*2.4. Water Balance Approach*

Based on the schema in Figure 1, water balance analysis in each plot was analyzed by the following equation:

$$
\Delta \text{WL}(\mathbf{t}) = \mathbf{I}(\mathbf{t}) + \mathbf{P}(\mathbf{t}) - \text{DR}(\mathbf{t}) - \text{ETa}(\mathbf{t}) \tag{1}
$$

where ΔWL is the change of water level in the bucket (mm), I is irrigation (mm), P is precipitation (mm), DR is drainage or overflow from the plot (mm) and ETa is actual evapotranspiration (mm). Here, the plot was designed with zero percolation and seepage. In the inlet and outlet, there was a water meter to measure irrigation and drainage. However, there was water loss by overflow when heavy rain events occurred and low pressure of water flow was not recorded. Therefore, the Excel Solver and ETa adjusted the parameters of I and DR by minimizing the following objective function:

$$\mathbf{F(x)} = \sum\_{\mathbf{t}=1}^{n} |\Delta \mathbf{W} \mathbf{L}\_{\mathbf{o}}(\mathbf{t}) - \Delta \mathbf{W} \mathbf{L}\_{\mathbf{m}}(\mathbf{t})| \tag{2}$$

The constraints:

$$I \ge 0; DR \ge 0; ETa \ge 0\tag{3}$$

where ΔWLo is the change of observed water level (mm), and ΔWLm is the change of estimated observed water level by the Excel Solver (mm), t is the day after transplanting (DAT) and n is total cultivation days. Since the Excel Solver only estimated 200 data in one process, the adjustment process was performed four times according to plant growth stages. They were initial (1–24 DAT), crop development (25–64 DAT), mid-season (65–87 DAT) and late-season stages (88–110 DAT).

Weather data were used to determine reference evapotranspiration according to a standard model by the FAO Penman-Monteith equation [28], which is derived based on the aerodynamic and canopy resistance, given by the following equation:

$$ET\_o = \frac{0.408\Delta (R\_\text{n} - G) + \gamma \frac{900}{T\_{\text{ave}} + 273} \mu (\mathbf{e}\_s - \mathbf{e}\_a)}{\Delta + \gamma (1 + 0.34\mu)} \tag{4}$$

where *ETo* is reference evapotranspiration on a daily basis (mm), *Rn* is net radiation received at crop surface (MJ/m2/d), *G* is soil heat flux density (MJ/m2/d), *Tave* is air temperature (◦C), *u* is wind speed at 2 m height (m/s), *es* is saturation vapor pressure (kPa), *ea* is actual vapor pressure (kPa), *γ* is psychrometric constant (kPa/◦C) and Δ is the slope of vapor pressure curve (kPa/◦C). The input data to calculate *ETo* were solar radiation, minimum, average and maximum air temperature, relative humidity and wind speed at 2 m height on a daily basis. Moreover, the information regarding the location (elevation and latitude) was needed, as well as Julian's day. Detailed derived equations and their procedure calculations of *ETo* can be referred to by Allen et al. [28].

*ETo* and *ETa* can be used to determine and adjust crop coefficient (*Kc*) by the following equation:

$$K\_c = \frac{ET\_a}{ET\_o} \tag{5}$$

Water productivity and water-use efficiency index were used to evaluate the performance of each regime. There are two definitions of water productivity adopted in this study. Firstly, water productivity is defined as total production per total water input; secondly, water productivity is total production per total water evaporated and transpired and they are expressed in g grain/kg water [29]. Meanwhile, the water-use efficiency index is crop yield per unit of water supplied [30]. Accordingly, the equation of water productivity and water-use efficiency index is given by the following equation:

$$\text{WP}\_{\text{I}+\text{P}} = \frac{100\,\text{Y}}{\text{I}+\text{P}} \tag{6}$$

$$\text{WP}\_{\text{ETa}} = \frac{100\,\text{Y}}{\text{ETa}}\tag{7}$$

$$\text{WUE} = \frac{100\text{Y}}{\text{I}}\tag{8}$$

where Y is grain yield (ton/ha), 100 is a conversion factor, WPI+P is water productivity by total inflow (irrigation and precipitation) (g grain/kg water), WPETa is water productivity by actual evapotranspiration (g grain/kg water) and WUE is water use efficiency index (g grain/kg water).

The water level of the setpoint was compared to the observed to evaluate the performance of evapotranspirative irrigation by root mean square error (*RMSE*):

$$RMSE = \sqrt{\sum\_{i=1}^{n} \frac{\left(WL\_{set} - WL\_{\circ}\right)^{2}}{n}} \tag{9}$$

where *WLset* is water level setpoint (cm), *WLo* is actual water level (cm) and *n* is cultivation days.

A significant test was performed by a single factor analysis of variance (ANOVA) to elucidate the effects of irrigation regimes on crop performance, water productivities and water use efficiency. The differences among regimes on all parameters' means were then compared using the least significant difference (LSD) at the 0.05 probability level (α = 0.05).

#### *2.5. Weather Condition during the Season*

Figure 2 shows the fluctuations in weather parameters, especially air temperature, relative humidity and wind speed. Air temperature is presented in minimum, maximum and average values. Despite fluctuating, air temperature conditions remain relatively constant throughout the growing season. It can be referred to as the gradient value of the linear equation, which was relatively low (<0.01). The maximum air temperature reaches 36.3 ◦C, while the minimum and average air temperatures reach 20.5 ◦C and 26.8 ◦C, respectively.

**Figure 2.** Air temperature, relative humidity, and wind speed fluctuations during planting season.

On the other hand, the relative humidity was found to decrease slightly. The gradient value was higher than the linear equations of air temperature; however, the value was low (<0.1). During one growing season, the consecutive minimum, average and maximum relative humidity values were 65.4%, 80.1%, and 91.0%, respectively. For the wind speed, the fluctuation was between 0 and 0.5 m/s, which indicated low wind speed in the field location (<1 m/s). In addition, its gradient was also relatively low (<0.01) by means there that even fluctuated; however, there was no significant change in the trends. The minimum, average and maximum wind speed values were 0, 0.1, and 0.5 m/s, respectively.

Another parameter, solar radiation, also showed a slightly decreasing trend, as depicted in Figure 3. The reference evapotranspiration is also presented in the figure. At the beginning of the growing season, solar radiation reached around 15 MJ/m2/d with reference evapotranspiration of 3 mm. Then, the reference evapotranspiration and solar radiation fluctuated; however, the trend was similar to other weather parameters. The gradient of the linear equation was low (<0.01), which represented no significance in raising and decreasing those parameters. At the end of the season, the value of solar radiation was around 14 MJ/m2/d with reference evapotranspiration being lower than 3 mm. The maximum, average and minimum values of solar radiation were 19.9, 14.1 and 5.9 MJ/m2/d, respectively. At the same time, the reference evapotranspiration values were 1.1, 2.9, and 4.2 mm for minimum, average and maximum, respectively.

**Figure 3.** Solar radiation and reference evapotranspiration fluctuation during the planting season.

The linear relationship between the reference evapotranspiration and the weather parameters is presented in Figure 4. Among the four parameters, solar radiation has the most substantial relationship to the reference evapotranspiration, represented by the highest R<sup>2</sup> value. The value of R<sup>2</sup> was close to 0.95, indicating that solar radiation has the highest contribution to the variability of evapotranspiration. The second parameter that has a major influence on the reference evapotranspiration was relative humidity, followed by the air temperature and the wind speed. This relationship indicated that solar radiation most influences the evapotranspiration process through the soil surface and plants [31]. Based on the sensitivity analysis study, solar radiation is the most sensitive parameter to changes in evapotranspiration [32].

**Figure 4.** The linear correlation among reference evapotranspiration and weather parameters: (**a**) reference evapotranspiration vs. solar radiation; (**b**) reference evapotranspiration vs. air temperature; (**c**) reference evapotranspiration vs. relative humidity; (**d**) reference evapotranspiration vs. wind speed.

### **3. Results**

#### *3.1. Performance of Evapotranspirative Irrigation*

The actual condition of the water levels in the CFI regime for replications 1 and 2 (CFI-1 and CFI-2) are presented in Figure 5. In this regime, inundation with a water level of 2 cm above the soil surface was used as the set point. The water level fluctuated and was close to the set point; however, high fluctuation occurred when there was a high rain intensity event. There was a significant increase in water level, especially at 20 DAT, both in CFI-1 and CFI-2. Heavy rainfall of 26.2 cm caused an increase in water level from 1.5 cm to 4.9 cm in CFI-1 and from 2.5 cm to 6.4 cm in CFI-2. The same situation occurred at 26 DAT when 50.2 mm of rainfall contributed to raising in water level from 2 cm to 4.6 cm of CFI-1 and 1 cm to 5.5 cm of CFI-2.

**Figure 5.** The actual field condition of water levels: (**a**) CFI-1; (**b**) CFI-2.

On the other hand, water levels tend to be lower when no rain event occurs for several days. As at 32–42 DAT, the water level decreased from 3.6 cm to 0.8 cm. Although it was set at the same setpoint, CFI-1 showed better performance. The average water levels were 2 cm and 2.4 cm for CFI-1 and CFI-2, respectively. Even though they fluctuated, the water levels were close to the desired level, indicating that the evapotranspirative control system worked well in this regime.

Figure 6 shows the fluctuations in water levels of the MFI regime in both the first replication (MFI-1) and the second one (MFI-2). The actual water levels fluctuated and were a little bit far from the set point. The actual water level is higher than that of the setpoint, particularly MFI-2. The water levels were lower to the setpoint only at the end of the growing season. When rainfall with high intensity occurred, it caused water levels in the field to increase. As at 22 DAT, after 70.8 mm of rain, the water level increased by 4.6 cm and 1.8 cm for MFI-1 and MFI-2, respectively. As per the same situation on the CFI regime, lower water levels generally occurred when no rain event happened, such as from 32 to 42 DAT. At this time, the water level tended to decrease from 1.8 cm to 0.5 cm. The average water levels were 1 cm and 0.9 cm for MFI-1 and MFI-2, respectively. This indicated that the evapotranspiration control system was slightly accurate in controlling the water level.

**Figure 6.** The actual field condition of water levels: (**a**) MFI-1; (**b**) MFI-2.

As previously mentioned, there were two setpoints in the WSI regime, i.e., 0 cm at 0–20 DAT and −5 cm afterward. As presented in Figure 7, the water level in both the first replication (WSI-1) and second replication (WSI-2) was well controlled at 0–20 DAT with the first set point. There were no significant fluctuations, and the water levels were close to the setpoint even though there was low rain intensity. The average water level in this phase is −0.1 and 0.5 cm for WSI-1 and WSI-2, respectively. Then, high fluctuation occurred when the water level was dropped to −5 cm. In this period, as per the same situation in two other regimes, high rainfall events occurred. The average water levels in the stage were

−4.6 cm and −3.3 cm for WSI-1 and WSI-2, respectively. These results indicated that the performance of WSI was slightly accurate, and both plots can be conditioned to be drier than the other two regimes.

**Figure 7.** The actual field condition of water levels: (**a**) WSI-1; (**b**) WSI-2.

RMSE values of the CFI regime showed the lowest level, indicating that the CFI plot was the best in controlling water level (Table 2). Its values were 1.17 cm, 15%, and 26% lower than that of MFI and WSI plots; however, the differences were not significant. The water level can generally be controlled as their values close to the setpoint, with RMSE below 1.6 cm. The biggest challenge in implementing the evapotranspirative irrigation was high rain intensity during one growing season. In hydrology, rainfall is always correlated to the water level as many models have been developed [33,34]; thus, rainfall becomes the most important factor in predicting water level under natural conditions.

**Table 2.** The performances of evapotranspirative irrigation in each regime.


Note: The presented data are the mean ± SD, where different letters in a row indicate a significant difference at α < 0.05 level.

Precipitation contributed to most of the water balance component by 79–88% of the inflow (Table 3). The largest contribution of precipitation was found in the WSI regime with less irrigation water. However, the rainfall affected more drainage or water loss. It was counted for 67–69% of outflow. For irrigation, CFI requires the most irrigation water to maintain flooded conditions in the field. The CFI regime required 27% and 49% more irrigation water than the MSI and WSI regimes. Flooded conditions in the CFI and MSI regimes also contributed to the higher value of actual evapotranspiration. The values were about 8% higher than that of the WSI regime. High actual evapotranspiration also correlated with higher crop coefficients in the CFI and MSI regimes. Several studies have shown similar results; flooding increases water used through actual evapotranspiration and consequently increases the crop coefficients [35–37].

**Table 3.** Water budget in each regime.


Note: The presented data are the mean ± SD, where a different letter in a row indicates a significant difference at α < 0.05 level.

#### *3.2. Effects of Irrigation Regimes on Crop and Water Productivities*

Plant height during one growing season in the three regimes is presented in Figure 8. At 10 DAT, the average plant height of the CFI, MFI, and WSI regimes was 26.5 cm, 23.4 cm, and 22.3 cm, respectively. The higher plant height of the CFI regime showed that standing water in the initial growth stage stimulated the crop to grow taller. At the beginning of the mid-season stage (64 DAT), there was a proportional and consistent increase in plant height of 98.8 cm under the CFI regime, while in the MFI and WSI regimes they were, consecutively, 93.4 cm and 92.2 cm. Finally, the highest average plant height at the end of the season was found in the CFI regime. It was 3.2% and 4.8% higher than those of the MFI and WSI regimes, respectively.

**Figure 8.** The Average plant height among the regimes.

Comparable results in the number of tillers were found among the regimes, particularly in the early growth stage. At 10 DAT, the regimes produced the same number of tillers (Figure 9). A significant increase in the number of tillers occurred from the vegetative growth stage (25–30 DAT). In this phase, the number of tillers was 11, 8, and 8 in the CFI, MFI, and WSI regimes, respectively. The tiller formation ended at 70 DAT in the generative state, in which the paddies focused on grain filling. An appealing occurrence happened at the end of the late-season stage, where the MFI regime produced more tillers than the two other regimes. The number of tillers in the MFI regime was 34. It was 3.8% and 10.8% greater than the CFI and WSI regimes, respectively. Thus, saturated soil conditions (water level at soil the soil surface) were more effective in tillers formation.

**Figure 9.** The average plant height among the regimes.

Based on statistical analysis, there was no significant difference in crop growth performance, including in plant height, number of tillers, number of panicles, biomass (straw) weight and grain yield (Table 4). Indeed, the CFI regime produced the highest plant height, which correlated to the heaviest straw weight. However, it was only about 5.6% higher than those of the others two regimes. Meanwhile, the MFI regime, although it produced lower plant heights than the CFI regime, it produced the greatest number of tillers and number of panicles. Its quantity was 5–10% higher than those of the CFI and WSI regimes, respectively. The exciting things occurred in the WSI regime that produced the highest grain yield. Although not significant, it was 6% and 7.5% higher than the CFI and MFI regimes, respectively. The increased grain yield seems to be due to the high grain density [38].

**Table 4.** Yield, water productivity, and water use efficiency among the regimes.


Note: The presented data are the mean ± SD, where the different letters in a row indicate a significant difference at α < 0.05 level.

However, the WSI regime produced the heaviest weeds biomass, reaching 3.7 tons/ha. Therefore, it was challenging to implement water-saving irrigation such as intermittent irrigation of the SRI method [39]. The WSI regime produced weed biomass more than three-times higher than the other regimes, and they were significantly different (Table 4 and Figure 10). Indeed, rice inundation was an alternative to prevent weed growth, especially in the vegetative phase [40]. However, as previously mentioned, it was wasteful in the water use since the paddies supplied more than they needed.

**Figure 10.** Weed collection after harvesting in each regime: (**a**) CFI; (**b**) MFI; (**c**) WSI.

The drier fields with the low water level caused the roots to grow more profound, as in the WSI regime (Figure 11). This situation is in line with the previous observations by Setiawan et al. [18] and Aziez et al. [41]. The water deficit conditions spur roots to grow vertically downwards in deeper soil layers to get water or nutrients. Deeper root formation may cause stronger paddy growth in the SRI with intermittent irrigation than in conventional farming with continuously flooded irrigation. Hence, SRI plant growth may be better than conventional systems with continuous waterlogging [42]. On the other hand, when the field is flooded, the roots grow sideways horizontally around the soil surface, as found in the CFI regime (Figure 11).

**Figure 11.** Root development of randomized hill of paddy in each regime: (**a**) CFI; (**b**) MFI; (**c**) WSI.

The minimum water irrigation in the WSI regime had implications in increasing water productivities, both in terms of total inflow (WPI+P) and actual evapotranspiration (WPETa). WPI+P of the WSI regime increased up to 14%; however, it was not significant because precipitation became dominant in water inflow. The same is true for water productivity

from the perspective of plant evapotranspiration. Although actual evapotranspiration was the lowest, the WPETa of WSI regime still increased up to 14.5% since the highest grain yield. Moreover, the WSI regime had the highest water use efficiency index due to the lowest irrigation. Its value index was 34% and 52% higher than those of the MFI and CFI regimes. This lead showed that maintaining water level at the soil surface at the beginning of plant growth is one alternative to raise water-use efficiency. This result is similar to that from an alternate wet and dry irrigation (AWD) experiment conducted previously to improve water use efficiency [43].

#### **4. Discussion**

Along with the effect of climate change, water resource availability changes and tends to decrease, particularly in runoff and water levels due to changes in the hydrological cycle [44]. Climate change is commonly characterized by increasing temperatures, rainfall patterns and the frequent occurrence of extreme weather [45]. The concept of evapotranspirative irrigation is an effort to find an adaptive strategy to climate change and easier application in the fields. The performance showed it was satisfactory with fairly small RMSE values (Table 2). However, the system inaccuracies were raised when there was a heavy rainfall event (Figures 5–7), and therefore precipitation became a constraining factor affecting the performance. The precipitation was also found as the main factor that reduced accuracy in water level control application in Indonesia [46,47].

Although we utilized advanced technology such as sensors, actuators and microcontrollers, inaccuracies were found during rainfall whenever the drainage system was not controlled properly. Sirait et al. [46] developed a solar power pipe irrigation automation system to control water levels. The performance of the system was very satisfactory from the beginning and early late-season; however, error increased, as raising the gap between setpoint and observed water level in the late season since the heavy rain event. An identical situation was found by Nurfaijah et al. [47]. They developed an on-off water level control system by utilizing an Arduino microcontroller for three irrigation regimes. There was an increase in error during the precipitation. Therefore, it is highly recommended to control the drainage rate for areas with high rainfall, such as by utilizing a subsurface drainage system [48]. The subsurface drainage technology was able to increase water-use efficiency up to 20% while maintaining the yield [49].

Among the regimes, the evapotranspirative irrigation was suitable with the WSI regime in producing more rice. The key to increasing grain yield was seemingly attributed to the lower water level below the soil surface after 20 DAT. The field was on aerobic conditions that allowed more oxygen availability in the soil [50]. In addition, in the initial stage, the field was wet. Thus, the WSI regime was similar to the moderate wetting and drying regime (MWD) [51] or alternate wetting and drying irrigation (AWDI) [52]. The regime was effective in water use and able to increase the yield [53]. The key in increasing grain yield is increasing oxidation activity in roots, raising the photosynthetic rate in leaves and increasing enzyme activities in the converting process of sucrose to starch in rice grains [51]. Moreover, the system allows the roots to grow larger. It will transport cytokinins through the xylem to the leaves in maintaining the photosynthesis process [54]. The longer root, as presented in Figure 11, seemingly shows more activities inside under the WSI regime. More biomass and grain were also developed when more oxygen absorption occurred by root activities, particularly in the reproductive stage [55].

However, the aerobic condition also has potential yield reduction when the lower water level is not well controlled, causing the extreme driest of the soil. Setiawan et al. [18] reported that the yield could be maintained at a water level of 3.2–7 cm below the soil surface. However, if the water level is deeper than those intervals of water levels, it can significantly reduce the yield due to stress on the crop. Based on a field experiment by Zhang et al. [51], the yield reduction of 32% occurred when the soil was extremely dry. The reason is due to abiotic factors such as increased soil pH, ammonia toxicity and nutritional deficiencies in aerobic conditions [56]. In addition, aerobic conditions also stimulate significant weed growth, as shown in Figure 10. More weed production can potentially reduce the yield, thus integrated weed management became important to deal with this obstacle [57]. In Indonesia, weed growth can be suppressed using an active herbicide containing 10% ethyl pyrazosulfuron applied after tillage and a mechanical power weeder [58].

Under the absence of inundation, such as in the WSI regime, the rate of evapotranspiration is low [25]; consequently, the total actual evapotranspiration was lowest compared to the two other regimes (Table 3). Then, the average actual coefficient of this regime was also lowest. The finding was supported by Linquist et al. [37]. They performed a 3-year field experiment and found that continuously flooded irrigation resulted in higher actual evapotranspiration and crop coefficient than that drier, and vice versa. Kadiyala et al. [59] recorded a 19% lower crop coefficient in aerobic conditions. Commonly, the lower the actual evapotranspiration, the lower the yield [18,60], according to the basic equation reported by van Lier et al. [61]. However, several experiments showed different results [10,62–64]. It seems there is an inconsistent correlation between crop coefficient and grain yield. According to Zhang et al. [65], the relationship between evapotranspiration and yield can be represented by a parabolic trend. By means, the higher evapotranspiration may lead to higher yield within a particular range. Then, after the parabolic peak point, the opposite trend is found. It is important to optimize the irrigation regime to find the peak point of actual evapotranspiration and yield so that water can be efficiently used.

The WSI regime improved water productivities and a significant water-use efficiency index. The similar result was also found by Choudhury et al. [66], that SRI improved water productivity both in evapotranspiration and total water supplied perspectives. In addition, the SRI saves 18–21% of water input [67]; thus, the regime is suitable for the areas with limited water resources such as upland and in combination with SRI cultivation [68]. The strategies to improve those two parameters are by reducing percolation and evaporation. Percolation can be reduced by minimizing the inundation water level (or at least at saturated level) and increasing the duration of unsaturated conditions at 80–90% field capacity water content [69]. In other words, the water level should be kept between 0 and 5 cm on the surface [70]. Under this setup, plant growth was not significantly impaired (Figure 8), and it is an effective strategy to reduce evaporation from the soil surface [71].

#### **5. Conclusions**

A model of evapotranspirative irrigation has reasonable prospects of application because of its simplicity and easiness of use. Good performance of the system was achieved as indicated by low RMSE in all irrigation regimes during one rice planting season. The performance can be improved by developing a better drainage system. According to the experiment under the developed technology, the water-saving irrigation (WSI) regime of the system of rice intensification (SRI) was most efficient in water use. It was able to increase water productivity by up to 14.5% without reducing the yield. In addition, it has the highest water-use efficiency index, which is 34% and 52% higher than the moderate flooded irrigation (MFI) and continuous flooded irrigation (CFI) regimes. In the near future, the system should be implemented at the field levels under various climate conditions.

**Author Contributions:** Conceptualization C.A. and B.I.S.; methodology C.A. and B.I.S.; data collection C.A., S.K.S. and M.T.; writing—original draft preparation, C.A.; writing—review and editing, C.A., W.B.S., M.T. and M.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Ministry of Research and Technology/National Research and Innovation Agency by the project title "Developing Artificial Intelligence Based Smart Evaporative Irrigation for Environmental Friendly Precision Farming" according to the contract number 2090/IT3.L1/PN/2021 and 1/E1/KP.PTNBH/2021.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available upon request from the corresponding author. The data are not publicly available due to shared ownership between all parties that contributed to the research.

**Acknowledgments:** We thank the Ministry of Research and Technology/National Research and Innovation Agency for their funding support to this study. Also, we thank 3 anonymous reviewers for their constructive comments and suggestions.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**

