Performing Department
irrigation for the future, inc
Non Technical Summary
This project involves sophisticated decision support software for optimum irrigation management in a future of accelerating competition for water, increasing energy costs and concerns about sustainability of irrigation water use.The decision support system analyzes how applied irrigation water translates into crop water availability and how crop water availability relates to crop yields. With these analytical capabilities farm managers can plan irrigation schedules to maximize water use efficiency, minimize production costs and capitalize on utility incentives for energy load shifting. And the farm manager will also be better equipped to deal with ground water resource management and water quality protection.The analytical models at the heart of the system depend upon a number of input parameters, such as drip line emitter rates, soil water holding capacities, crop root depths, canopy cover, etc. The accuracy of the models largely depend on the accuracy of the parameter values. But, while parameter values can be estimated reasonably well based on prior knowledge of specific soil types, crop characteristics and irrigation system configurations, the precise values for any given field will always be somewhat different from the prior estimates, and the collective effect of even small differences in parameter values can compromise model accuracy.Maximizing the effectiveness of thedecision support software therefore requires locally specific calibration ofkey parameters. Experience has shown that critical parameter values can be refined using field data in a feedback loop to calibrate the model. However, the calibration process requires more technical expertise than most irrigators have, and it can be a computationally intensive process. While large, well capitalized farms often have adequate resources to deal with calibration, smaller farms generally do not. This project will deploy an automated calibration system utilizing an optimizing search tool to minimize the technical burden and reduce the overall effortin order to make the decision support tool readily accessible and practical for use by smaller farms.
Animal Health Component
75%
Research Effort Categories
Basic
(N/A)
Applied
75%
Developmental
25%
Goals / Objectives
The fundamental goal of this project is to facilitate use of advanced irrigation management software by small and medium sized farms. The project is specifically focused on a decision support system known as Irrigation Management Online (IMO) that supports three fundamental goals of the SBIR program; (i) improving the efficiency of water and energy use; (ii) increasing the profitability of small irrigated farms through better planning and risk management; and (iii) utilizing renewable energy more effectively.IMO relies on sophisticated modeling of the disposition and fate of applied water to characterize crop water availability in spatially variable fields. The modeling accounts for crop characteristics, irrigation system performance, soil water dynamics and operational constraints of individual farms, and provides farm managers with detailed, optimized schedules for water and energy use weeks or months into the future. These analytical capabilities necessarily involve substantial modeling complexity, and the accuracy of the system depends to some extent on local, farm-specific calibration.Once calibrated, IMO enables individual irrigators to anticipate how different levels of applied water and different irrigation timing strategies will translate into seasonal patterns of crop water availability for the unique circumstances of their specific fields. With those capabilities irrigators can increase net economic returns to water and energy in a variety of ways, such as: (i) planning irrigation schedules to maximize water use efficiency; (ii) recognizing and capitalizing on the opportunity costs of water; (iii) participating in incentive programs for energy load shifting to minimize electrical grid congestion; and (iv) aligning irrigation schedules with the natural timing of on-farm renewable energy generation. The analytical tools of IMO can also be used for sustainable ground water management or to minimize nutrient leaching.The user interface is being simplified to the point where - once set up and calibrated - a farm manager will need no special technical training to use it. However, the calibration process requires a relatively high level of technical training and experience, and that presents an obstacle to widespread use of the system, particularly for smaller farms. The central purpose of this project is therefore to simplify the calibration processfor irrigators with diverse irrigation experience and educational backgrounds.The first phase of the project will be to develop an intelligent, automatic calibration procedure using simulated annealing augmented with heuristic spread sheets through which to input locally specific knowledge and experience. The second phase of the project will involve working closely withirrigators on small and mid-sized farms, and with irrigation scheduling companies and crop consultants who serve those farms, to continue refining the software to meet their operational conditions without adding undue burdens of higher management costs or added labor.The specific goals are:develop a simulated annealing algorithm for calibration of the decision support system;characterize the algorithm performance in terms of processing speed and user demand scenarios;develop an excel spreadsheet interface to simplify transfer of farm parameters and field data into the decision support system;prototype heuristic tools for introducing local knowledge and experience into the analysis to accelerate the search procedure;test the efficacy of the search tools and heuristics with currently engaged cooperators on small and medium farms where two or three years of high quality data are available.
Project Methods
The IMO system is unique among commercial irrigation management tools in that it integrates estimates of soil moisture derived from two different sources: (i) estimates based on modeling of evapotranspiration and soil water dynamics, and (ii) in situ measurements of soil moisture. In theory, both of these sources should produce the same estimate. But each of these sources has some associated error, so, in practice, the estimates are never the same. But the nature of the error sources is not the same either, and because the errors are from different causes the two estimates can be combined to produce a third estimate with a lower standard error than either of the original two. The magnitude of the difference between the two original estimates also gives us some indication of potential errors in the estimates.In operational use of IMO the model estimate is 'corrected' by revising the estimate to a weighted average ofmeasured and modeled values of soil moisture. The improved estimates of soil moisture and soil moisture measurements are displayed togetherto make the user aware of the differences between them. The resulting seasonal record of daily crop water availability enables the user to improve crop water management, relate other commonly used plant-based measurements (e.g. stem water potential) to field-wide crop water availability, and, in general, improve the performance of the overall system. The calibration procedure will use simulated annealing to minimze thedifferences between the water balance-based estimate and the in situ measurementby modifying the values of relevant parameters. An Excel based interface will establish the appropriate ranges of parameter values and provide heuristic guidelines on selection of parameters to be adjusted according to the nature and scale of differences between model estimates and field measurements.Themethodologiesspecific to each ofthe five elements listed earlier are discussed below:1. Development of optimization codeThis part includes the development of an application of the Simulated Annealing[1] algorithm, software infrastructure to support execution of the method, and database extensions to store results and integrate them into the user's datasets.2. Microsoft Excel-based interfaceThis interface will serve two purposes. First, it will facilitate implementation of the Simulated Annealing algorithm. There is some uncertainty about how to identify which parameters will significantly impact calibration. It is possible to use sensitivity analysis to identify which parameters will have the biggest impact on calibration. However, we have found through testing and manual calibration that the most impactful parameters also depend on the farm's configuration and especially on differences between what the operators believe to be the system configuration and the actual conditions on the ground. For example, a common scenario in our field trials is where the farmer tells us the flow rate of their irrigation system and when we try to calibrate the system we find that the given flow rate cannot produce an acceptable calibration. Further investigation usually reveals that the irrigation system's flow rate is different from what the farmer originally thought. We have seen this scenario several times during our field trials, and our technicians can recognize the symptoms which indicate when the system flow rate might be erroneous. The Microsoft Excel-based interface will enable rapid prototyping of heuristics (e.g., a guide for choosing system flow rate) that will help the user select the most impactful system parameters. These heuristics will enable a non-technical user to exploit the benefits of the Auto-Calibration system.The second purpose of the Excel-based interface is to provide a more flexible platform for expert users. We expect that a small but important segment of our user base will be technical service providers. These customers will have more specialized UI expectations and more sophisticated technical aptitude. These users are, almost without exception, expert Excel users. Building an Excel-based interface provides a tool that enables these users to integrate the DSS into their workflow.3. Cloud-based backendThe Simulated Annealing algorithm provides a mechanism to search complex parameter spaces but does so at the expense of computation time. One approach to mitigate runtime is to use more computational resources. We cannot expect our users to purchase more powerful computers. Cloud-based computing provides a convenient and, more importantly, scalable solution for running the Simulated Annealing algorithm. A cloud-based solution enables application of more computing resources as needed and produces a corresponding speedup of the SA algorithm.This part of the project will involve the development of a cloud-based system that will execute the Simulated Annealing algorithm on demand. Additionally, the system will have a simple API that enables the use of the system both from the Excel-based interface (item 2 above) and a browser-based interface (item 5 below). The implementation will be deployed on Azure and will be capable of scaling based on user demand.4. Execution monitoring infrastructureThe Simulated Annealing algorithm could potentially result in run times longer than a typical user is willing to wait. The scalable cloud-based implementation will help reduce run time, but long runs may still occur. The complex nature of the Simulated Annealing algorithm therefore necessitates a robust mechanism for monitoring its execution. The fourth part of this project will develop a mechanism that enables the user to monitor the algorithm's progress and allows unattended execution.5. Browser-based interfaceThe current IMO system has a Vue.js[2] based interface. The fifth part of this project will involve building extensions to the IMO interface that support using the Auto-Calibration system. The functionality of these extensions will be based on the prototyping work done with the Excel-based interface described in part 2. The Vue.js based interface is intended for non-technical users.[1] Brooks, S. P., and B. J. T. Morgan. "Optimization Using Simulated Annealing."Journal of the Royal Statistical Society. Series D (The Statistician), vol. 44, no. 2, 1995, pp. 241-257.JSTOR, JSTOR, www.jstor.org/stable/2348448.[2] https://vuejs.org/