We investigated seasonal water quality changes on different land uses in a small agricultural watershed dominated with rice and tea crops. Observed NO3–N in tea plantations and tea drainage channels was sufficiently high enough to cause downstream water pollution throughout the year. On contrary, significant NO3–N reductions were observed in abandoned and active paddy fields, and were highest in active paddy fields during rice growing seasons. Accordingly, a question is posed whether deliberately diverting NO3–N contaminated drainage water from tea plantations to paddy fields would be effective a strategy for reducing downstream water pollution. NO3–N ratios reduced in paddy fields are however neither distinct nor known in advance, and this uncertainty could consequently lead to further environmental risks like NO3–N leaching and production of greenhouse N2O gas. We assume that NO3–N reduction rates of paddy fields and drainage canals are constrained in an ellipsoidal set of uncertainty. A portfolio optimisation problem is formulated to determine optimal rates of diversion from tea drainage channel into paddy fields. The problem is numerically solved in robust optimisation approach by maximising minimum NO3–N reduced under uncertainty. An application demonstrated that robust approach allocates different optimal rates of diversion to hedge the risk of possible malfunctioning of NO3–N reduction processes. Overall worst-case NO3–N reduction is highest in rice growing season, due to increased drainage water allocations to active paddy fields. The robust optimisation approach has potential to support decision-making processes for reducing NO3–N pollution and water demand pressure on water resources, and cost of rice production thereof.