Modelling the response of paddy water balance on groundwater level fluctuations in Central Punjab

Long-term abstraction of groundwater resources, in particular for irrigation, has resulted in severe decline and deterioration of water resources in many arid and semi-arid regions of the world (Konikow & Kendy 2005; Rodell et al. 2009; Scanlon et al. 2012; Feng et al. 2013; Yang et al. 2013; Famiglietti 2014; Chinnasamy & Ganapathy 2018). Rice (Oryza sativa L.), an important cereal and staple food for more than half of the global population, accounts for 40% of all global irrigation, and 17% of global groundwater depletion. The global average water footprint per kilogram of rice is 2,500 litres (Bouman 2009), with India at 2,800 litres and the State of Punjab has the highest footprint of 4,500 litres. Here, rice covers nearly 75% of the total cropped area and requires nearly 800–2,000 mm of irrigation water depending upon climate and soil type. The rainfall and surface water resources are not sufficient to meet the total irrigation needs. The percentage of irrigated area from canals and tube wells in the state is 28 and 72%, respectively. This has led to a ‘silent revolution of intensive groundwater use’ (Lamas & Martinez-Santos 2005) with the number of energized tube wells increasing from 0.19 in 1970 to 1.475 million in 2019 (Anon 2019). Consequently, there is huge pressure on the groundwater aquifers, more so in the central districts of the state where the water table has registered a steep decline (>1 m/year). Many researchers in the past have emphasized water saving in rice through adoption of improved technologies, such as laser land levelling, alternate wetting and drying (AWD), use of tensiometers for irrigation scheduling, delayed transplanting, shorter duration rice varieties and even drip irrigation systems (Singh et al. 2001a, 2001b; Kukal et al. 2005; Aggarwal et al. 2009; Minhas et al. 2010; Eberbach et al. 2019). It is often likely that these practices may reduce deep drainage, with very little effect on evapotranspiration (ET). Reducing deep drainage may not ‘save water’ or reduce the rate of decline of the water table (Humphreys et al. 2010). Researchers have also argued that the major loss, i.e., ET in rice, ranges about 30–40% (Jalota et al. 2018) and the drainage from paddy fields may not be considered as a loss because the water can be captured and reused downstream (Bouman et al. 2007). In a study elsewhere, in Iran, it was concluded that water saving irrigation systems (e.g,. drip and sprinkler) conserve water but they can reduce potential groundwater recharge (PGR) compared with traditional systems (e.g., furrow) and may impair sustainability, particularly in arid and semi-arid regions (Porhemmat et al. 2018). Further, rice is grown in kharif season and the region receives nearly 80% of the annual rains during this period. As rice is a tolerant crop, its yield is not adversely affected and the standing rainwater in the fields slowly joins the water table. Seasonal trends of the region also showed that the water table rose after kharif season (mid-June to mid-October) and fell during the rabi season (mid-October to mid-June) (Jalota et al. 2018). It becomes imperative to establish a relationship between water balance dynamics and groundwater changes (Yang et al. 2006), but systematic studies on this aspect are seldom seen for the region. This requires complex information about the soil vegetation system, which is usually lacking, and there are financial and technical limitations involved in gathering a large amount of data. In most of the earlier studies, groundwater use has been related to the number of tubewells and unit draft (Aggarwal et al. 2005; Miglani et al. 2015), and these assessments do not indicate the desired saving in water in order to decrease or stop groundwater decline which require continuous measurements of long-term agricultural water use. Alternatively, process-based crop models such as CERES (Crop Environment Resource Synthesis), DSSAT (Decision Support System for Agrotechnology Transfer), EPIC (Environmental Policy Integrated Climate), etc., can provide insights into crop water use and drainage phenomena from a simplified input of climate, soil and water management. For instance, Jalota et al. (2014) used CropSyst to simulate the effect of climate change on the global irrigation water use under designed climate change scenarios. Dar (2016) examined the applicability of the DSSAT model in analysing the impact of water-related options on rice yield and water use for enhancing water productivity in irrigated environments of Punjab and found the model performed reasonably well. With the above background, the present research was planned to evaluate the DSSAT model for quantifying the impact of water balance components on groundwater levels in rice crop.

The field experiment was conducted at the Research Farm of the Department of Soil Science, Punjab Agricultural University, Ludhiana (30°56′ N, 75°52′ E and 247 m above mean sea-level) during the rainy (kharif) season of 2014. The soil of the experimental field was deep alluvial loamy sand Typic Ustipsamment developed under hyperthermic regime (USDA classification). At the start of the experiment, depthwise soil physical and chemical properties of the field were determined following the standard procedures (Table 1). The meteorological data (daily maximum and minimum temperatures, rainfall, pan evaporation, sunshine hours, morning and evening relative humidity) were collected from the meteorological observatory of the Punjab Agricultural University, Ludhiana located at a distance of 100 m from the experimental field.

Rice was grown in sandy loam soil under different irrigation conditions, since the genetic parameters for cultivar ‘PR111’ of rice have been calibrated previously (Vashisht et al. 2015). In the experiment, the effects of three sowing dates (June 5, D1, June 20, D2 and July 5, D3) and two irrigation regimes (2-day drainage (I₁) and tensiometer-based, (I₂) level were evaluated on growth and yield of rice.

Each treatment was replicated three times with a plot size 6 × 4 m² in a split-plot design. Irrigation treatments were started following continuous flooding for 15 days after transplanting. The amount of irrigation each time was 75 mm. For tensiometer-based irrigation, the soil water suction was measured with tensiometers installed at 20 cm soil depth. In 2-day drainage, irrigation was applied 2 days after the ponded water had infiltrated into the soil. Fertilizer dose of N, P, K was applied following the procedure given in a package of practices for crops of Punjab (www.pau.edu). Plant parameters (plant height, number of tillers m⁻² and dry matter accumulation) and yield attributes (grain loop length, per cent filled grains, thousand grain weights and grain and straw yield) were recorded.

To get the above constants of A and B, a bunded and fully saturated plot on the bare soil was covered with a polyethylene sheet in order to stop the evaporation completely. Profile soil moisture storage was estimated from the plot at the depths of 15, 30, 45, 60, 90, 120, 150 and 180 cm at days 1, 3, 5 and 7 up to 39 days. From the relation between W and T, the values of A and B were 38.13 and 0.09 (Figure 1).

Simulations for yield and water balance were run using the DSSAT model (Jones et al. 2003). In the model, certain crop parameters were modified slightly (within the given range) to match the periodic biomass and yield and the rest of the parameters were equivalent to those previously reported (Vashisht et al. 2015). The location parameters included longitude, latitude, daily weather data files and ET models. The methods selected in the simulation were Priestley–Taylor – evapotranspiration, Soil Conservation Service – infiltration, canopy curve (daily) photosynthesis, Ritchie water balance – hydrology, CERES–Godwin – organic matter and Ritchie-CERES – soil evaporation.

Treatment-specific irrigation management operations, performed on different dates in the experiments, were entered in the crop management file. The model performance was evaluated for yield and water balance component in rice transplanted during 2014 using the Nash–Sutcliffe modelling efficiency (ME) (Nash 1970) and the root mean square error (RMSE), respectively.

In the present study, the difference in subsurface inflow and outflow is taken as nil (Kaur et al. 2015) and irrigation from groundwater was not considered as a fraction of the total withdrawal contribution to water table decline which is used as ET. The specific yield was assumed to be 0.20 (Kaur et al. 2014).

Site-specific groundwater level data (pre- and post-monsoon) from 1998 to 2014 was obtained by employing krigging interpolation technique (Dhillon et al. 2018) on the nearby observation wells.

Rice yield was significantly influenced by the date of transplanting. The maximum yield of rice was observed on June 20 (D2), which was higher than that in June 5 (D1) and July 5 (D3), i.e., 24 and 31% (Table 2). These observations confirm the field and simulated results reported by other researchers (Singh et al. 2001a, 2001b; Chahal et al. 2007; Safdar et al. 2008; Jalota et al. 2009). Late planting (July) showed a decline in grain yield due to delayed panicle formation and grain filling in the season where temperature and solar radiation are less (IRRI 1993). With respect to irrigation, I₂ recorded 11% more yield than I_1, although irrigation was 21% higher in 2-day drainage-based treatment. Kukal et al. (2005) also reported that irrigation at 160 ± 20 cm soil metric suction helped save 30–35% irrigation water as compared to 2-day drainage-based irrigation with no yield penalty. ET observed was higher in D₁ (550 mm) than D₂ (514 mm) and D₃ (546 mm).

The water balance components and WUE changed with treatments (Table 2). Precipitation was nearly the same for all the dates of transplanting_. D₂ received 16 and 4% more irrigation than that of D₃ and D₁, respectively. Measured ET by water balance equation was higher in D₁, i.e., 550.5 mm than D₂ and D₃. The highest water use efficiency in D₂ is ascribed to relatively greater decrease in ET (denominator) compared to the yield (numerator). Averaged across irrigation treatments, WUE in I₂ was 13% more than I₁.

Simulated grain yield and ET had trends similar to measured responses, although there were differences in their magnitude (Figure 2 and Table 3). Statistical parameters such as RMSE and Nash–Sutcliffe ME during validation were 0.09 and 0.99 for grain yield. The coefficient of determination (R²) between simulated and observed yield was 0.96 and was 0.86 in the case of ET.

The model simulated crop yield, irrigation, daily ET and drainage from 1998 to 2014 for the site (Table 4), using the historic climate data. The highest yield of paddy was obtained in 2014, i.e., 6.6 t/ha, the lowest yield in 2010, i.e., 5.5 t/ha and the average yield of 17 years is 6.0 t/ha which closely agrees with the data provided by government agencies (Statistical Abstracts Punjab, Various issues).

The average kharif season's rainfall computed from the daily records (for the years 1989–2014) was 594 mm. The irrigation requirements of rice ranged from 1,020 to 1,994 mm with an average of 1,443 mm. The ET, which is one of the major losses, from the rice field varied from 387 to 597 mm during different years with an average loss as 438 mm. The drainage component which is assumed equal to potential recharge ranged between 61 and 70% of total water applied (precipitation + irrigation) during different years, with an average drainage of 66%, i.e., 1,342 mm (Table 4), of which, precipitation contributes 19.2%. However, a significant variation in contribution to potential recharge individually from precipitation and irrigation is witnessed for different years (Figure 3). Increased precipitation bears a linear positive relationship for contribution to potential recharge and vice versa relationship for irrigation.

An average post-monsoon rise of 0.74 m in groundwater levels at the site was observed with exceptions in the years 2000, 2002 and 2012, when the water table dropped post-monsoon. These exceptional years were due to deficit rainfall coupled with high irrigation water use. Although a graph between groundwater fluctuations in response to rainfall and irrigation water use for the experimental site indicates that June to October is the period when groundwater level responds relative poorly to irrigation water requirement, since soil water from the rainy season is also absorbed, yet, there was general agreement between potential recharge and seasonal groundwater level changes (Figure 4).

The simulated change in storage (Equation (5)) was able to fairly estimate the groundwater water fluctuations rather than potential recharge (drainage − irrigation) alone (Figure 5), more so when years with rainfall greater than 800 mm were excluded. In a recent study by Jalota et al. (2018), the annual water table fluctuations were approximated by computing the water deficit [evapotranspiration − (rainfall + surface water)], which can approximate the water table decline/rise rather than groundwater withdrawal alone.

In view of shrinking water resources and the anticipated effects of climate change, there is a need to have a better understanding of crop water balance relationship with groundwater dynamics. The outcome of the study indicates that delaying the transplanting date to June 20 and tensiometer-based irrigation can result in higher yield (34%) and water saving (10%) in comparison to June 5 and 2-day drainage-based irrigation. The findings are in tune with the government recommendations of delaying the transplanting date of paddy to mid-June. Simulated CERES-Rice model was then applied to quantify change in storage while analysing groundwater fluctuations. The estimated change in storage was able to explain groundwater level fluctuations in terms of trends/magnitude in most of the years barring the high rainfall as there are still limitations when we directly try to link groundwater level changes and crop water use. First, the simulation results might be more accurate for recent years than for earlier years. Second, the effect of different crop cultivars was not used in historical simulations and the genetic parameters of PR 121 were used for the whole simulation period. Third, if actual groundwater level observations were available we could have computed more realistic regression. Aside from the application of the agricultural model, a more accurate estimation of groundwater use to reflect the change of groundwater level in multiple aquifers through the application of a groundwater model might be necessary. However, with all these limitations, such type of study still provides information on crop behaviour under different conditions, and thus can be helpful for formulating management intervention for increasing productivity and water use efficiency and also give insights on agricultural water use and groundwater fluctuations.

The authors are very grateful to the Indian Institute of Water Management, Bhubneshwar – ICAR for funding the research under All India Coordinated Research Project of Irrigation Water Management.

All relevant data are included in the paper or its Supplementary Information.