基于无人机多光谱的玉米与大豆盐胁迫响应指数及盐分管理干预比较评估

1. Introduction

Soil salinization is one of the major abiotic stressors constraining sustainable agricultural development worldwide [1]. Excessive soil salinity not only impairs water and mineral nutrient uptake and transport but also directly suppresses photosynthetic activity, thereby inducing coordinated changes in a range of morphological and physiological traits, including plant height, leaf area, chlorophyll content, and dry matter accumulation. These changes ultimately manifest as distinct phenotypic characteristics, such as inhibited plant growth and altered canopy spectral properties [2,3]. Traditionally, salt stress has been diagnosed through manual field observations based on phenotypic traits, whereby crop growth status is visually assessed using geometric characteristics, morphology, and color. However, this approach is labor-intensive, time-consuming, and often yields results with limited spatial representativeness. More importantly, the pronounced spatial heterogeneity of soil salinity leads to highly discontinuous and patchy crop growth patterns that cannot be adequately captured by conventional field observations [4]. In China, where saline cropland is widely distributed, considerable efforts have been devoted to developing chemical, biological, and agronomic amelioration strategies. The effectiveness of these approaches has typically been evaluated using field-based observations of crop growth, yield, and soil properties. Nevertheless, their practical performance under conditions of pronounced spatial variability remains insufficiently understood [5]. Therefore, accurate and efficient monitoring of salt stress through the response of crops is essential for providing a scientific basis for optimizing soil amelioration and crop management strategies, improving the utilization efficiency of saline cropland, and identifying effective amelioration practices.

Current remote sensing approaches for salinity monitoring have been developed across multiple platforms. Satellite sensors provide large-scale and long-term observations but are often limited by spatial resolution when characterizing field-scale salinity heterogeneity. UAV platforms provide centimeter-level spatial information and flexible data acquisition, making them suitable for crop-scale stress monitoring; however, most UAV-based studies still rely on individual spectral indices or limited indicators. Although hyperspectral and ground-based measurements can provide detailed physiological information, their operational efficiency and scalability remain limited. Therefore, integrating multi-dimensional crop responses with UAV observations remains necessary for more comprehensive salt stress characterization. At the crop scale, UAV multispectral observations can capture stress-related changes in canopy structure and pigment status through variations in spectral reflectance, providing a basis for quantitative monitoring of crop salt stress [6]. Stress-induced response in crop internal structure and chlorophyll content can lead to regular variations in reflectance at specific spectral bands, which lays the theoretical foundation for using UAV-based multispectral data to monitor salt stress and provides a technical means for objectively evaluating the effectiveness of soil amelioration. To date, various abiotic and biotic stresses, such as drought, nitrogen deficiency, and disease, have been extensively explored in major crops, including maize, soybean, and wheat [7,8,9,10,11,12,13]. For salt stress monitoring, Zhao et al. used biomass as an indicator and constructed an alfalfa salt stress index with UAV multispectral data, realizing quantitative diagnosis of alfalfa salt stress [14]. Shi et al. established an evaluation system for saline soil improvement effectiveness using UAV-derived leaf area index at the jointing stage of winter wheat [15]. Although UAV-based stress monitoring has evolved from single spectral indicators to multi-trait crop responses, existing composite indices often determine trait contributions based on statistical variance, empirical assumptions, or yield-related relationships, rather than the response relationships between crops and environmental stressors. This limitation is particularly relevant because crop growth indicators exhibit distinct sensitivities and response dynamics under salt stress. Previous studies have shown that salt stress initially affects leaf area index, followed by fractional vegetation cover and photosynthetically active radiation, whereas physiological indicators such as leaf water content and chlorophyll-related traits generally respond at later stages [16]. These asynchronous responses highlight the difficulty of accurately characterizing salt stress based on individual traits, which may fail to capture the integrated effects of salinity on crop morphological structure and physiological function [17]. Therefore, an index that integrates multiple crop indicators according to their responses to soil salinity is needed to better characterize crop salt stress and facilitate comparative assessment of management practices.

This study develops a salt stress response index (SSRI) by integrating multiple crop growth indicators according to their responses to soil salinity. We hypothesized that integrating multiple growth indicators could provide a more comprehensive characterization of crop salt stress than individual traits and that UAV-based SSRI retrieval performance would vary across crop growth stages due to differences in canopy development and stress expression. The specific objectives of this study are to (i) quantify the responses of key crop growth traits to soil salinity; (ii) identify salt stress-sensitive spectral features and develop models for SSRI retrieval; and (iii) compare crop salt stress responses under different management practices using the proposed SSRI.

2. Materials and Methods

2.1. Study Area

The study area is located at the experimental base of the Shandong Yellow River Delta Agricultural High-tech Industry Demonstration Zone, China, spanning 118°50′–119°03′ E and 37°36′–38°17′ N (Figure 1). This region is representative of a coastal saline soil transition zone, where chloride-dominated saline soils prevail and are characterized by high salinity and pronounced spatial heterogeneity. The area experiences a warm-temperate, semi-humid continental monsoon climate, with a mean annual temperature of 13.3 °C, an accumulated temperature above 10 °C of approximately 4300 °C, and an average annual precipitation of 537 mm. The local cropping system follows a two-year, three-crop rotation, primarily consisting of maize, soybean, and winter wheat. Baseline soil physicochemical properties were characterized to describe the initial soil conditions. The 0–20 cm soil layer had a pH of 8.93, electrical conductivity of 1716 μS cm−1, bulk density of 1.55 g cm−3, organic carbon content of 7.81 g kg−1, available phosphorus of 7.3 mg kg−1, and available potassium of 204.4 mg kg−1. The soil texture consisted of 16.97% sand, 61.30% silt, and 21.73% clay. As a key research base for saline land improvement and sustainable agricultural production, this region provides a critical setting for remote sensing-based monitoring of salt stress. As shown in Table 1, 14 experimental plots were established for maize and another 14 for soybean, each measuring 15 m × 50 m. Each management practice was assigned to one experimental plot.

Figure 1. Geographic location of the study area and distribution of sampling sites. (a) Location of the Shandong Yellow River Delta Agricultural High-tech Industry Demonstration Zone. (b) Location of the experimental sites. (c) Distribution of the experimental treatments and sampling sites.

Table 1. Experimental treatments for maize and soybean.

The experimental management practices consisted of soil amendment practices, cultivar selection, and agronomic management practices. For maize, the evaluated practices included shared practices (Treatments 1–10) and maize-specific practices (Treatments M1–M4). Treatment 1 represented conventional management and served as the reference practice, whereas Treatment 2 received 75% of the conventional fertilizer rate. Treatments 3–10 involved soil amendment practices, including a waste-derived soil conditioner, phosphorus-rich mineral amendment, and other additive measures. The maize-specific practices included cultivar selection (M1 and M2, Zhongdan 1158 and Zhongdan 1191), ridge-furrow planting (M3), and rhizosphere application of a composite microbial agent (M4). Except for M1 and M2, all maize practices used Zhengdan 958 as the cultivar. For soybean, Treatments 1–10 followed the same shared management practices as those applied to maize, whereas Treatments S1–S4 represented cultivar selections involving four soybean varieties. These management practices represented different approaches for regulating crop growth conditions under saline environments. Notably, soybean Treatment 9 was excluded from subsequent analyses because this plot underwent separate replanting during the growing season due to field management issues. The replanting resulted in inconsistent crop establishment and growth timing compared with the other plots, making it unsuitable for direct comparison. The evaluated practices originated from different research teams and were compared exploratorily under the field conditions of this study.

Maize was sown on 22 May 2024 and harvested on 2 October 2024 at a density of 82,500 plants ha−1; soybean was sown on 30 May 2024 and harvested on 15 October 2024 at a density of 210,000 plants ha−1. All plots received the same irrigation management, whereas Treatment 2 differed from the other practices in fertilizer application rate. The conventional fertilizer regime was 140 kg N ha−1, 60 kg P 2 O 5 ha−1, and 60 kg K 2 O ha−1 for maize, and 60 kg N ha−1, 75 kg P 2 O 5 ha−1, and 60 kg K 2 O ha−1 for soybean. All fertilizers were applied as basal fertilizer by broadcasting followed by rotary tillage before sowing, and Treatment 2 received 75% of the conventional fertilizer rate. After sowing, one irrigation event was conducted using drip irrigation belts with an irrigation amount of 60 mm. No further irrigation was applied during the growing season, and the irrigation water salinity was 484 mg L−1.

2.2. Data Acquisition and Processing

2.2.1. Soil Sampling Data

Three sampling sites were established within each plot to capture within-plot spatial variability. Soil samples were collected in May 2024, prior to crop sowing and following the seasonal peak of soil salt accumulation, and were used to characterize baseline soil salinity. Three sampling sites were evenly distributed within each plot (Figure 1). At each sampling site, three surface soil samples (0–20 cm) were collected and thoroughly mixed to form one composite sample. Therefore, each plot contained three independent soil sampling locations, and the resulting composite samples were used for soil salinity (SC) determination. Accordingly, 42 SC observations for maize and 39 SC observations for soybean were obtained and used in subsequent analyses. The samples were air-dried, passed through a 2 mm sieve, and prepared as soil–water extracts at a soil-to-water ratio of 1:5. After shaking for 5 min and allowing the extracts to stand for 8 h, electrical conductivity (EC) was measured using a FE38 conductivity meter (Mettler Toledo, Zurich, Switzerland). The measured EC values (EC 1:5, mS/cm) were converted to soil salt content (SC, g/kg) using the following conversion equation established for the same saline soil region and soil salinity type [18]: SC = 2.9863·EC 1:5 − 0.1019. The equation was established based on saline soils with SC ranging from 0.78 to 39.54 g/kg.

The soil salt content across the experimental plots ranged from 0.79 to 2.72 g/kg. According to the Chinese soil salinity classification standard [19], the soils were classified as non-saline (<1 g/kg), slightly saline (1–2 g/kg), and moderately saline (2–4 g/kg). No artificial salt gradient was imposed in this study. Instead, the baseline salinity of the experimental site encompassed the typical range of soil salinity found in regional maize and soybean production areas [20]. For each plot, the reported SC value was calculated as the mean of three sampling sites within the plot.

2.2.2. Crop Sampling Data

Crop growth measurements were collected at each sampling site on 18 July and 23 August 2024. These dates corresponded to the maize jointing stage and soybean branching stage and to the flowering stage of both crops, respectively. A total of 42 sets of growth measurements were collected for maize and 39 for soybean at each sampling campaign.

Given the relatively small size of each experimental plot, all measurements were obtained using non-destructive in situ methods to avoid disturbing crop growth and maintain the integrity of the field experiment. Four key growth indicators were measured directly in the field: leaf area index (LAI), plant height (PH), relative chlorophyll content (SPAD), and fractional vegetation cover (FVC). These indicators were selected because of their sensitivity to salt stress and their suitability for retrieval from UAV-based remote sensing data. LAI characterizes canopy density and is closely associated with photosynthetic productivity [21]. FVC reflects horizontal canopy coverage and effectively captures reductions in canopy cover caused by salt stress [22]. SPAD provides an indirect measure of leaf chlorophyll content and serves as an indicator of salt-induced changes in photosynthetic physiology [23,24]. PH, a fundamental morphological trait, is highly responsive to salt stress and complements the assessment of crop growth status [25,26]. Collectively, these four indicators represent complementary aspects of crop structure, function, and physiology, while all can be retrieved from remote sensing data, making them well suited for field-scale monitoring.

Specifically, LAI was measured using an LAI-2200C Plant Canopy Analyzer (LI-COR Biosciences, Lincoln, NE, USA). Three measurements were taken at each sampling site and averaged to obtain the site-level LAI value. SPAD was measured using a SPAD-502Plus Chlorophyll Meter (Konica Minolta, Tokyo, Japan). Readings were collected from the midpoint of the fully expanded functional leaf of 10 randomly selected plants at each sampling site. PH was determined by measuring the vertical distance from the soil surface to the apex of 10 randomly selected plants at each site using a ruler.

FVC was estimated using the digital photography method described by [27]. Three RGB images were captured along the diagonal of each sampling site using a Huawei P20 smartphone camera (1632 × 1224 pixels) with the camera positioned approximately 2 m above the canopy under nadir viewing geometry. Images were processed in MATLAB R2016b (MathWorks Inc., Natick, MA, USA) by converting RGB images into the CIE Lab* color space. The a* channel was used to separate vegetation from background, and vegetation pixels were extracted using a fixed threshold (a* = −8.0). FVC was calculated as the proportion of vegetation pixels within each image, and the mean value of the three images was used as the site-level FVC.

At crop maturity, grain yield was determined at both sampling-site and plot scales through manual harvesting. For site-level yield measurement, 10 representative plants were harvested at each sampling site, whereas the entire experimental plot was harvested for plot-level yield determination. After threshing and cleaning, the grains were air-dried to constant weight, and yield was expressed as kg/ha based on dry grain weight.

2.2.3. UAV Multispectral Data

UAV multispectral imagery was acquired on the same dates as the field measurements using a DJI Phantom 4 Multispectral (P4M) UAV (SZ DJI Technology Co., Shenzhen, China). The basic parameters of the P4M camera are summarized in Table 2. Flights were conducted between 10:00 and 14:00 under stable illumination conditions with a flight speed of 3 m s−1. Before each flight, radiometric calibration was performed using the DJI P4M reflectance calibration panel. The UAV was flown at an altitude of 40 m, resulting in a ground sampling distance (GSD) of approximately 2.1 cm, with 75% forward and side overlaps. The RTK positioning module of the UAV was enabled during data acquisition for image georeferencing. Images were acquired with the multispectral camera oriented nadir to the ground. Orthomosaic generation, radiometric processing, and inter-band alignment were performed using DJI Terra to generate reflectance mosaics for the blue, green, red, red-edge, and near-infrared bands, with a spatial resolution of approximately 2.1 cm. Regions of interest (ROIs) corresponding to each sampling site were delineated as 3 m × 3 m square areas around the sampling locations in ENVI 5.3. The mean reflectance values of each spectral band within each ROI were extracted for subsequent analysis.

Table 2. Basic parameters of the P4M camera.

2.3. Method

This study developed a framework for estimating crop salt stress and exploring differences among salinity-management practices by integrating UAV remote sensing, crop growth measurements, and soil data, as illustrated in Figure 2. First, partial least squares path modeling (PLS-PM) was used to derive relative weights for LAI, PH, SPAD, and FVC based on their relationships with baseline soil salinity, and these weighted indicators were integrated to con