Nitrogen monitoring and inversion algorithms of fruit trees based on spectral remote sensing: a deep review

Ruibin Xi, Yixin Gu, Xiaoqian Zhang, Zhenhui RenFrontiers in Plant Science 2024 · DOI

This deep review, published in Frontiers in Plant Science, provides a comprehensive roadmap for using spectral remote sensing to monitor nitrogen levels in fruit trees. By synthesizing research across satellite, UAV, and ground-based platforms, the authors highlight how advanced inversion algorithms—ranging from classical statistics to deep learning—are transforming traditional orchards into data-driven precision systems that optimize yield while protecting the environment.

What Problem Does This Paper Address?

Nitrogen is the "engine" of fruit tree growth, driving protein synthesis and photosynthesis. However, orchard managers face a difficult balancing act: nitrogen deficiency leads to yellowing leaves and stunted yields, while excessive fertilization causes soil degradation and water pollution.

Traditional monitoring relies on destructive chemical analysis (like the Kjeldahl method) or point-based portable meters (like SPAD). These are either too slow, too expensive, or unable to capture the spatial variability of a large orchard. The challenge lies in creating a non-destructive, real-time, and large-scale monitoring system that can "see" nitrogen levels across complex tree canopies.

Methods

The paper categorizes the research framework into three main pillars:

  1. Remote Sensing Platforms:
    • Satellite: Best for large-scale monitoring but often limited by low spatial resolution and "data intermittency" due to cloud cover.
  2. UAV (Drones): The current "sweet spot" for orchards, offering high resolution, flexibility, and the ability to capture data during critical growth stages.
  3. Ground-based: High-precision instruments (spectroradiometers) used primarily for leaf-level validation and developing baseline models.
  4. Indicator Parameters: The authors summarize various Vegetation Indices (VIs) used as proxies for nitrogen, including the ubiquitous NDVI, the more nitrogen-sensitive NDRE (Red-Edge), and soil-adjusted indices like SAVI and MSAVI which help filter out background noise from the orchard floor.
  5. Inversion Algorithms:
    • Conventional: Linear and Partial Least Squares Regression (PLSR).
  6. Machine Learning: Random Forest (RF), Support Vector Machines (SVM), and Gaussian Process Regression (GPR).
  7. Deep Learning: Convolutional Neural Networks (CNNs) used for processing complex spatial and spectral patterns.
  8. Hybrid Models: Combining physical Radiative Transfer Models (RTM) with statistical methods to improve model interpretability and mobility.

Datasets

As a review paper, it does not introduce a single new dataset but analyzes dozens of existing studies. Key datasets mentioned include:

  • Apple Orchards: Extensive studies using UAV-mounted hyperspectral cameras (e.g., UHD 185-Firefly) with sample sizes ranging from 70 to 4,000 leaf samples.
  • Citrus and Olive: Research utilizing multispectral data from drones to map nitrogen status across Mediterranean climates.
  • Peach and Persimmon: Ground-based spectral libraries used to correlate leaf reflectance with biochemical nitrogen content.

Key Insights and Ideas

  • The "Red Edge" Advantage: The paper emphasizes that indices using the "red edge" (the region of rapid change in reflectance between red and near-infrared) are significantly more effective at monitoring nitrogen than standard NDVI, which tends to saturate at high biomass levels.
  • Phenology Matters: Nitrogen levels are not static; they fluctuate based on the tree's life cycle (flowering vs. fruit expansion). The authors argue that models must be "phenology-aware" to be truly accurate.
  • The Hybrid Frontier: While machine learning offers high accuracy, it is often a "black box." Hybrid models that incorporate physical laws of light-matter interaction (RTMs) are identified as the most promising path for creating models that work across different regions and species.
  • Data Fusion: Moving beyond just "spectral" data, the most robust models now integrate texture, shape, and environmental data (soil moisture/weather) to improve nitrogen inversion.

Key Results

  • Algorithm Performance: In many cited UAV studies, Random Forest (RF) and XGBoost consistently outperformed linear models, achieving $R^2$ values between 0.73 and 0.77 for nitrogen retrieval in apple and olive orchards.
  • Timing is Everything: Research showed that the accuracy of nitrogen inversion is highest during the fruit expansion period, providing a critical window for precision fertilization.
  • UAV Superiority: For orchard-scale management, UAVs were found to be superior to satellites because they can fly under clouds and provide the sub-meter resolution necessary to distinguish individual tree canopies from the soil background.

Limitations and Future Work

The authors acknowledge several hurdles that still prevent "plug-and-play" nitrogen monitoring:

  • Universality: Most current models are "site-specific." A model trained on an apple orchard in China may not work for a vineyard in Spain without significant recalibration.
  • Vertical Distribution: Remote sensing primarily sees the "skin" of the canopy. Future work needs to address how nitrogen is distributed vertically within the tree structure.
  • Shadow and Noise: The complex architecture of fruit trees creates significant shadow noise, which can skew spectral data. Improved "de-shadowing" algorithms are needed.
  • Multi-Source Integration: The next step is the seamless integration of ground-based sensors, drone imagery, and satellite data into a single "intelligent orchard" management platform.

Summary updated

Abstract

Nitrogen, as one of the important elements affecting the growth and development of fruit trees, leads to slowed protein synthesis and reduced photosynthesis, resulting in yellowing of the leaves, poor tree growth, and decreased yield under nitrogen-deficient conditions. In order to minimize losses and maximize fruit yield, there is often an occurrence of excessive fertilization, soil structure degradation, and water pollution. Therefore, accurate and real-time monitoring of nitrogen content in fruit trees has become the fundamental prerequisite for precision management of orchards. Furthermore, precision orchard management is crucial for enhancing fruit quality by maintaining the optimal growth conditions necessary for trees. Moreover, it plays a vital role in safeguarding the ecological environment by mitigating the overuse of fertilizers and pesticides. With the continuous development and application of spectral remote sensing technology in agricultural monitoring and land management, this technology can provide an effective method for monitoring nitrogen content. Based on a review of relevant literature, this paper summarizes a research framework for monitoring and inversion of nitrogen content in fruit trees, which provides help for further research. Firstly, based on different remote sensing platforms, the application was discussed, on spectral remote sensing technology in the acquisition of nitrogen content in fruit trees. Secondly, the index parameters that can reflect the nitrogen content of fruit trees are summarized, which provides practical guidance for remote sensing monitoring. Additionally, the regression algorithms and application situations based on spectral data for nitrogen content were introduced. In conclusion, in response to the current issues and technological limitations, future research should focus on studying the nitrogen content characteristics of fruit trees during different phenological periods, integrating multi-type data information, and thereby improving the universality of the nitrogen content inversion model for fruit trees.
BibTeX
@inproceedings{xi2024nitrogen,
  author = {Ruibin Xi and Yixin Gu and Xiaoqian Zhang and Zhenhui Ren},
  title = {{Nitrogen monitoring and inversion algorithms of fruit trees based on spectral remote sensing: a deep review}},
  booktitle = {Frontiers in Plant Science},
  year = {2024},
  doi = {10.3389/fpls.2024.1489151},
}
Figure