Scholarlink

Background <p>Leaf Nitrogen Content (LNC) is an important indicator of the nutritional status of fruit trees. To facilitate precise nutrient management in orchards within Xinjiang’s arid region, this study focused on walnut (<i>Juglans regia</i> L.), jujube (<i>Ziziphus jujuba</i> Mill.), and apricot (<i>Prunus armeniaca</i> L.) trees in the Weigan–Kuqa River Delta Oasis (Weiku Oasis) of southern Xinjiang (50 samples per species), and used Unmanned Aerial Vehicle (UAV) multispectral imagery to develop an optimal inversion method for LNC. Spectral indices and Texture Features (TF) related to LNC were extracted, including eight Two-Band Optimized Spectral Indices (2B-OSI), eight Three-Band Optimized Spectral Indices (3B-OSI), and eight TF. These were integrated with Random Forest Regression (RFR), Support Vector Regression (SVR), and Quadratic Polynomial Regression (QPR) models to assess their performance.</p> Results <p>The RFR model integrating 2B-OSI, 3B-OSI, and TF achieved the best performance, with validation set Coefficient of Determination (R<sup>2</sup>) ranging from 0.89 to 0.93, Root Mean Square Error (RMSE) from 0.84 to 1.13&#xa0;g/kg, and Ratio of Performance to Deviation (RPD) from 2.81 to 3.12. After TF were incorporated, R<sup>2</sup> increased by 1.95%, RMSE reduced by 6.73%, and RPD increased by 5.89%, indicating that TF further improved model robustness. Robustness and uncertainty analyses indicated that, under small-sample conditions, the spectral–textural fusion framework still exhibited relatively stable predictive performance across different validation conditions. Spatial analysis revealed clear species-specific distribution patterns of LNC: jujube showed clustered high values, apricot exhibited a relatively balanced distribution, and walnut had overall lower nitrogen levels.</p> Conclusions <p>The integration of spectral indices and TF based on UAV multispectral imagery can effectively achieve inversion of LNC in fruit trees in arid regions, with the RFR model showing the best performance. However, the relatively limited sample size and validation uncertainty should be considered when interpreting model performance. These findings can provide theoretical support and technical guidance for precise fertilization and nutrient management in orchards in the arid region of Xinjiang.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Estimation of leaf nitrogen content in fruit trees within Arid Oasis Orchards using UAV-based spectral–texture fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • Jiaxi Liang,
  • Jiawen Yu,
  • ue Wu,
  • Muqiao Li,
  • Jiansuo Liu,
  • Kai Nian,
  • Aihua Long

摘要

Background

Leaf Nitrogen Content (LNC) is an important indicator of the nutritional status of fruit trees. To facilitate precise nutrient management in orchards within Xinjiang’s arid region, this study focused on walnut (Juglans regia L.), jujube (Ziziphus jujuba Mill.), and apricot (Prunus armeniaca L.) trees in the Weigan–Kuqa River Delta Oasis (Weiku Oasis) of southern Xinjiang (50 samples per species), and used Unmanned Aerial Vehicle (UAV) multispectral imagery to develop an optimal inversion method for LNC. Spectral indices and Texture Features (TF) related to LNC were extracted, including eight Two-Band Optimized Spectral Indices (2B-OSI), eight Three-Band Optimized Spectral Indices (3B-OSI), and eight TF. These were integrated with Random Forest Regression (RFR), Support Vector Regression (SVR), and Quadratic Polynomial Regression (QPR) models to assess their performance.

Results

The RFR model integrating 2B-OSI, 3B-OSI, and TF achieved the best performance, with validation set Coefficient of Determination (R2) ranging from 0.89 to 0.93, Root Mean Square Error (RMSE) from 0.84 to 1.13 g/kg, and Ratio of Performance to Deviation (RPD) from 2.81 to 3.12. After TF were incorporated, R2 increased by 1.95%, RMSE reduced by 6.73%, and RPD increased by 5.89%, indicating that TF further improved model robustness. Robustness and uncertainty analyses indicated that, under small-sample conditions, the spectral–textural fusion framework still exhibited relatively stable predictive performance across different validation conditions. Spatial analysis revealed clear species-specific distribution patterns of LNC: jujube showed clustered high values, apricot exhibited a relatively balanced distribution, and walnut had overall lower nitrogen levels.

Conclusions

The integration of spectral indices and TF based on UAV multispectral imagery can effectively achieve inversion of LNC in fruit trees in arid regions, with the RFR model showing the best performance. However, the relatively limited sample size and validation uncertainty should be considered when interpreting model performance. These findings can provide theoretical support and technical guidance for precise fertilization and nutrient management in orchards in the arid region of Xinjiang.