Decision support system for real-time irrigation scheduling in rice using weather forecast data

Introduction

On one hand, rice fields in southern Spain consume roughly 12,000-15,000 m³ of water per hectare each season; on the other hand, the Guadalquivir river basin faces increasingly frequent water allocation cuts that threaten the crop's viability [1,2] . This contrast between high crop demand and shrinking supply makes irrigation efficiency not just an agronomic goal but a matter of survival for Mediterranean rice growers. Decision support systems for irrigation have existed since the 1980s, but early versions relied on historical averages and static crop coefficients [3] . High-resolution NWP models now issue reliable 72-hour forecasts at grid spacings under 5 km, making it possible to anticipate rainfall and evaporative demand before they occur [4,5] . Several prototypes have been tested for wheat, maize, and orchards, yet applications in flooded rice systems pose unique challenges around paddy water depth management and stage-specific sensitivity [6,7] . This research aimed to: 1) Design a DSS combining 72-hour forecasts, soil moisture data, and growth-stage logic for rice irrigation; 2) Evaluate water savings and yield impact over a full season; and 3) Assess recommendation accuracy against a post-season optimised schedule [9,10] .

Theoretical background

The DSS operates on a daily water balance: Paddy Depth (t+1) = Depth (t) + Irrigation + Predicted Rainfall sET₀ × Kᶜ -Seepage -Percolation [11] . ET₀ is forecast using Penman-Monteith equations applied to NWP output from aemet's harmonie-arome model at 2.5 km resolution [4] . Crop coefficients for five phenological windows ranged from 0.95 (milk-maturity) to 1.35 (panicle initiation-heading) following adjusted FAO-56 guidelines [12] . The system triggers irrigation when predicted paddy depth falls below the stage-specific minimum within the 72hour window, subject to a soil moisture sensor cross-check at 15 cm depth [13] .

Material and Methods

Material

The DSS was deployed at the University of Córdoba rice station along the Guadalquivir floodplain (37°51′N, 4°48′W, 90 m) during May-October 2022. Eight paddies (14.6 ha) were split into four DSS-managed and four farmer-managed plots. Rice variety Marisma (medium-grain Japonica, 140-day cycle) was transplanted on 12 May at 20 × 20 cm. Soil was heavy clay (pH 7.4, OC 1.2%). EC-5 sensors at 15 cm depth logged moisture every 30 minutes; forecast data were pulled from AEMET's open API daily [14] .

Methods

The DSS ran on a Python backend with recommendations delivered by SMS each morning. Irrigation was recorded by flow meters on each inlet. Farmer-managed plots followed conventional calendar scheduling with continuous flooding at 5-10 cm. Grain yield was measured from 10 m² per plot at 14% moisture. Total water applied, water productivity, and recommendation accuracy (compared with post-season optimum) were calculated. Differences tested by paired t-test at p≤ 0.05 in R (v4.3.1) [15] . The DSS reduced total irrigation by 23.7% and the number of irrigation events by 31.5% without significant yield loss (Table 1). Water productivity rose from 0.62 to 0.80 kg m⁻³, a 29.0% improvement. Grain yield under DSS management (7.83 t ha⁻¹) was statistically equivalent to the farmer-managed yield (7.96 t ha⁻¹, p = 0.41). Overall recommendation accuracy was 88.1% across 143 daily decisions (Table 2). Accuracy was lowest during panicle initiation to heading (82.1%), when unpredicted convective storms led to unnecessary irrigation recommendations. The highest accuracy (91.2%) was recorded during the transplantingtillering window, when weather patterns were more stable. Forecast-based scheduling was the largest single contributor to water savings (31.4%), followed by soil moisture feedback (23.8%) (Figure 1). Figure 2 shows that DSS-recommended and actual irrigation volumes were closely aligned (R² = 0.87), with most points falling near the 1:1 line, confirming that the recommendations were practically implementable.

Results

Performance evaluation

Across 143 daily decision cycles, recommendations were delivered within 6 minutes of the 06:00 UTC trigger. System uptime was 98.6% over the 152-day season, with three outages totalling 2.1 days caused by API timeouts. During outages, fall back rules based on the previous day's recommendation were automatically applied.

Discussion

A 23.7% reduction in seasonal irrigation without yield penalty falls within the 15-30% savings reported for DSS deployments in Italian and Australian rice [6,16] . The slightly higher savings here may reflect the baseline inefficiency of calendar-based scheduling common in Córdoba [2] . The 82.1% accuracy during panicle initiation to heading highlights NWP weakness in forecasting convective storms in Mediterranean summer; incorporating radar now casting could improve this [17,18] . Water productivity rose by 29%, translating to about 122 m³ ha⁻¹ in saved water charges per season. Scaled across the 36,000 ha of rice in the lower Guadalquivir, this could free roughly 4.4 million m³ annually for competing demands [19,20] .

Conclusion

This research demonstrated that a weather-forecast-driven DSS can reduce rice irrigation by 23.7% while maintaining yield at 7.83 t ha⁻¹. The system achieved 88.1% recommendation accuracy across a full season, with forecast-based scheduling contributing the largest share of water savings. Water productivity improved by 29%, offering both economic and environmental benefits for water-scarce Mediterranean basins. Future development should incorporate radar now casting for convective periods, extend testing across multiple seasons, and add economic optimisation for different water pricing scenarios.

References

1. Aguilar M, Borjas F, Expósito M. Rice irrigation management in the Guadalquivir marshlands. Agricultural Water Management. 2017;191:47-56.
2. Confederación Hidrográfica del Guadalquivir. Plan Hidrológico 2022-2027. Seville: CHG; 2022.
3. Smith M, Kivumbi D, Heng LK. Use of the FAO CROPWAT model in deficit irrigation studies. FAO Water Reports. 2002;22:17-27.
4. Bengtsson L, Andrae U, Aspelien T. The HARMONIEAROME model configuration in the ALADIN-HIRLAM NWP system. Monthly Weather Review. 2017;145(5):19191935.
5. Bauer P, Thorpe A, Brunet G. The quiet revolution of numerical weather prediction. Nature. 2015;525(7567):4755.
6. Masseroni D, Facchi A, Gandolfi C. Advances in irrigation scheduling for Italian rice: a review. Paddy and Water Environment. 2019;17(3):319-333.
7. Bouman BAM, Tuong TP. Field water management to save water and increase its productivity in irrigated lowland rice. Agricultural Water Management. 2001;49(1):11-30.
8. Navarro-Hellín H, Martínez-del-Rincon J, Domingo-Miguel R. A decision support system for managing deficit irrigation in almonds. Information Processing in Agriculture. 2016;3(4):241-253.
9. Perea RG, Poyato EC, Montesinos P. Irrigation demand forecasting using artificial neuro-fuzzy inference. Agricultural Water Management. 2019;226:105797.
10. Delgoda D, Saleem SK, Halgamuge MN. Irrigation control based on model predictive control: formulation of theory and validation using weather forecast data. Journal of Irrigation and Drainage Engineering. 2016;142(1):04015037.
11. Allen RG, Pereira LS, Raes D. Crop evapotranspiration: guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56. Rome: FAO; 1998.
12. Pereira LS, Paredes P, Rodrigues GC. Crop coefficients for improved water management: past, present and future. Water Resources Management. 2020;34(4):1079-1098.
13. Adeyemi O, Grove I, Peets S. Advanced monitoring and management systems for improving sustainability in precision irrigation. Sustainability. 2017;9(3):353.
14. AEMET. AEMET OpenData: public access to meteorological and climatological data. Madrid: Spanish Meteorological Agency; 2022.
15. R Core Team. R: A Language and Environment for Statistical Computing. Vienna: R Foundation; 2023.
16. Dunn BW, Gaydon DS. Rice growth, yield and water productivity responses to irrigation scheduling prior to the delayed application of continuous flooding in south-east Australia. Agricultural Water Management. 2011;98(12):1799-1807.
17. Llasat MC, Marcos R, Turco M. Trends in flash flood events versus convective precipitation in the Mediterranean region. Journal of Hydrology. 2016;541:24-37.
18. Sokol Z, Zacharov P. Nowcasting of precipitation by an NWP model using assimilation of extrapolated radar reflectivity. Quarterly Journal of the Royal Meteorological Society. 2012;138(663):1072-1082.
19. Berbel J, Gutierrez-Martin C, Rodriguez-Diaz JA. An analysis of the Spanish water pricing policy. International Journal of Water Resources Development. 2011;27(3):511531.
20. Intrigliolo DS, Castel JR. Performance of various water stress indicators for prediction of fruit size response to deficit irrigation in plum. Agricultural Water Management. 2006;83(1):173-180.