Monitoring Nitrogen Uptake and Grain Quality in Ponded and Aerobic Rice with the Squared Simplified Canopy Chlorophyll Content Index

Abstract

Remote sensing tools have been proposed to assist with rice crop monitoring but have been developed and validated on ponded rice. This two-year study was conducted on a commercial rice farm with irrigation automation technology aimed to (i) understand how canopy reflectance differs between high-yielding ponded and aerobic rice, (ii) validate the feasibility of using the squared simplified canopy chlorophyll content index (SCCCI²) for N uptake estimates, and (iii) explore the SCCCI² and similar chlorophyll-sensitive indices for grain quality monitoring. Multispectral images were collected from an unmanned aerial vehicle during both rice-growing seasons. Above-ground biomass and nitrogen (N) uptake were measured at panicle initiation (PI). The performance of single-vegetation-index models in estimating rice N uptake, as previously published, was assessed. Yield and grain quality were determined at harvest. Results showed that canopy reflectance in the visible and near-infrared regions differed between aerobic and ponded rice early in the growing season. Chlorophyll-sensitive indices showed lower values in aerobic rice than in the ponded rice at PI, despite having similar yields at harvest. The SCCCI² model (RMSE = 20.52, Bias = −6.21 Kg N ha⁻¹, and MAPE = 11.95%) outperformed other models assessed. The SCCCI², squared normalized difference red edge index, and chlorophyll green index correlated at PI with the percentage of cracked grain, immature grain, and quality score, suggesting that grain milling quality parameters could be associated with N uptake at PI. This study highlights canopy reflectance differences between high-yielding aerobic (averaging 15 Mg ha⁻¹) and ponded rice at key phenological stages and confirms the validity of a single-vegetation-index model based on the SCCCI² for N uptake estimates in ponded and non-ponded rice crops.

1. Introduction

The traditional water management in rice farms, in which the crop is ponded for most of the growing season, is threatening the sustainability of the rice industry due to high water use, particularly in arid and semi-arid regions [1,2,3,4]. Aiming to increase rice water use productivity, irrigation strategies that shorten to a greater or lesser extent the period the crop remains flooded have been suggested as alternative techniques to traditional management [5,6,7,8,9]. Aerobic rice production is one of the strategies that has been shown to increase water productivity in comparison with ponded rice [10,11,12,13].

In aerobic rice production, the crop is watered when the soil is in an unsaturated state and drained later so permanent water ponding is not maintained at any time during the growing season. The effect on grain yield for aerobic rice, and water-saving strategies that allow soil to dry to a certain level, is influenced by environmental and management factors, with soil water potential being the main driver of yield variation [14]. The average soil water potential threshold where rice grain yield is not reduced by applying water-saving management strategies has been estimated in a recent meta-analysis as −19 kPa, with variations among rice growing areas depending mainly on climate and soil type [14].

In temperate areas of Australia, aerobic rice production is being considered as a water management strategy of interest for the industry due to the recurrent annual fluctuations in irrigation water availability [15]. The main barriers to the adoption of this irrigation strategy by farmers at commercial scale are the high labor required for water management compared with ponded rice, and famers’ perception of a high crop susceptibility to water stress when the soil is not saturated and/or to cold stress at the microspore stage [16,17], both associated with yield reductions. However, achieving high yields under aerobic rice management at commercial scale is possible with adequate water and nitrogen (N) management. Automation technology for gravity surface irrigation systems is key for reducing labor requirements and enabling timely watering of crops [18,19,20].

Over recent years, a variety of remote sensing tools have been proposed to assist in the monitoring of rice crops for optimizing different aspects of crop management. As the main irrigation water strategy employed for rice production, the majority of these tools have been developed and validated on ponded rice crops.

Due to the differences generally observed in canopy development between aerobic and ponded rice, with the former taking longer to reach full canopy cover [5] and in image background (differences in canopy light extinction coefficient [21] and presence or absence of surface water [22,23]), remote sensing tools developed for ponded rice (e.g., prediction of panicle initiation, nitrogen uptake, or grain yield) may not be appropriate for rice grown aerobically. Carracelas et al. [24] indicated that common chlorophyll-sensitive vegetation indices reported as optimum for N uptake estimates were significantly influenced by the level of surface water in bays at the moment of image collection. These authors observed that among the vegetation indices they assessed the squared simplified chlorophyll canopy content index (SCCCI²) was the least affected by the level of surface water and observation time, suggesting it could be used as a general index for N uptake estimates regardless of the irrigation strategy undertaken. Validation of the model reported by Carracelas et al. [24] in different conditions to those in which it was developed (site, growing season, and variety) would demonstrate the SCCCI² validity for N uptake estimates and provide the industry with a valuable tool for optimizing N fertilizer management in rice farming.

Apart from being essential for maximizing rice grain yield, optimum N management is important for producing high grain quality [25,26,27,28]. Plant N availability, along with other factors, affect grain physical qualities [29]. It is hypothesized that vegetation indices that are sensitive to chlorophyll content could explain crop variability in grain quality. However, there are few studies dealing with rice grain quality monitoring from remote sensing measurements. In a study with two cultivars and a range of N application rates, Tsukaguchi et al. [30] reported a good correlation between the chlorophyll index green (Clg) obtained from drone imagery around heading and the percentage of imperfect grains. Hou et al. [31] reported a good correlation between spectral measurements and brown rice rate, moisture content, and taste value. In Australia, price penalties are imposed to crops with a percentage of cracked and immature grain exceeding established thresholds as both are the main parameters used to estimate the overall quality score. Thus, monitoring grain quality variability within bays could be of interest for growers in order to reduce within crop variability.

The objective of this study conducted at a commercial rice farm was to understand how canopy reflectance can differ between high-yielding ponded and aerobically grown rice crops over the growing season, validating the feasibility of using the SCCCI² for N uptake estimation regardless of the water management strategy undertaken, and exploring the potential use of the SCCCI² and similar chlorophyll-sensitive indices for grain quality monitoring.

2. Materials and Methods

2.1. Study Site Description and Field Management

The study was conducted in two bays (7.4 ha and 5.4 ha) of a commercial rice farm located in the Murrumbidgee rice growing region in New South Wales, Australia (Latitude: 340 21′ 21.9″S, Longitude: 1470 10′ 14.8″E) with different water management (aerobic and traditional) during two consecutive rice growing seasons (2022/2023 and 2023/2024) (Figure 1).

The climate in the Murrumbidgee rice growing region is temperate with hot dry summers and low humidity. A description of selected weather parameters obtained from the nearest weather station (~15 Km) for the period from 1 October to 15 April in each rice growing season is reported in

Table 1. Description of weather parameters obtained from the nearest meteorological station to the rice farm located at Griffith Airport, NSW, Australia (Bureau of Meteorology, Australian Government, www.bom.gov.au/watl/eto/, accessed on 10 May 2025) from the 1 October to the 15 April in each growing season.

Weather Parameters	Rice Growing Season
	2022–2023	2023–2024
Total Evapotranspiration (mm)	1098.9	1263.2
Daily Evapotranspiration (mm d−1)	5.4	6.2
Total Rain (mm)	373.0	332.4
Maximum Temperature (°C)	27.7	30.0
Days with Max. temp. above 35 °C	29.0	42.0
Minimum Temperature (°C)	12.8	14.3
Days with Min. temp. below 15 °C	121.0	94.0
Days with Min. temp. below 15 °C (from PI to Flowering)	24.0	11.0
Maximum Relative Humidity (%)	88.2	83.4
Minimum Relative Humidity (%)	35.2	28.7
Wind-speed at 10 m (m/s)	4.1	4.0
Solar Radiation (MJ m−2 d−1)	21.8	22.4

. Total evapotranspiration was 164.3 mm lower in the first growing season than in the second one. The first growing season had 13 fewer days with maximum temperatures above 35 °C, and 13 more days with minimum temperatures below 15 °C from PI to flowering than the second growing season (Table 1).

A soil corer sample machine (Christie Engineering, Delaware Road, Horsley Park, NSW, Australia) was used prior to sowing to collect a total of 16 soil samples (0–30 cm depth) to determine soil chemical and physical properties in both bays. Soil samples were analyzed at the professional laboratory (EAL Environmental Analysis Laboratory, Lismore, NSW, Australia) (

Table 2. Selected soil properties determined prior to sowing at the Environmental Analysis Laboratory—EAL (www.scu.edu.au/eal, accessed on 1 December 2024). The data shown is the average of the two rice growing seasons and monitored bays.

Soil Parameters	Value	Soil Parameters	Value
pH (water)	7.6	Zn-Zinc (mg/kg)	0.9
pH (CaCl2)	7.0	Mn-Manganese (mg/kg)	20.5
Estimated Organic Matter (% OM)	1.5	Fe-Iron (mg/kg)	78.8
Phosphorus (mg/kg P) (Colwell)	41.4	Cu-Copper (mg/kg)	3.2
Nitrate Nitrogen (mg/kg N)	4.7	B-Boron (mg/kg)	1.1
Ammonium Nitrogen (mg/kg N)	3.4	Si-Silicon (mg/kg Si)	44.0
Sulfur (mg/kg S)	27.6	Total Carbon (%)	0.9
Electrical Conductivity (dS/m)	0.1	Total Nitrogen (%)	0.1
Ca—Exchangeable Calcium (mg/kg)	2.6	Carbon/Nitrogen Ratio	8.8
Mg—Exchangeable Magnesium (mg/kg)	1.2	Texture (ISSS classification)	Clay
K-Exchangeable Potassium (mg/kg)	309.7	Sand > 20 µm	38.3%
Na-Exchangeable Sodium (mg/kg)	102.1	Silt (2−20 µm)	8.9%
Al-Exchangeable Aluminum (mg/kg)	1.9	Clay (< 2 µm)	52.8%
ECEC—Effective Cation Exchange Capacity (cmol+/kg)	25.3	Gravel > 2 mm	0.2%
		Basic Color	Brownish

). The soil type was brownish clay in both bays, and was classified as Transitional red brown earth [32], commonly used for rice production in the area. Relative to the indicative guidelines for a heavy clay soil [33,34], the soil organic matter percentage, phosphorus and nitrate nitrogen on average were low. Potassium levels were higher in relation to the indicative guidelines. Most micronutrients (Zn, Mg, Cu) were on average below the indicative guidelines for a heavy clay soil type.

2.2. Field Management

The study was conducted during the second and third year of consecutively growing rice crops at the farm. Minimum tillage was performed on leveled bays. All tillage operations, herbicides applications and the first N fertilizer applications were undertaken on dry soil before there was any irrigation event.

In the first year of study, due to a wet start of the growing season (204 mm recorded in October, 55% of the total rainfall received throughout the entire crop cycle) the variety V071 (medium grain, Japonica) was broadcasted with pregerminated seed on the 19 October 2022 with an average seeding rate of 158 kg ha⁻¹ (Table 1). Bays were irrigated the following day after sowing and remained flooded for 15 days to promote plant establishment. During the second rice growing season, the same variety was direct drilled (row spacing of 0.20 m) on the 2 October 2023 with an average seeding rate of 142 kg ha⁻¹. The first irrigation event occurred on the 10 October 2023, and it was followed by two more irrigation events to promote emergence. On average for both rice growing seasons, plant establishment was 169 plants per m², ranging from 40 to 305 plants per m² (n = 41).

The fertilization management included a basal starter (custom blend fertilizer—rice bouncer: N8%-P17%-S7% Ca8% Zn0.6%) application of ~220 kg ha⁻¹ in 2022 and 150 kg ha⁻¹ in 2023. This was followed by urea applications in late November and early in January. In the 2022/23 rice growing season, urea was broadcast at 355 kg ha⁻¹ and 145 kg ha⁻¹ in the first and second applications at both bays (500 kg ha⁻¹ in total). In the 2023/24 rice growing season, 275 and 350 kg ha⁻¹ of urea were applied in the aerobic and traditionally managed bays, respectively, in late November, and 125 kg ha⁻¹ were applied in both bays early in January. To level the total amount of urea applied to the aerobic and traditional bays, an additional 100 kg ha⁻¹ of urea were applied to the aerobic bay. In total in this second season, the aerobic and ponded rice received 500 and 475 kg ha⁻¹ of urea.

2.3. Irrigation Management

The traditional water management consisted in maintaining the soil under anaerobic conditions. This was achieved by maintaining a layer of surface water that fluctuated over the growing seasons as reported in Figure 2. At microsporogenesis, as the standard practice followed in the rice growing region, the water layer was increased up to 0.20–0.25 m to protect the crop against the incidence of low temperatures (Figure 2b). The bay was drained in both growing seasons late November early December for the first urea application.

In the aerobic rice management, water was applied to the bay and drained to let the soil dry and return to an aerobic condition before the next irrigation event took place. Irrigation frequency varied during the season depending on water needs in the contiguous bay and other operations like fertilizer applications. However, irrigation was scheduled to try to avoid soil water potential values below −15 kpa at 0.10 m depth (Figure 3). This was monitored with Watermarks (Model 200SS, Irrometer Company Inc., Riverside, CA, USA) that were installed in pairs at the beginning of each growing season. Watermark readings showed that soil matric potential decreased in the aerobic bay between irrigation events (Figure 3). On average, water was applied to the aerobically managed bay every 3–5 days. At the end of the growing season when grain moisture was below 26%, all the bays at the farm were drained off for harvesting the crop on dry soil.

2.4. Multispectral Measurements

Multispectral images were collected four times across the growing season in the first and second year of study, corresponding approximately with the phenological stages of tillering, PI, flowering and maturity. Images were collected from an unmanned aerial vehicle (UAV; Matrice 100, DJI, Shenzhen, China) with the Altum camera (AgEagle Aerial Systems Inc., Wichita, KS, USA) that captures information from the blue, green, red, red edge, and near-infrared (NIR) spectral bands centered at 475, 560, 668, 717 and 840 nm, respectively. Flights were automated using the Pix4Dcapture app that was set up to collect images at 120 m altitude with an 80% overlap. These settings provided a pixel resolution of ~5.6 cm. The spectral bands used to calculate the vegetation indices in this study were the green, red, red edge and NIR bands. An image of a calibration panel was collected before and after each flight to remove the effects of sunlight variation. Information from the radiometric target, camera gain and exposure settings and a Downwelling Light Sensor (DLS2) that provides information about the illumination at the moment each image was collected, was used for the radiometric corrections when the images were processed with the Agisoft software (version 1.8.0, Agisoft Metashape, Saint Petersburg, Russia). A schematic illustration of the methodology followed in this study is provided in Figure 4. The Geo-TIFF files generated for each set of images with information of the multispectral bands were post-processed in QGIS (version 3.26). Four areas of interest of ~397 m² each (radius = 11.25 m), where soil moisture and plant measurements were conducted at different times during the growing seasons, were delimited in both bays. Since there was no treatment to cause variability in the present study, the location of the four areas of interest within each bay were selected based on the (low) variability in NDVI observed in the previous growing season from satellite imagery. The zonal statistical tool was used in QGIS to obtain the average reflectance of the five spectral bands captured with the camera on each measurement date for each area of interest. This information was exported from the attribute table and used to compute the vegetation indices in Excel. The formulation of the multispectral indices used in this study are shown in

Table 3. Vegetation Indices computed from the multispectral images with their respective formulations. G, R, RE, and NIR correspond to reflectance in the green, red, red edge and near-infrared regions of the spectrum.

Vegetation Index	Formulation	References
Squared of simplified canopy chlorophyll content index	SCCCI2 = ${\left.\left({\displaystyle \frac{NDRE}{NDVI}}\right.\right)}^2={\left.\left({\displaystyle \frac{\left(\frac{NIR-RE}{NIR+RE}\right)}{\left(\frac{NIR-R}{NIR+R}\right)}}\right.\right)}^2$	[24]
Squared of normalized difference red edge	NDRE2 = ${\left.\left({\displaystyle \frac{NIR-RE}{NIR+RE}}\right.\right)}^2$	[35]
Chlorophyll green	Clg $=\left({\displaystyle \frac{NIR}{G}}\right)-1$	[36]
Normalized difference vegetation index	NDVI = $\left.\left({\displaystyle \frac{NIR-R}{NIR+R}}\right.\right)$	[37]

.

2.5. Plant Measurements

In both rice growing seasons, four above-ground biomass samples were collected per bay within each area of interest delimited for the remote sensing measurements around PI. Panicle initiation was defined as the moment when 30% of the main tillers had a developing panicle 1–3 mm long [38]. Plant samples were collected within 1 m² in each area of interest by manually cutting plants at ground level. Samples were dried in an oven at 60 °C until they reached a constant weight. Five whole plants including leaves and tillers weighing ~100 g were randomly picked per sample for the NIR tissue test at the Australian grain storage facility (AGS) to determine the nitrogen plant concentration (N%) [39]. Nitrogen uptake was estimated using the following equation:

$$Nuptake=N\%\times DM$$

(1)

where DM is the plant biomass above ground expressed in kg of dry matter per hectare at PI.

Rice grain yield was assessed in late and mid-April in 2023 and 2024, respectively. In the first year of study, eight samples were manually harvested at the areas of interest of each bay using a sampling ring with an area of 0.2 m² [39]. In the second growing season when the crop was direct drilled, however, a quadrant of 1 m² (5 rows × 1 m) was used in both bays. A total of 16 samples were manually harvested in the two rice growing seasons when grain moisture was lower than 22% according to rice industry recommendations. Grain yields were normalized to 14% moisture. The whole bays were harvested with a commercial John Deere header with yield harvest monitor.

Grain moisture content was measured with FOSS (Infratec) equipment [40]. Each sample was threshed using a Pfeuffer grain sample cleaner and a visual check was performed for contaminants.

Grain milling quality was measured using the PaddyVision^® [41] instrument to quickly assess dry or wet paddy rice at the Australian Grain Storage lab, part of Sunrice Group in Leeton, Australia [42]. A total of ~6000 grains were scanned across the sixteen samples (373 grains per sample) collected at harvest in the two evaluated rice growing seasons. The key quality parameters assessed were the percentage of cracked and immature grains, and the quality score, which is estimated from the previous two parameters.

2.6. MultispeQ Measurements

To monitor the possible occurrence of abiotic stress on the aerobic rice, chlorophyll fluorescence and photosynthesis-related parameters were periodically measured in both monitored bays with a hand-held MultispeQ device (V 2.0, Photosynq Inc., East Lansing, MI 48823, USA). Measurements were taken using the protocol Photosynthesis RIDES 2.0, between 11 a.m. and 3 p.m. in the flag leaf of 10 to 20 plants per area of interest within each bay on the same dates when the multispectral images were collected. Measurements were taken five meters from the soil moisture sensors starting from the North side and following counterclockwise to avoid duplicate measurements. A total of 206 individual leaf measurements were taken over the two rice growing seasons. Among the parameters measured with the MultispeQ, only seven were reported here: ambient temperature, relative humidity, leaf temperature differential, chlorophyll SPAD, maximum quantum yield of PSII under light conditions (Fv/Fm), ratio of the incoming light energy used for the photochemistry (Phi2), ratio of the incoming light energy that is lost to unregulated processes (PhiN0), and ratio of the incoming light that is dissipated as heat (PhiNPQ).

2.7. Data Analysis

The statistical analysis was conducted using R software (R version 4.2.2) [43]. Analyses of variance were performed to evaluate the influence of irrigation techniques and rice growing season on multispeQ measurements, above-ground biomass, N%, N uptake, rice grain yield and grain milling quality parameters (total cracked %, immature%, and quality score). Means were compared using Tukey’s tests. Linear regression models were used to analyze the relationships between vegetation indices and rice grain yield and grain milling quality parameters. The best fitting model was selected based on the coefficient of determination (R²) and Akaike information criterion (AIC). Pearson correlation analysis was performed using the cor.test function (R Core Team, 2022). The quadratic and linear regression models were fitted using the lm function in R [44].

Previously published models for N uptake estimates at PI using vegetation indices were assessed by comparing observed and predicted N uptake. A linear–plateau model was fitted for the relationship between NDRE² and N uptake recently published in Carracelas et al. [24], using the nls function in R [45,46]. A linear model was obtained between NDRE² and N uptake up to a plateau threshold of 166 ± 6.3 kg N uptake ha⁻¹ (Figure S1). The obtained model equation with an R² = 0.80 and significance of p < 0.001 was:

$$Nuptake\left(\mathrm{k}\mathrm{g}\mathrm{N}{\mathrm{h}\mathrm{a}}^{-1}\right)=311.{NDRE}^2+29.7$$

(2)

The R², Bias, the root mean square error (RMSE), mean absolute percentage error (MAPE), and the concordance correlation coefficient (CCC) were used to evaluate the performance of different models in estimating N uptake at PI. The CCC was calculated using desctools package in R [43]. The RMSE, Bias and MAPE were calculated using metrics package in R [47] with the following equations:

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}}{\sum }_{i=1}^n{\left(xi-yi\right)}^2$$

(3)

$$Bias={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\left(xi-yi\right)$$

(4)

$$MAPE={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\frac{(xi-yi)}{xi}$$

(5)

where xi and yi are the predicted and measured values, respectively, and n is the number of observations.

3. Results

3.1. Canopy Reflectance and Seasonal Evolution of the Vegetation Indices

Early in the growing season, canopy reflectance in the blue, green, red, and red-edge regions of the spectrum was higher in the aerobic than in the ponded rice in both growing seasons while the opposite was observed for the NIR region in the second rice season (Figure 5). The differences in canopy reflectance in the visible region were more evident in the first rice growing season when crop development in the aerobic rice was lacking behind the ponded rice than in the second growing season. No differences were observed in the visible region between water management strategies later in the growing season except for flowering in 2023/24 and maturity in the 2022/23 growing season, when a slightly lower canopy reflectance was observed in the green and red edge regions in the aerobic rice than in the ponded rice (Figure 5). Canopy reflectance in the NIR was generally lower in the aerobic rice than in the ponded rice at maturity in the first rice season and at tillering and PI in the second rice season. However, these differences were minimized around flowering in both growing seasons (2 March 2023 and 15 February 2024), and maturity in the second growing season.

The evolution of the vegetation indices SCCCI², NDRE², Clg and NDVI reported of interest for rice crop monitoring are illustrated for the aerobically and traditionally managed bays in Figure 6. Mean values of all the vegetation indices were lower in the aerobic rice than in the ponded rice on measurement dates occurring before flowering. This was observed in both rice growing seasons despite differences in the moment when PI occurred in each water management strategy were only detected in the first rice growing season. However, the differences were not statistically significant for the NDVI at PI and for the SCCCI² at PI in the 2022/23 growing season. The observed differences between water management strategies decreased for all the vegetation indices from flowering onwards. The aerobic rice showed slightly higher mean values than the ponded rice at flowering and showed practically no differences with the ponded rice at maturity (Figure 6).

3.2. Biomass, N% and N Uptake at PI

Average above-ground dry biomass and N uptake in 2023 were 3311 kg ha⁻¹ and 100 Kg N ha⁻¹, respectively, while in 2024 above-ground dry biomass was 5733 kg ha⁻¹ and N uptake was 155 Kg N ha⁻¹ (

Table 4. Plant biomass (DM kg ha⁻¹), Nitrogen percent (N%) and Nitrogen uptake (Kg N ha⁻¹) parameters for the water strategy and rice growing season. The interaction irrigation x season and the mean value for each parameter is also indicated.

Treatments	Biomass (DM kg ha−1)	Nitrogen Percent (N%)	N Uptake (Kg N ha−1)
Irrigation
Aerobic	3686	3.03	112
Traditional	5358	2.79	144
Irrigation effect	ns	ns	ns
Season
2022–2023	3311 b	3.01	100 b
2023–2024	5733 a	2.81	155 a
Season effect	.	ns	*
Irrigation × Season effect	***	**	**
Mean	4522	2.91	128

Means followed by different letters are significantly different. Asterisks indicate statistical significance at p < 0.001 ‘***’, p < 0.01 ‘**’, p < 0.05 ‘*’, p < 0.1 ‘.’, ‘ns’ indicates non-significant differences.

).

The irrigation effect was not statistically significant for any of the parameters evaluated (p > 0.1). However, there was a statistically significant interaction between the irrigation strategy and growing season (p < 0.01) (Table 4 and Figure 7). In the first growing season, the ponded rice reached PI before (6 January 2023) the aerobic rice (~25 January 2023). The traditionally and aerobically managed bays had similar above-ground biomass, N%, and N uptake at PI with no significant differences observed between them (Figure 7). In the second growing season, no noticeable differences were observed in the date PI occurred between water management strategies (~5 January 2024). In this growing season, plants from the traditionally irrigated bay had significantly higher biomass, N% and N uptake than plants from the aerobically managed bay (Figure 7).

3.3. Nitrogen Uptake Estimates

A strong linear relationship was observed between N uptake and SCCCI² when data for both sites and rice growing seasons were plotted together (Figure 8). The equation and the statistically significant slope parameter obtained was very similar (1.9 kg N uptake ha⁻¹ higher per 0.1 increase in SCCCI²) to the one reported in Carracelas et al. [24] that was developed in a different site, variety and growing season and that we aimed to validate. The observed versus predicted plot using the model developed in Carracelas et al. [24] (N uptake (kg ha⁻¹) = 273.SCCCI², R² = 0.73, p < 0.001) is illustrated in Figure 9g, along with published single-vegetation-index linear models for N uptake estimates using the chlorophyll-sensitive indices Clg, Clre and NDRE².

Among the models assessed, the model using the SCCCI² had the lowest RMSE (20.5 kg N ha⁻¹), Bias (−6.21 kg N ha⁻¹), and MAPE (11.95%) (Figure 9g). Both the 2021 and 2025 NDRE² models had an R² of 0.83. However, compared with the SCCCI² model, the RMSE observed for the 2021 NDRE² model was 68% higher while only 7% higher for the 2025 NDRE² model (Figure 9c,f). In both cases, the Bias and MAPE were higher than those observed for the SCCCI² model. The SCCCI² and the 2025 NDRE² were the models with the highest CCC values, 0.82 and 0.83, respectively. The models using the indices Clg and Clre, had the highest RMSE, Bias and MAPE (Figure 9a,b,d,e).

3.4. Environmental Conditions at Canopy Level and Photosynthesis-Related Parameters

The MultispeQ measurements showed that temperatures at canopy level were similar for the aerobically and ponded managed bays, but relative humidity was significantly higher in the latter (

Table 5. Temperature, relative humidity, leaf to air temperature differential, ratio of the incoming light energy used for the photochemistry (Phi2), ratio of the incoming light energy that is lost to unregulated processes (PhiN0), ratio of the incoming light that is dissipated as heat (PhiNPQ) and SPAD obtained from the MultispeQ measurements.

Treatments	Ambient Parameters		Leaf Temperature Differential (°C)	Fv/Fm	Fraction of Energy Light Captured by PhotosystemII			SPAD
	Temperature	Humidity			Phi2	PhiN0	PhiNPQ
Irrigation
Aerobic	31.4	40.2 b	−4.98	0.68	0.32	0.3	0.38	51.3
Traditional	31.6	42.8 a	−5.30	0.67	0.32	0.29	0.4	52.4
Irrigation effect	ns	**	ns	ns	ns	ns	ns	ns
Rice Season
2022–2023	32.2 a	38.0 b	−5.29	0.71 a	0.41 a	0.3	0.29 b	52.7
2023–2024	30.7 b	45.0 a	−4.98	0.64 b	0.22 b	0.29	0.49 a	51
Season effect	***	***	ns	***	***	ns	***	ns
Irrigation × Season effect	ns	ns	ns	ns	ns	ns	ns	ns
Mean	31.5	41.5	−5.14	0.67	0.32	0.30	0.39	51.9

Means followed by different letters are significantly different. Asterisks indicate statistical significance at p < 0.001 ‘***’, p < 0.01 ‘**’ and ‘ns’ indicates non-significant differences.

). No statistically significant difference was observed between water management strategies in leaf temperature differential, which on average for all measurements was around −5.0 °C. No statistically significant difference was observed either between water management strategies in Phi2, PhiN0, PhiNPQ and the SPAD measurements (Table 5).

Significantly higher temperatures and lower relative humidity values were recorded in the first growing season than in the second on measurement dates (Table 5). The Phi2 and PhiNPQ measurements were also significantly higher and lower, respectively, in the first growing season than in the second. No statistically significant differences were observed between rice growing seasons for the other parameters measured.

3.5. Grain Yield and Milling Quality

Rice grain yield was significantly higher (10.5%) in the aerobic bay than in the ponded bay Grain milling quality parameters and the overall quality score were not significantly influenced by the water strategy (

Table 6. Rice grain yield (Mg ha⁻¹) and grain milling quality parameters for the water strategy and rice growing season. The interaction irrigation x season and the mean value for each parameter is also indicated.

Treatments	Rice Grain Yield (14%) Mg ha−1	Grain Milling Quality
		Total Cracked %	Immature %	Quality Score
Irrigation
Aerobic	15.03 a	12.9	2.76	8.43
Traditional	13.60 b	22.8	3.43	7.37
Irrigation effect	**	ns	ns	ns
Season
S1 2022–2023	14.35	10.6 b	0.34 b	8.90
S2 2023–2024	14.29	25.1 a	5.85 a	6.91
Season effect	ns	*	***	ns
Irrigation × Season effect	ns	*	ns	**
Mean	14.32	17.85	3.09	7.90

Means followed by different letters are significantly different. Asterisks indicate statistical significance at p < 0.001 ‘***’, p < 0.01 ‘**’, p < 0.05 ‘*’ and ‘ns’ indicates non-significant differences.

).

A significant interaction was observed between the irrigation water strategy and rice growing season for the percentage of cracked grain and quality score (Table 6), indicating these parameters were higher and lower, respectively, in the ponded rice than in the aerobic rice in the second rice growing season (Figure 10). The percentage of cracked grains and immature grains were significantly higher in the second rice growing season than in the first one, and therefore, the quality score was lower.

3.6. Relationship Between Chlorophyll Sensitive Vegetation Indices and Grain Quality Parameters

All the assessed vegetation indices had a strong correlation with the percentage of cracked and immature grain, and the quality score (Figure 11).

All indices were linearly correlated with the immature percentage. For this grain quality parameter, the SCCCI² had the highest R² (0.71), followed by the NDRE² (R² = 0.66) and Clg (R² = 0.61) indices. The lowest R² value was observed for the NDVI. The relationship observed for the percentage of cracked grain and overall quality score fitted a polynomial quadratic model. The NDRE², SCCCI², and Clg indices had a high (>0.75) and similar R² among them for both quality parameters. The NDVI was the index with the lowest R² for both percentage of cracked grain and quality score.

No significant relationship was observed between rice grain yield and any of the vegetation indices (p > 0.1).

4. Discussion

Differences in canopy reflectance in the visible and NIR regions between high-yielding aerobic and ponded rice were more evident early in the growing season when the crop had not reached full canopy cover than later in the season. At tillering, aerobic rice had a higher reflectance in the visible and red edge regions than ponded rice, but the opposite was observed in the NIR region, particularly in the second rice growing season. The differences observed in canopy reflectance early in the growing season were more likely to be related to the advanced development of the above ground part of the plant generally observed in ponded rice with respect to aerobic rice or similar water management practices [5,6,21], thus less background exposed in the image. This could be associated with a promotion of root development in non-ponded strategies during the vegetative phase [49]. Carrijo et al. [50] reported that root length density was significantly higher (100%) under alternate wetting and drying safe in relation to continuously ponded, with no yield loss or plant water stress, as deeper roots accessed deeper water at 25–35 cm. The differences observed in canopy reflectance between aerobic and ponded rice in the visible and NIR regions early in the growing season, and mainly in the NIR region at tillering, PI and maturity, could also be attributed to the aerobic rice being water stressed on days when images were captured [51,52]. However, the MultispeQ measurements showed that leaf to air temperature differential on days when images were collected was similar and negative (leaves cooler than air temperature) in both water management strategies. Further, no statistically significant differences were observed between aerobic and ponded rice in chlorophyll fluorescence (Fv/Fm) and the photosynthesis-related parameters Phi2, PhiN0 and PhiNPQ in any date of measurement, suggesting the aerobic rice was not water stressed on those dates. The fact that grain yield at the monitored locations was higher in the aerobic rice than in the ponded rice at harvest (Table 6), corroborates the inference that aerobic rice did not experience water stress levels detrimental for crop production in any rice growing season. This was achieved by limiting the lowest soil water potential to −14 kPa and −25 kPa in the first and second growing seasons, respectively. These values were close to the soil water potential threshold (−19 kPa) reported by Bo et al. [14] and ≥−20 kpa reported by Carrijo et al. [8] at which rice yield was not reduced in the alternate wetting and drying strategy.

Differences in canopy reflectance between ponded and aerobic rice led to obtaining lower values of the vegetation indices SCCCI², NDRE², Clg and NDVI in the aerobic rice than in the ponded rice up to flowering (March 2 in the first growing season and February 15 in the second). At this phenological stage, differences in canopy reflectance were minimal between the aerobic and ponded rice, and the vegetation indices were either similar for both water management practices or slightly higher in the aerobic rice. As expected, the differences in NDVI observed between ponded and aerobic rice were smaller than those observed for the chlorophyll-sensitive indices because of the tendency of the NDVI to saturate earlier in the growing season in rice crops not deficient in N [35,53,54,55].

In agreement with that observed by Carracelas et al. [24] who first reported the use of the SCCCI² for N uptake monitoring in experimental plots, the SCCCI² was significantly correlated with N uptake at PI in a commercial rice farm. The R² of the linear relationship was slightly lower in this study (0.69; p < 0.001) than in Carracelas et al. [24] (0.73; p < 0.001), but still high despite the fact that differences in N uptake in the present study were caused by the crop management and the inherent variability within the bays as opposed to having N rate treatments. Furthermore, the equation and the statistically significant slope parameter obtained for SCCCI² were similar in both studies, even though it was conducted in other rice season with a different variety. Differences in predicted N uptake using either of these models are extremely low and would not affect the N recommended topdressing rates based on the guidelines tables developed for N topdressing for main rice varieties in Australia [56].

4.1. Validation of the SCCCI² Model for N Uptake Estimates

When assessing the performance of the model recently reported in a study by Carracelas et al. [24] that relies on SCCCI² measurements and that was developed in the same rice growing region of this study, but in a different growing season and using a different variety (Sherpa), there was a good agreement between the measured and estimated N uptake (Figure 9). The RMSE, Bias and MAPE obtained with the SCCCI² model (20.52, −6.21 Kg N ha⁻¹ and 11.95%) were notably lower than those obtained when using the Clg and Clre models reported by Brinkhoff et al. [48] (Figure 9a,b) and Carracelas et al. [24] (Figure 9d,e). Other single-vegetation-index models reported in the literature using the NDRE and the ratio index NIR/RE [57] and ratio NIR/R [58], provided worse results with RMSE and Bias values ≥113 and ≥112 Kg N ha⁻¹, respectively. Models using the NDRE² index [24,48] (Figure 9f,c) showed more similar results to those obtained with the SCCCI² model, but especially the model obtained from Carracelas et al. [24] (Figure S1), that had a RMSE of 21.95, Bias of 14.96 Kg N ha⁻¹, MAPE of 15.84% and the highest CCC of all the models, 0.83 The better result observed with the NDRE² model reported in Carracelas et al. [24] was more likely because it was developed using images collected at PI at five different times of the day (decreasing the variability observed in NDRE measurements through the day) on similar weather conditions (clear-sky days) and with the same multispectral camera used in this study.

To better confirm the validity of the SCCCI² model for N uptake estimates, the results obtained from using this single-vegetation-index model were compared with the results obtained using a stepwise multiple linear regression model (SMLR) developed by Zha et al. [59] from UAV-based multispectral images based on the Green Optimized Soil Adjusted Vegetation Index (GOSAVI), Modified Normalized Difference Red Edge (MNDRE2), and Modified Chlorophyll Absorption in Reflectance Index1 (MCARI1). The N uptake estimates using the SMLR model (Plant N uptake = −198.601 + 353.387 GOSAVI + 132.397 MNDRE2 − 91.552 MCARI1) had an RMSE of 19.68 and a Bias of −4.96 Kg N ha⁻¹ (R² = 0.77, p < 0.001), which were only 0.84 kg N ha⁻¹ and 1.25 kg N ha⁻¹ lower than those obtained using the SCCCI² model. These results validate the single-vegetation-index model reported by Carracelas et al. [24] and confirm the SCCCI² as a useful vegetation index for rice N uptake estimations at PI. The best performance of the SCCCI² model in estimating N uptake in comparison with the other single-vegetation-index models tested here could be due to the low influence that water depth and time of image collection have on this index with respect to the other chlorophyll-sensitive indices used in this study [24].

Additional validation studies across different locations, rice growing seasons, and varieties are required to further improve the reported SCCCI² model for broadly monitoring cultivated areas on commercial farms within the rice industry.

4.2. Remote Sensing Monitoring of Grain Milling Quality

All the vegetation indices explored at PI in this study were significantly correlated with the percentage of cracked grain, percentage of immature grain and quality score when data for both growing seasons were pooled together. Tsukaguchi et al. [30], reported a correlation between the Clg index, immature grains and chalky grain rate, which has been shown to be affected by basal N fertilization [60,61] and negatively associated with N uptake at heading [62]. Zhao et al. [63] also reported that optimal nitrogen application (~270 kg N ha⁻¹) improved rice grain quality, while excessive nitrogen application (345 kg N ha⁻¹) reduced it by lowering the average grain filling rate and prolonging the grain filling period. In the present study, no differences in chalky grain rate were observed between seasons and water management strategies.

The NDVI was the index that had the lowest R² for the relationship with all grain quality parameters, which agrees with that observed by Tsukaguchi et al. [30]. This result was most likely due to the early saturation of the NDVI in no N deficient crops in comparison with the other indices (Figure 6 and Figure 11) as also reported in previous studies [35,53,55,64], the role that plant N availability plays in grain quality [25,26,27,28] and the fact that the SCCCI², NDRE² and Clg indices are all more sensitive than NDVI to leaf chlorophyll content and indirectly to N content. All the chlorophyll-sensitive indices had a similar R² when plotted against the percentage of cracked grain (0.75–0.78) and the quality score (0.84–0.86). However, the strongest relationship with the percentage of immature grain was observed for the SCCCI². The findings of this study suggest that grain milling quality parameters could potentially be associated with N uptake at PI since all SCCCI², NDRE² and Clg were correlated with N uptake at this phenological stage. The percentage of grain cracking and quality score increased and decreased, respectively, when the SCCCI² reached a value between 0.4 and 0.5 (Figure 11). This value of the SCCCI² corresponded to a N uptake between 150 and 160 Kg N ha⁻¹, which could possibly indicate that N uptake values at PI above this threshold could lead to a higher percentage of grain cracked and grain milling quality. Previous research has indeed shown that high doses of N topdressing may increase the cracking percentage [65] and induce changes in the grain protein composition [25,27]. However, grain quality parameters are influenced not only by plant N status, but also genetic factors, agronomic practices [28], weather conditions before and after heading [28,30,66], harvest time [67], and post-harvest handling practices [28].

The study showed that there is a relationship between spectral indices sensitive to chlorophyll and grain quality parameters, although further research on this topic is needed to better understand how weather conditions and crop management at PI relates to remotely sensed indices.

5. Conclusions

The findings of the present study show that advanced canopy development in ponded rice with respect to high-yielding aerobic rice leads to differences in canopy reflectance (mainly in the NIR region) up to flowering when these differences are minimized. These canopy reflectance variations led to having lower values of the SCCCI², NDRE², and Clg indices in the aerobic than in the ponded rice even though yield was not reduced in the non-ponded strategy.

This study validated the SCCCI² model for N uptake estimates at panicle initiation in a different variety, site, and growing seasons from the ones the model was developed. The better performance of the SCCCI² model in estimating N uptake in comparison with single-vegetation-index models using other chlorophyll-sensitive vegetation indices and the similar performance to that of a stepwise multiple linear regression model confirm the suitability of the SCCCI² for N uptake monitoring.

Moreover, this work suggests that grain milling quality parameters could potentially be associated with N uptake at PI since all the chlorophyll-sensitive vegetation indices tested here (SCCCI², NDRE², and Clg) were highly and similarly correlated with grain milling quality parameters. Among the quality parameters assessed, the best correlation for all the chlorophyll-sensitive indices was observed with the quality score, with the SCCCI² providing the highest R² (0.85). Further research is required to better understand how the management of the crop at panicle initiation could affect the overall grain milling quality score and how this relates to remotely sensed indices and weather conditions.