Document Type : Original Research Paper
1 Ph.D. School of Engineering, College of Engineering, Computing and Cybernetics, Australian National University, Canberra, ACT Australia and Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC Australia
2 Ph.D., Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC Australia.
3 Assistant Professor, Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC Australia.
4 Professor, Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC Australia
Introduction: Irrigation water is an expensive and limited resource. Previous studies show that irrigation scheduling can boost efficiency by 20-60%, while improving the water productivity by at least 10%. A key aspect of irrigation scheduling is accurate estimation of crop water use and soil water status, which often require modelling with good information on soil, crop, climate and field management. However, this input information is often highly uncertain. Our study aims to obtain a comprehensive understanding of uncertainties in irrigation scheduling that arise from individual model inputs, from which identifying the key contributor of uncertainty. Our study aims to understand the uncertainty in model-based irrigation scheduling and the key model inputs that contribute to this uncertainty. To achieve this, we first performed a comprehensive literature review to identify the key sources and the expected ranges of uncertainty in individual model inputs. Secondly, a global sensitivity analysis was conducted to quantify the influence of each model input on the total uncertainty of the modelled irrigation scheduling decision, across 14 climatically different locations in Australia.
Materials and Methods: To achieve this, we used a global sensitivity analysis to assess the relative importance of the uncertainty in each model input to the total uncertainty in output. This analysis focused on the modelled irrigation scheduling (summarized with irrigation amount per day during an irrigation cycle) with a single-bucket soil water balance model following the Food and Agriculture Organisation (FAO). The key input variables required by the model include weather data, crop parameters (i.e., crop coefficient and root depth), soil parameters (plant available water capacity) and management factors (depletion factor).
Results: To define the uncertainty in each model input, we first performed a comprehensive literature review to summarize the key sources of uncertainty in estimating each of these model inputs, and the expected range of uncertainty in the data of each input. Based on these uncertainty ranges, we ran the global sensitivity analysis with the soil water scheduling model. In this analysis, a large number of random samples were drawn for each input variable within its expected range of uncertainty, to produce ensemble simulations of soil water status and thus irrigation scheduling decisions. The total uncertainty in these scheduling decisions were then analysed with respect to that of each input variable, to establish the relative importance of the uncertainty in individual input variables. The sensitivity analysis was performed at 14 climatically different locations in main irrigation districts across Australia to provide a comprehensive understanding of sensitivity.
Conclusions: Our results highlight the crop coefficient as the most important contributor to the total uncertainty in irrigation scheduling simulation, across different climate zones in Australia. The uncertainty in crop coefficient can be potentially reduced by better representation of its spatial and temporal variation, as well as considering alternative approaches such as remote sensing estimates. Our findings are useful to inform the future direction of research to improve irrigation scheduling in Australia. Further, our modelling approach is transferable to other irrigation regions to better understand the uncertainties associated with irrigation scheduling and the key data sources that lead to these uncertainties.