics is a key enabler for data-driven decision support in precision agriculture. Normalized Difference Vegetation Index (NDVI) forecasting from satellite observations, however, remains challenging due to sparse and irregular sampling caused by cloud coverage, as well as the heterogeneous climatic conditions under which crops evolve. In this work, we propose a probabilistic forecasting framework specifically designed for field-level NDVI prediction under clear-sky acquisition constraints. The method leverages a transformer-based architecture that explicitly separates the modeling of historical vegetation dynamics from future exogenous information, integrating historical NDVI observations with both historical and future meteorological covariates. To address irregular revisit patterns and horizon-dependent uncertainty, we introduce a temporal-distance weighted quantile loss that aligns the t
arXiv:2602.17683v1 Announce Type: new Abstract: Accurate short-term forecasting of vegetation dynamics is a key enabler for data-driven decision support in precision agriculture. Normalized Difference Vegetation Index (NDVI) forecasting from satellite observations, however, remains challenging due to sparse and irregular sampling caused by cloud coverage, as well as the heterogeneous climatic conditions under which crops evolve. In this work, we propose a probabilistic forecasting framework specifically designed for field-level NDVI prediction under clear-sky acquisition constraints. The method leverages a transformer-based architecture that explicitly separates the modeling of historical vegetation dynamics from future exogenous information, integrating historical NDVI observations with both historical and future meteorological covariates. To address irregular revisit patterns and horizon-dependent uncertainty, we introduce a temporal-distance weighted quantile loss that aligns the training objective with the effective forecasting horizon. In addition, we incorporate cumulative and extreme-weather feature engineering to better capture delayed meteorological effects relevant to vegetation response. Extensive experiments on European satellite data demonstrate that the proposed approach consistently outperforms a diverse set of statistical, deep learning, and recent time series baselines across both point-wise and probabilistic evaluation metrics. Ablation studies further highlight the central role of target history, while showing that meteorological covariates provide complementary gains when jointly exploited. The code is available at https://github.com/arco-group/ndvi-forecasting.
Executive Summary
This article proposes a probabilistic forecasting framework for field-level Normalized Difference Vegetation Index (NDVI) prediction using satellite observations and weather covariates. The method leverages a transformer-based architecture to integrate historical NDVI observations with historical and future meteorological covariates, addressing irregular revisit patterns and horizon-dependent uncertainty. The proposed approach consistently outperforms various statistical and deep learning baselines, and ablative studies highlight the importance of target history and meteorological covariates. The code is available, enabling reproduction and further development of the method. This work has significant implications for precision agriculture, as accurate short-term NDVI forecasting can inform data-driven decision support.
Key Points
- ▸ Probabilistic forecasting framework for NDVI prediction
- ▸ Transformer-based architecture for integrating historical and future covariates
- ▸ Temporal-distance weighted quantile loss for addressing irregular revisit patterns and horizon-dependent uncertainty
- ▸ Cumulative and extreme-weather feature engineering for capturing delayed meteorological effects
Merits
Strength
The proposed framework effectively addresses the challenges of sparse and irregular satellite data, and its performance is consistently better than various baselines.
Strength
The introduction of temporal-distance weighted quantile loss and cumulative and extreme-weather feature engineering improves the model's ability to capture delayed meteorological effects.
Strength
The availability of the code enables reproduction and further development of the method.
Demerits
Limitation
The study focuses on European satellite data, and it is unclear whether the proposed framework would generalize to other regions and data sources.
Limitation
The impact of cloud coverage on the model's performance is not explicitly addressed.
Limitation
The study does not explore the potential applications of the proposed framework beyond precision agriculture.
Expert Commentary
The proposed framework is a significant contribution to the field of precision agriculture, as it addresses the challenges of sparse and irregular satellite data and provides a probabilistic forecasting approach. The use of a transformer-based architecture and temporal-distance weighted quantile loss improves the model's ability to capture delayed meteorological effects, and the availability of the code enables reproduction and further development of the method. However, the study's limitations, such as the focus on European satellite data and the impact of cloud coverage, should be addressed in future work. Additionally, the potential applications of the proposed framework beyond precision agriculture should be explored.
Recommendations
- ✓ Future work should focus on adapting the proposed framework to other regions and data sources, and exploring its potential applications beyond precision agriculture.
- ✓ The impact of cloud coverage on the model's performance should be explicitly addressed in future studies.
- ✓ The proposed framework can be integrated with other big data analytics techniques to improve its performance and inform the broader area of big data analytics in agriculture.
