Is It Possible to Predict the Timing of Mid−Season Drainage by Assessing Rice Canopy Light Interception?

Abstract

Mid−season drainage (MSD) is a widely used water management practice in rice (Oryza sativa L.) cultivation. However, the timing of the initiation and termination of MSD is highly arbitrary and subjective, and a quantitative indicator is lacking in precision agronomic practice. In this study, datasets (91 cases) from previous field experiments were obtained and used to fit adjusted exponential growth models, incorporating rice canopy light interception (CLI) and tillering development dynamics. Different criteria for initiating and terminating MSD were developed based on CLI indicators. The results showed that the CLI indicator at 80% of the projected panicle could be used to predict the initiation of MSD; however, it was highly variable, depending on the growing season and rice cultivated variety. The values for Indica rice were 0.26 to 0.31 and 0.31 to 0.42 in the late and single seasons, respectively, while the values for Japonica rice were 0.15 to 0.29 and 0.23 to 0.33, respectively. In addition, the CLI values at 7 to 10 days prior to panicle initiation (PI) ranged from 0.77 to 0.87 and from 0.56 to 0.83 for the Indica and Japonica varieties, respectively, and were calculated to determine MSD termination. The CLI indicators for MSD were greatly dependent on the cultivated variety, growing season, and planting method. The results of the correlation study and principal component analysis (PCA) indicated that the differences in CLI values attributed to rice type and growing season were driven by the tillering and canopy characteristic traits, respectively. Therefore, the current parameters could provide a reference for subsequent field applications in specific areas, but further optimization is needed to increase their robustness. To evaluate the usefulness of the CLI indicator for determining MSD initiation and termination, a set of devices for monitoring canopy light interception and water level was developed, and an on-farm trial was carried out in the middle and lower reaches of the Yangtze River in China in 2022. The field application demonstrated that MSD could be scheduled automatically based on the current system, and that the effect was consistent in practice.

1. Introduction

Mid-season drainage (MSD) is a water management practice that has been widely adopted in rice (Oryza sativa L.) cultivation in China [1], Japan [2], East South Asia [3], and other rice−producing countries or areas where irrigation and drainage are well managed. It was reported that the practice was initiated in northeastern China rice farming (early 1960s) in response to irrigation water shortages. MSD was not only reduce to water use, but also to increase crop yield [4].In general, MSD is carried out before the stage of maximum tillering with the goal of supplying oxygen to soil [5,6], keeping soil conditions oxidative temporally [7], removing toxic substances [3,8], and maintaining healthier roots [9,10], thus reducing ineffective tillers and decreasing CH₄ production potential in the soil [3,7] under anaerobic soil conditions.

However, there are various strategies for the timing of MSD for various purposes. In practice, Chinese agronomists generally recommend planning MSD based on the number of tillers (approximately 80–120% of expected panicles at harvest, which depends on the rice cultivated variety and/or growing season) for high yield output [11]. For example, MSD could be carried out when the tiller number reaches 80% of the projected panicle when growing the cultivar with great tillering producing capacity and heavy panicle within an adequate growing period, as with the hybrid and/or Indica varieties during a single season [12]. The threshold may be increased with improved tillering capacity and/or a reduction in the growing period, as with the inbred cultivated variety (especially Japonica rice) during the double season.

However, determining the timing of MSD requires in situ tillering counting, which is time−consuming and labor-intensive work, as well as good agronomic experience in the field. Furthermore, farmers often determine the need for MSD by observing the sparseness of the canopy and combining their field experiences, which makes the scheduling of MSD highly empirical and prone to error. Therefore, there is a demand for precision tools to assist farmers in improving the accuracy and reliability of MSD decisions.

Is it possible to use canopy characteristics to determine the timing for MSD? The canopy light interception (CLI) of solar radiation represents a crop’s ability to capture sunlight, which is a major component for determining the photosynthetic capacity of a crop [13]. The interception of solar radiation or light, as well as its distribution within a crop canopy, depends on the size, shape and orientation of canopy components [14]. For rice, the CLI is generally related to the canopy leaf area per unit of ground area, i.e., leaf area index (LAI), while the LAI for a given environment is generally affected by planting density, tillering emergence and leafiness per tiller [15], and modulated by water or mineral nutrition limitation during the vegetative period [16]. The evaluation of the CLI is generally carried out by measuring the solar radiation values of the upper and inside canopy using a solar radiation determining probe (i.e., AccuPAR or SunSCAN). Briefly, the light probe is placed above the canopy and 10 to 20 cm above the water or soil surface in succession to measure the light intensity above and inside the canopy, respectively, and the CLI is calculated as the percentage of incoming light intensity that was intercepted by the canopy [17]. The sensor for capturing photosynthetically active radiation is expensive, but light sensors for field stratification are relatively inexpensive to assemble, which makes it possible to customize canopy sensors for MSD decisions. Therefore, we hypothesized that for a given rice cultivated variety or rice type (with similar leaf characteristic traits), unmanned monitoring of the CLI can be implemented to predict the rice tillering dynamics for MSD decisions. However, few studies have been reported about using the canopy traits to predict the timing of MSD. In the current study, model simulations were used to quantify the tiller dynamics and light canopy interception during rice growth, and parameters were evaluated to predict the MSD based on CLI and tillering dynamic models.

In the current study, CLI and tillering dynamic data from two field experiments, differing in rice season, planting density, and rice varieties, were collected (approximately 91 datasets of rice growing for each condition). A logistic model and adjusted exponential model were adopted to simulate the CLI and tillering dynamics, respectively, in the target growing situation. The objectives of the study were as follows: (1) to evaluate the relationship between the CLI and tillering dynamics and define the scenarios for MSD, and (2) to explore the use of the CLI for determining the timing of MSD initiation and termination in each scenario. With the above objectives, we can quantify the judgement of MSD use in practical applications and implement machine control.

2. Materials and Methods

2.1. Site Description

Field experiments were carried out from 2015 to 2018 at the experimental farm of the China Rice Research Institute (120.2 E, 30.3 N, with an altitude of 11 m). The site is located in Southeast China, which is characterized by a subtropical monsoon climate with an annual mean temperature of 13 to 20 °C and a mean annual precipitation of 1200 to 1600 mm. The soil is a ferric−accumulic stagnic anthrosol with soil pH of 6.5 ± 0.5, soil organic matter of 22.3 ± 0.3 g kg⁻¹, total N of 1.98 ± 0.23 g kg⁻¹, available P of 39.4 ± 5.6 mg kg⁻¹, and available K of 35.6 ± 5.9 mg kg⁻¹. The meteorological data were obtained using a self-recording tintype meteorological station (R800, Technosolutions Ltd., Beijing, China).

2.2. Rice Cultivars and Experimental Design

2.2.1. Field Experiment I

Field experiment I was carried out with two inbred cultivars and two hybrid cultivars with five planting densities (40, 26.7, 20, 16.7, and 13.3 hills m⁻²) in 2015 and 2016 during a single season (May to October). Pregerminated seeds were sown in seedbeds on 22 June and 24 June and transplanted on 18 July and 19 July in 2015 and 2016.

The field experiment was arranged following a split−plot randomized complete block design in triplicate. The main plot treatments were planting densities of 40–26.7–20–16.7–13.3 hills m⁻². The split−plot was rice cultivars. Seeds were sown on the nursery bed in mid-May, and seedlings (approximately 3- to 4-leaf age) were transplanted into the puddle field at a designed density with 1 to 2 seedlings per hill for hybrid rice and 2 to 4 seedlings per hill for inbred rice. Urea, calcium superphosphate, and potassium chloride were applied in all the plots to yield the amounts of 142.5 kg ha⁻¹ N, 135 kg ha⁻¹ P₂O₅, and 101.3 kg ha⁻¹ K₂O, respectively. Approximately 50% N was incorporated as basal fertilizer one day before transplanting and topdressing at the tillering and booting stages at rates of 30% and 20% of the total N, respectively. Potassium fertilizer was applied at 50% and 50% as basal dressing and topdressing at panicle initiation, respectively. Phosphorous fertilizer was applied to each plot, with 100% as basal dressing. The other crop managements (i.e., irrigation) followed local high−yield practices. Weeds, insects, and diseases were controlled as required to avoid yield loss.

2.2.2. Field Experiment II

Field experiment II was performed with four rice types (namely, Indica rice, inbred Japonica rice, hybrid Japonica rice, and hybrid Indica/Japonica rice) from 2016 to 2018 during the late season (from June to November). Specifically, there were 16, 15, and 24 cultivars in 2016, 2017, and 2018, respectively.

Field experiments were conducted following a randomized block design in triplicate. Pregerminated seeds were sown in seedbeds on 23 June (±3) and transplanted on 20 July, 14 July, and 19 July in 2016, 2017, and 2018, respectively. The hill spacing was 26.7 × 13.3 cm, with 1 to 2 seedlings per hill for hybrid rice and 2 to 4 seedlings per hill for inbred rice. The total N application rate was 202.5 kg ha⁻¹; 50% was applied 1 day before transplanting, 30% was applied 14 days after transplanting (DAT), and 20% was applied at the panicle initiation (PI) stage. Approximately 100.0 kg ha⁻¹ P₂O₅ and K₂O were applied as a basal fertilizer, and an additional 65.0 kg ha⁻¹ K₂O was applied as a topdressing at PI. Management practices, such as pest and weed control and irrigation, were performed as needed to avoid yield losses.

The details of the two rice field experiments are listed in

Table 1. Detailed description of the datasets from two rice experiments used in the current study.

Experiment		Year	Growth Season	Cultivars	Planting Dense (Hills per m2)	No. of Dataset
Exp 1	2015	Middle season	Huanghuazhan, Chunyou84, Tianyouhuazhan, Xiushui09		27 20, 16, 13	16
	2016	Middle season	Huanghuazhan, Yongyou12, Tianyouhuazhan, Xiushui09		40, 27, 20, 16, 13	20
Exp 2	2016	Late season	Huanghuazhan, Jia58, Jia67, Nangeng46, Nangeng5055, Nangeng9108, Ninggeng1, Ninggeng2, Ninggeng3, Ninggeng4, Tianyouhuazhan, Xiushui09, Xiushui134, Yongyou1640, Yongyou2640, Yongyou538		24	16
	2017	Late season	Cliangyouhuazhan, Huanghuazhan, Jia58, Changyou5, Chunyou84, Jiaheyou218, Nangeng46, Nangeng9108, Ninggeng1, Tianyouhuazhan, Xiushui134, Yongyou1540, Yongyou538		24	15
	2018	Late season	Chunyou927, Huxianggeng151, Huanghuazhan, Jia58, Jiahe218, Jiayou5, Nangeng46, Nangeng9108, Ninggeng4, Tianyouhuazhan, Wuyugeng24, Wuyugeng6567, Wuyungeng6571, Xiushui134, Yongyou12, Yongyou15, Yongyou1540, Yongyou538, Yongyou540, Yongyou8		24	20

, and a total of 91 sets of rice growth data were obtained.

2.3. On−Farm Trial

A field trial was carried out with two treatments, namely, flooding irrigation (CK) and irrigation with MSD (T). An inbred Japonica cultivar (Jia58) was used and, based on previous data, CLI values of 0.35 and 0.80 were set for the initiation and termination of MSD, respectively. For irrigation management, the field water level was controlled at 3 to 5 cm above the soil in flooding irrigation and 0 to 5 cm above the soil before and after MSD in the MSD treatment group. Other agronomic functions were implemented according to the local farmers’ practices.

To ensure that the MSD can be controlled by real−time CLI changes in the field, the researchers developed a device for real−time CLI and field water level monitoring. The device consists of three parts: (1) the host control system, which is used to receive/store/process sensor data; (2) the light sensor group, which is used to collect CLI data; and (3) the field water level module (Figure 1). The canopy monitoring device was approximately 140 to 160 cm in height with 10 light sensors, which were divided into two groups: an upside light sensor group with two sensors 100 to 110 cm above the soil and a downside light sensor group with eight sensors 20 to 30 cm above the soil. The field water level sensor was developed based on the laser sensor, with millimeter−scale sensitivity. The skeletons of both devices were made of PVC, and the LoRa protocol was adopted for wireless data transmission. All parts were powered by solar panels. The process of controlling MSD was as follows (Figure 2). The irrigation was operated by a stabilized pressure pump and an electrically operated valve based on the signal from the system.

2.4. Sampling and Measurement

Panicle initiation was determined by observing the development of the panicle primordia by stripping the stem. The full heading stage was defined as the date when 95% of the panicles had emerged, and the plant reached physiological maturity when 95% of spikelets had turned yellow. The growth period was defined as the days from seed sowing to plant physiological maturity.

For each plot, a concessive 15 to 20 hills were labeled, and the number of tillers was recorded at intervals of 7 to 10 days after plants had recovered from transplanting shocking.

At maturity, the number of panicles per m² was evaluated by counting approximately 15 to 20 successive hills and converting the values’ unit area numbers based on planting density. The hills from the center part (approximately 5 m²) were used to determine the grain yield. Unhulled (rough) rice grain was obtained after reaping, threshing, and wind selection. The weight of the rice grain was adjusted to a moisture content of 14%.

CLI was measured between 1100 and 1300 h at intervals of 7–10 days and 15–20 days before and after heading, respectively, using the AccuPAR LP−80 (LP−80, Decagon Devices Ltd., Pullman, WA, USA). In each plot, the light bar was placed above the canopy and 10 to 20 cm above the water surface in succession to measure the light intensity above and inside the canopy, respectively. Three measurements were taken within rows and another three between rows.

The CLI rate was calculated as the percentage of incoming light intensity that was intercepted by the canopy [18].

$$\mathrm{CLI}=1-\frac{{\mathrm{LI}}_{below}}{{\mathrm{LI}}_{above}}$$

(1)

where LI_below is the light intensity value measured at 10 to 20 cm above the field surface and LI_above is the light intensity value measured above the canopy.

The tillering dynamic was fitted by an algorithmic model with one independent variable, i.e., the degree−day statistical model [19]:

$$y=a_0+\left(af-a_0\right)e^{\frac{-c{\left(d-T\right)}^2}{T^2}}$$

(2)

where y is the tiller number per unit area; a₀ is the tiller number per unit area at transplanting; af is the maximum tiller number per unit area; T is the cumulative accumulated temperature after transplanting; and c and d are cultivar parameters.

The degree−day statistical model predicts tillering dynamics based on the effect of degree−days on tillering dynamics. When sufficient fertilizer and water are supplied and seedlings are transplanted at the same spacing density, temperature and solar radiation are the two key factors affecting rice tillering dynamics [20,21]. Because temperature and solar radiation are highly correlated, several tillering models have been based on the relationship between tillering dynamics and degree days [20,22]. The CLI dynamic was fitted by an adjusted logistic model with the degree−day statistical model [20]:

$$\mathrm{CLI}=\frac{CL{I}_{max}\times CL{I}_{init}\times e^{rT}}{CL{I}_{max}+CL{I}_{init}\times (e^{rT}-1)}$$

(3)

where CLI is the canopy light interception rate and CLI_max is the maximum canopy light interception rate. CLI_init is the canopy light interception at transplanting; T is the cumulative accumulated temperature after transplanting; and r is the steepness of the curve, which is cultivated variety−dependent.

Three indicators were calculated for evaluation in determining the initiation of MSD:

(1)

The CLI value at the time when the tiller number reached 80% of the projected panicle number. In the current study, the panicle number at harvest was assumed to be the expected panicle number.

$${\mathrm{T}}_{\mathrm{TL}\_0.8}=\arg {\min }_T\left|a_0-{\mathrm{y}}_{0.8}+\left(af-a_0\right){\mathrm{e}}^{-\frac{\mathrm{c}{\left(\mathrm{d}-\mathrm{T}\right)}^2}{{\mathrm{T}}^2}}\right|,\mathrm{T}ϵ\left[{\mathrm{T}}_{a0},{\mathrm{T}}_{af}\right]$$

(4)

where T_{TL_0.8} is the cumulative accumulated temperature at the time when the tiller number reached 80% of the projected panicle number; y0.8 is the tiller number per unit area of 80% of the projected panicle number; a₀ is the tiller number per unit area at transplanting; af is the maximum tiller number per unit area; c and d are cultivar parameters; and T is the cumulative accumulated temperature after transplanting.

(2)

The CLI value at the time of maximum increases with tillering production. The derivative of the tillering−producing model between the stages of initiation and maximum tillering was calculated, and the CLI value at the maximum derivative was calculated.

$${\mathrm{T}}_{\mathrm{TLPmax}}=\arg {\max }_T\frac{2\mathrm{d}\times \mathrm{c}\left(af-a_0\right)\left(\mathrm{T}-\mathrm{d}\right)}{{\mathrm{T}}^3}{\mathrm{e}}^{-\frac{\mathrm{c}\left(d-\mathrm{T}\right)}{{\mathrm{T}}^2}},\mathrm{T}ϵ\left[{\mathrm{T}}_{a0},{\mathrm{T}}_{af}\right]$$

(5)

where T_TLPmax is the cumulative accumulated temperature at the time at which the maximum increase in tillering occurs; a₀ is the tiller number per unit area at transplanting; af is the maximum tiller number per unit area; c and d are cultivar parameters; and T is the cumulative accumulated temperature after transplanting.

(3)

The CLI value at the time of maximum increases with CLI development. The derivative of the CLI development model between the stages of initiation and maximum CLI was calculated, and the CLI value at the maximum derivative occurred was calculated.

$${\mathrm{T}}_{LIPmax}={\arg \max }_{\mathrm{T}}\frac{-\mathrm{r}CL{I}_{max}CL{I}_{init}{\mathrm{e}}^{rT}\left(CL{I}_{max}-CL{I}_{init}\right)}{{\left[CL{I}_{max}+CL{I}_{init}\left({\mathrm{e}}^{\mathrm{rT}}-1\right)\right]}^2},\mathrm{T}ϵ\left[{\mathrm{T}}_{CLIinit},{\mathrm{T}}_{CLImax}\right]$$

(6)

where T_LIPmax is the cumulative accumulated temperature at the time at which the maximum increase in CLI occurs and CLI_max is the maximum canopy light interception rate. CLI_init is the canopy light interception at transplanting; T is the cumulative accumulated temperature after transplanting; and r is the steepness of the curve, which is cultivated variety−dependent.

In addition, three CLI values were calculated to determine the end of MSD using Equation (3).

(1)

The CLI value PI;

(2)

The CLI value at 7 days prior to PI;

(3)

The CLI value at 10 days prior to PI.

The T values for CLI calculation were based on the date of PI and the climate data.

2.5. Data Analysis

Analysis of variance (ANOVA) was applied to the data using SAS statistical software version 8.0 (SAS Institute Inc., Cary, NC, USA). ANOVA is the most efficient parametric method available for the analysis of data from experiments; here, we used the linear model of an ANOVA [23]:

x_ij = μ+ ai + e_ij

(7)

The subscript ‘i’ is used to denote the group or class (i.e., the treatment group), with ‘i’ taking the values ‘1 to a’; whereas the subscript ‘j’ designates the members of the class, with ‘j’ taking the values ‘1 to n’ (hence, ‘a’ groups and ‘n’ replicates per group).

The model dynamic LI and tillering measures were implemented with the mclust package version 6.0.0 of R software version 4.2.1 (University of Auckland, Auckland, New Zealand), and the mco package version 1.15.6 with nondominated sorting genetic algorithm II (NSGA−II) [24] was used to estimate the parameters of the model, as well as the T value calculation based on the tillering model. NSGA−II is an algorithm for solving combinatorial optimization problems. In our study, we used NSGA−II to simulate 91 datasets from the tiller model and the canopy intercept rate model to obtain corresponding data with minimum mean squared error and to make graphs based on these 91 datasets. The packages boxplot version 2.4.0 and ggplot2 version 3.4.0 were used to generate the plots.

3. Results

3.1. Simulation of Dynamic Tillering DevelopmenSSt and Canopy Light Interception

A total of 91 datasets were analyzed, and the dynamics of tillering development and CLI for each dataset were simulated according to Jiang’s method [19] and the logistic function, respectively (Figure 3). A detailed summary of the simulation results is provided in

Table 2. Summary statistics for the CLI value of MSD initiation of different criteria.

Season	Cultivated Variety Type	Mean	SE	Median	Quantile (25%)	Quantile (75%)
		MSD start at 0.80 of tiller to panicle
Late	hybrid Indica	0.31	0.03	0.27	0.27	0.34
	hybrid Japonica	0.23	0.02	0.22	0.18	0.29
	Inbred Indica	0.28	0.03	0.31	0.26	0.31
	inbred Japonica	0.20	0.01	0.20	0.15	0.23
Single	hybrid Indica	0.35	0.02	0.36	0.32	0.38
	hybrid Japonica	0.25	0.01	0.23	0.23	0.25
	inbred Indica	0.38	0.02	0.40	0.31	0.42
	inbred Japonica	0.28	0.02	0.27	0.25	0.33
		MSD start at max deriv of TL model
Late	hybrid Indica	0.26	0.04	0.22	0.18	0.32
	hybrid Japonica	0.17	0.02	0.15	0.10	0.25
	inbred Indica	0.23	0.04	0.26	0.20	0.27
	inbred Japonica	0.11	0.01	0.10	0.07	0.13
Single	hybrid Indica	0.23	0.02	0.26	0.17	0.29
	hybrid Japonica	0.18	0.01	0.17	0.17	0.19
	inbred Indica	0.23	0.02	0.22	0.22	0.28
	inbred Japonica	0.15	0.01	0.15	0.14	0.18
		MSD start at max deriv of LI model
Late	hybrid Indica	0.49	0.00	0.49	0.48	0.49
	hybrid Japonica	0.48	0.00	0.48	0.47	0.49
	inbred Indica	0.49	0.00	0.49	0.49	0.49
	inbred Japonica	0.46	0.00	0.46	0.46	0.47
Single	hybrid Indica	0.47	0.01	0.47	0.46	0.48
	hybrid Japonica	0.48	0.00	0.48	0.48	0.49
	inbred Indica	0.47	0.00	0.47	0.47	0.48
	inbred Japonica	0.48	0.01	0.47	0.47	0.49

. The Root Mean Square Errors (RMSEs) for tillering development and the CLI dynamic simulation ranged from 4.24 to 56.9 tillers per m² and 2.24 to 7.72%, respectively. The parameters for the dynamic development of tillers and CLI varied with the growth season, rice type, and growing years.

3.2. The CLI Value for Initiation and Termination of MSD

Three criteria for determining the initiation of MSD were tested: 80% of the projected panicle at maturity, the points corresponding to the maximum increase in tillering production, and CLI. The CLI values for each criterion were calculated based on the models.

The results showed that CLIs at 80% of the projected panicle at maturity for MSD initiation varied with the growing season as well as the rice type. In the late season, the 25%−quantile to 75%−quantile of CLI was 0.27 to 0.34 (n = 5), 0.26 to 0.31 (n = 3), 0.18 to 0.29 (n = 19), and 0.15 to 0.23 (n = 28) for hybrid Indica, inbred Indica, hybrid Japonica, and inbred Japonica varieties, respectively. For a single season, the values were 0.32 to 0.38 (n = 9), 0.31 to 0.42 (n = 9), 0.23 to 0.25 (n = 9), and 0.24 to 0.33 (n = 9), respectively. The CLI value at MSD initiation was greater in the single season than in the late season, and higher in Indica rice than in Japonica rice. The CLI value of hybrid rice did not differ from that of inbred rice. Similar to the CLI at 80% of the projected panicle, the CLI values when MSD was initiated at the maximum tiller production rate were also related to the rice type and growing season. However, the values were smaller than those at 80% of the projected panicle; specifically, the 25%−quantile to 75%−quantile CLI values were 0.8 to 0.12 smaller. The CLI values when MSD was initiated at the maximum CLI rate averaged from 0.46 to 0.49, irrespective of the growing season and varieties.

The CLIs at the maximum CLI rate were consistent and cultivated variety and growing season had little effect. However, the tiller number at these CLIs varied greatly among experimental conditions (Figure 4), making the indicator of limited value in agronomic practice. The CLIs at the maximum tiller production rate were too small to determine MSD initiation and showed exhibited unstable performance depending on the cultivated variety and growing seasons, making it difficult to apply in practice. The CLI values at 80% of the projected panicle were relatively consistent across the groups, except for late−season Japonica rice (Table 2). Therefore, compared with the other two indicators, it showed more promise for practical use.

The Timing of the CLI Value for MSD Termination

In practice, the duration of MSD depends on the weather (i.e., temperature, precipitation), and the termination of MSD is based on tillering development (or performance of tillering control), which is difficult to quantify. However, rice panicle initiation is sensitive to water deficiency. Therefore, MSD must be terminated before PI; in practice, this was 7 to 10 days before panicle initiation. The CLI values were also calculated at 0, 7, and 10 days before PI (Figure 5).

The differences in CLI values for MSD termination varied depending on rice cultivated variety and growing seasons. In general, the CLI value for Indica (both inbred and hybrid) and hybrid Japonica cultivars did not differ remarkably at 0, 7, and 10 days before PI; the values were higher than those of the inbred Japonica rice type, irrespective of growing season. The average CLIs at PI were 0.92, 0.89, 0.91, and 0.79 for hybrid Indica, inbred Indica, hybrid Japonica, and inbred Japonica cultivars, respectively. The corresponding CLI values averaged 0.87, 0.82, 0.83, and 0.67 at 7 days before PI, and 0.83, 0.77, 0.76, and 0.58 at 10 days before PI. Therefore, according to common farming practice, the CLI for MSD termination ranged from 0.83 to 0.87, 0.77 to 0.82, 0.76 to 0.83, and 0.58 to 0.67 for hybrid Indica, inbred Indica, hybrid Japonica, and inbred Japonica, respectively.

However, large variations in CLI values were found among samples. The standard deviations of inbred Japonica rice values were 8.4%, 10.0%, and 11.0% at PI, 7 days before PI, and 10 days before PI, respectively, and were greater than those of the other rice types. This Indicated greater variability in tillering canopy development among inbred Japonica cultivars. The results for different varieties and growing seasons of Indica rice were relatively consistent. In addition, the CLI value at the three stages was also affected by planting density.

3.3. Analysis of Tillering Canopy Development Affects the Application of CLI

The tillering and CLI parameters in the dynamic simulation reflected the characteristic traits of tillering and canopy development during growth. The CLI value for MSD initiation was positively correlated with the CLI_init (p < 0.001) in the CLI model and the d value in the tillering model (p < 0.01) but negatively correlated with the r value in the CLI model (p < 0.001).

The CLI value 7 days before PI was positively correlated with the c value (p < 0.001) and init_LI (p < 0.01) in the tillering dynamic model and negatively correlated with the d value (p < 0.001) and ini_TL (p < 0.01). The correlation between CLI values and model parameters for the previous 10 days of PI was similar to that for the previous 7 days of PI.

The CLI value 10 days before PI was positively correlated with the c value (p < 0.001) and CLI_init (p < 0.001) of the tillering dynamic model, while the d value was negatively correlated with a0 (p < 0.01). The CLI values for MSD termination were closely related to the c and d values in the tillering dynamic model and CLI_init in the CLI model (

Table 3. Pearson’s correlation between the CLI values and tillering and canopy light interception dynamic simulation parameters.

	c_TL	d_TL	a0	af	r_CLI	CLIinit	CLImax
CLI MSD initiation	0.12 ns	0.30 **	−0.07 ns	−0.03 ns	−0.47 ***	0.56 ***	0.15 ns
CLI 7 days before PI	0.37 ***	−0.35 ***	−0.28 **	0.08 ns	0.05 ns	0.27 **	0.24 *
CLI 10 days before PI	0.38 ***	−0.27 **	−0.30 **	0.00 ns	−0.06 ns	0.34 ***	0.24 *

Note: Statistical significance is represented by asterisks (* p < 0.05, ** p < 0.01, *** p < 0.001); ns is the duration.

).

PCA was performed on the parameters of the LI model and TL model for different growing seasons and types of rice varieties. Cases from different growing seasons were distinguished by Dim 1, which was dominated by LI traits. Dim 1 was positively correlated with the c value but negatively correlated with the initiation of CLI in the LI model. The results suggested a larger CLI and quick emergence of the canopy in the late season, rather than a single canopy. In the analysis, different rice types were distinguished by Dim 2, mainly by the tiller dynamics model. Dim 2 was positively correlated with the c value and negatively correlated with the d value in the tiller dynamics model. The results showed that the CLI value difference due to rice type depended on the development of tillering (Figure 6).

3.4. On−Farm Trial

To evaluate the performance of the device in practice, a field trial with two treatments, namely, flooding irrigation (CK) and irrigation with MSD (T), was carried out in a farmer’s paddy in Pinghu, Jiaxing, Zhejiang, during a single season in 2022.

Table 4. Costs of CLI system and traditional irrigation.

	CLI Irrigation System		Traditional Irrigation
Items	Light sensor (¥)	300	Labor costs (¥)	30–50
	Water level sensor (¥)	150	Water and power costs (¥)	0–10
	Water pumps (¥)	500		−
	Equipment installation (¥)	400		−
	Controllable areas (667 m2)	20–50		1
	Available years (y)	6		1
Average (¥·y−1·667 m−2)		4.5–16.8		30–60

shows the costs of the current system versus the traditional irrigation management. The average costs per area per year of the system can be reduced by at least 42% in compared to the traditional management system. The daily CLI values during rice canopy growth are shown in Figure 7a. The date of MSD initiation for CK and T were similar, with a 2−day delay in CK. However, the date of termination of MSD for the two treatments differed dramatically, with the termination of T being 6 days earlier than that of CK. This might be associated with the canopy response to drought stress during MSD. The dates of MSD initiation were 14 to 19 days prior to PI, which was relatively robust for MSD termination. However, it satisfied the water requirement criteria for PI development.

The daily field water level data were also collected and are shown in Figure 7b,c. The water level for CK was maintained between 3 and 5 cm during the vegetative stage with frequent irrigation. The irrigation was terminated during MSD. The field water level showed a dramatic drop a few days after the initiation of MSD and maintained values ranging from −200 mm to −150 mm during MSD. After the recovery of irrigation, the field water level increased and remained in the range of 0 to 5 cm on average. These results demonstrated that the current method for MSD prediction could be used in practice.

4. Discussion

4.1. CLI Indicators for Predicting the Initiation and Termination of MSD

MSD is a water management method adopted at the maximum tillering stage with multiple purposes, including to control nonproductive tillers [3]. However, the timing of the initiation and termination of MSD is still very subjective in practice, and the decisions are not typically based on quantitative criteria. A previous report described a typical MSD protocol: the MSD started at a certain number of days (i.e., 25 days) after rice transplanting and remained drained for 7 days, and MSD was terminated when some small cracks were observed on the soil surface [25]. Overall, this is a labor−intensive and highly subjective method. Moreover, the maximum tillering stage is difficult to predict due to its varietal dependence. In the current study, we aimed to establish observable and predictable metrics for managing MSD performance. Three CLI indicators were developed based on the tillering and CLI models to predict the initiation of MSD (Figure 2). The CLI at the maximum tillering production rate, which was calculated to represent the initiation of the maximum tiller stage, however, was highly variable, depending on growing season and rice cultivated variety, and its value was too small in some samples (Figure 1 and Figure 2). The increase in maximum CLI ranged from 0.46 to 0.49 (Table 2) but due to the heterogeneity of canopy and tiller dynamics among varieties, it was difficult to predict the tillering stage based on this indicator. Therefore, we did not choose these two as candidates.

The number of tillers reached at 80% of the projected panicle is a theoretical criterion in agronomic practice in China [22]; however, it is difficult to implement and evaluate precisely. The CLI values calculated based on when the tillers reached 80% of the projected panicle also varied by growing season and rive cultivated variety; however, in the analysis, the values could be separated by rice cultivated variety and growing season. For the Indica cultivated variety, the 25% to 75%−quantile CLI values were 0.26 to 0.31 (n = 8) and 0.31 to 0.42 (n = 18) for the late and single season, respectively. For the Japonica cultivated variety, the values were 0.15 to 0.29 (n = 47) and 0.23 to 0.33 (n = 18) for the late and single seasons, respectively. These parameters provide a reference for the subsequent application and optimization of the model.

The CLI value for predicting the termination of MSD was also calculated based on the date of PI as well as the climate date. The basic criterion for resuming irrigation is to meet the water requirement during panicle initiation [26]. Therefore, initiating irrigation 7 to 10 days before PI is a robust strategy in common practice. The CLI values for MSD termination also varied based on cultivated variety and growing season. In summary, the CLI values from 10 to 7 days prior to PI ranged from 0.77 to 0.87 and 0.56 to 0.83 for the Indica and Japonica varieties, respectively. In addition to preventing panicle initiation from drought stress, MSD also needs to be interrupted due to excessive drought stress. Observing the small cracks on the soil surface [27] and testing the soil hardness by stepping on it [28] were also adopted as methods for determining the timing of MSD termination. In the current system, a protective lower water limit was set during MSD to prevent the plants from experiencing severe drought stress (Figure 7c). Therefore, only CLI values predicting the PI were needed.

4.2. Factors Influencing the Application of the Model

The CLI value for predicting the initiation and termination of MSD was closely associated with the plant architecture, which consists of the following: (i) the rate of emergence of leaves; (ii) the angle of leaves produced from each axis; and (iii) the tillering dynamics (emergence and growth arrest) [29]. The interplay between the dynamics of architecture and tillering production is perceived by each plant component [30]. However, few parameterizations of plant architecture dynamics that consider both leaf and tiller growth have been developed. A simple framework frequently used is that tillers emerge in coordination with the leaves on the main stem, the so−called “phyllochron” [31]. Therefore, if the canopy/leaf stage is frequently monitored, the arrest of the progress of the canopy/leaf stage is thus a potential early marker indicating the status of the tiller [32]. However, accurately reconstructing canopy/leaf development is a challenge due to variations among specific cultivars and growth conditions, and competition in plants is highly influenced by the availability of light, water and nutrients [33,34,35,36,37]. Therefore, a large number of observations might be needed to build a descriptive model for practical use [38]. In the current study, 91 datasets were employed to illustrate the tillering and CLI interplay effect. Based on the PCA, the differences in cases between growing seasons might be dominated by the CLI parameters, while parameters in the TL model might be used to distinguish the varietal differences (Figure 4). According to the current findings, providing a standard set of parameters (CLI values) is ideal for a sophisticated cropping system, including cultivated variety, density and fertilizer application, although this is time−consuming. However, it is also feasible to provide a set of CLI parameters for a typical rice cultivated variety in certain growing seasons (Table 3). In addition, the on−farm trial verified that the constant CLI value could be applied for managing MSD in practice (Figure 7), and that the system we used is less costly than traditional irrigation; thus, it could be used not only in subsequent studies but also for replication.

5. Conclusions

Canopy development and tillering dynamics in different growing seasons and for different rice varieties were used to fit two models. Three indicators (80% of the projected panicle at maturity, the maximum increase in tillering production, and CLI) were developed for predicting the timing of MSD. The results showed that the indicator at 80% of the projected panicle exhibited the best performance for predicting the initiation of MSD; furthermore, a set of reference CLI values of MSD controlling was proposed for different scenarios. To evaluate the CLI indicator for MSD initiation and termination, a set of devices for monitoring CLI and water levels was developed, and an on−farm trial was carried out in 2022. The field study demonstrated that MSD could be scheduled automatically based on the current system, and that the effect was consistent with actual practice.