A Simple and Robust Spectral Index for Identifying Lodged Maize Using Gaofen1 Satellite Data

Yuanyuan Chen, Li Sun, Zhiyuan Pei, Juanying Sun, He Li, Weijie Jiao, Jiong You2022 · DOI

TL;DR

This paper introduces a simple and robust spectral index (SSI) derived from Gaofen-1 satellite data to efficiently identify maize lodging, a significant agricultural disaster. The SSI method demonstrates high accuracy in distinguishing lodged from non-lodged maize, offering a scalable and cost-effective solution for rapid lodging monitoring.

What Problem Does This Paper Address?

Crop lodging, caused by severe weather events like typhoons, can lead to substantial yield losses in agriculture, particularly for tall crops like maize. Timely and accurate identification of lodged areas is crucial for assessing damage, informing crop management decisions, and mitigating economic losses. Traditional field-based assessment methods are labor-intensive and lack the spatial coverage needed for large-scale monitoring. While satellite remote sensing offers a promising alternative, existing methods for lodging detection are often complex, rely on expensive high-resolution data, or are not easily scalable for operational use. Furthermore, the spectral changes associated with lodging events are not fully understood, and the effectiveness of traditional vegetation indices (VIs) for this specific application is questionable due to the complex spectral behavior of lodged crops.

Methods

The researchers developed a novel Spectral Sum Index (SSI) to identify lodged maize. This index is calculated by summing the reflectance values across all four bands (blue, green, red, and near-infrared) of the Gaofen-1 satellite imagery:

$SSI = R_{blue} + R_{green} + R_{red} + R_{near-infrared}$

The rationale behind this approach is that maize lodging causes a significant and consistent increase in reflectance across all spectral bands. By summing these increased reflectances, the SSI amplifies the spectral difference between lodged and non-lodged maize, making discrimination more robust. A threshold value for the SSI was determined empirically using ground sample points to create a binary map distinguishing lodged from non-lodged areas. Maize mapping prior to lodging was performed using a Support Vector Machine (SVM) classifier to ensure that only maize areas were considered in the lodging identification process.

Datasets

  • Gaofen-1 (GF-1) Satellite Data: Used for the primary lodging identification. This data has a 16m spatial resolution and a 4-day revisit cycle, making it suitable for rapid monitoring. Images from September 6, 2020, were used for lodging detection, and images from August 16 and 20, 2020, were used for pre-lodging maize mapping.
  • Planet Dove Satellite Data: Utilized for validation purposes. This data offers a finer spatial resolution (3m) and was used to cross-compare the results obtained from GF-1 data.
  • Ground Truth Information:
    • In-situ Sample Points: 263 sample points (140 for Zhaodong, 123 for Ningjiang) were collected in the field, with their lodging status (lodged/non-lodged) and GPS coordinates recorded. A subset was used for training spectral responses, and the remainder for validation.
  • Farmer Survey Questionnaires: Data from 18 farmers in Zhaodong City and 17 in Ningjiang District were collected, detailing the proportion of lodged maize in their fields. This served as a third validation method.

Key Insights and Ideas

  • Lodging uniformly increases reflectance: Contrary to some expectations, lodging doesn't just affect specific bands; it leads to a general increase in reflectance across the visible and near-infrared spectrum. This is attributed to changes in canopy structure (flattening, increased reflective surface), reduced soil/shaded leaf contribution, and potential changes in biophysical parameters.
  • SSI outperforms NDVI for lodging detection: While traditional VIs like NDVI are common for crop monitoring, they are less effective for lodging. The simultaneous increase in red and near-infrared bands after lodging can weaken NDVI's sensitivity. The SSI, by summing all bands, effectively amplifies this increase, providing a clearer distinction.
  • Gaofen-1 is a viable platform for rapid lodging monitoring: The combination of GF-1's free accessibility, wide coverage, and short revisit cycle makes it an ideal and cost-effective tool for timely lodging assessments, especially in large agricultural regions.
  • A simple threshold is sufficient: The spectral behavior of lodged maize is distinct enough that a straightforward threshold on the SSI can reliably differentiate between lodged and non-lodged areas, simplifying the operationalization of the method.

Key Results

  • High Accuracy: The SSI method achieved high overall accuracies: 92.86% for Zhaodong City and 88.24% for Ningjiang District, based on validation with ground sample points.
  • Cross-Comparison Agreement: The lodged maize areas identified using GF-1 data showed minimal differences when compared to finer-resolution Planet data, with cross-comparison differences of only 1.01% in Zhaodong City and 1.13% in Ningjiang District.
  • Farmer Survey Consistency: The lodging proportions derived from GF-1 data were consistent with farmer survey results, further validating the method's practicality. For instance, Zhaodong City showed a lodged proportion of 54.56% from GF-1, aligning with farmer reports.
  • Lodged Area Estimates: The study identified significant lodged maize areas: 112.17 thousand hectares (54.56% of maize area) in Zhaodong City and 41.72 thousand hectares (81.79% of maize area) in Ningjiang District.

Limitations and Future Work

  • Data Availability: While GF-1 is suitable, its availability can be limited by cloud cover and rain, which is a common challenge for optical remote sensing.
  • Lodging Severity: The current study focuses on distinguishing between lodged and non-lodged maize. Future work should aim to derive lodging severity (e.g., using inclination angle) to better estimate yield losses.
  • Data Fusion: Merging GF-1 data with microwave data (e.g., SAR) is suggested to overcome the limitations of optical data and potentially capture different aspects of lodging events.
  • Broader Applicability: While the SSI method was tested in two distinct regions, further research is needed to confirm its robustness across diverse geographical areas and crop types.

Summary updated

BibTeX
@misc{chen2022simple,
  author = {Yuanyuan Chen and Li Sun and Zhiyuan Pei and Juanying Sun and He Li and Weijie Jiao and Jiong You},
  title = {{A Simple and Robust Spectral Index for Identifying Lodged Maize Using Gaofen1 Satellite Data}},
  year = {2022},
  doi = {10.3390/s22030989},
}
Figure