|
H.M. Malano, Principal Research Fellow, and H.N. Turral, Research Fellow, International Development Technologies Centre, and M.L. Wood, Postgraduate Student, Department of Civil and Environmental Engineering, University of Melbourne, Parkville, Victoria, Australia |
SUMMARY
An integrated real-time irrigation scheduling system has been developed for the typical crop and irrigation management conditions in southeastern Australia. The system comprises three main elements, namely: (i) a soil moisture monitoring device capable of measuring soil moisture level on a continued basis; (ii) a medium-term (four-day) weather forecast; and (iii) a decision support system to assist irrigators in making irrigation scheduling and water ordering decisions. Three recently developed soil moisture monitoring devices were tested over two irrigation seasons for accuracy and reliability, and during this time were used to assist in the irrigation scheduling task. The weather forecast variables were also analysed to assess the level of forecast reliability. While the weather forecast exhibits adequate levels of accuracy, the format of the current weather forecast requires changes to make it more applicable to evapotranspiration calculations. A Surface Irrigation Management Decision Support System (SIMDSS) was also developed and its general architecture is described.
Irrigation scheduling in surface irrigated farms requires additional knowledge to predict the performance (time, quantity, uniformity of application) of any event. The second part of the paper therefore summarizes current state-of-the-art techniques to assess infiltration and roughness characteristics in real time, as part of an integrated real-time management system for irrigated pasture in southeastern Australia.
BACKGROUND
Surface irrigation is the main method of irrigation in southeastern Australia, where it is practised in association with pasture production for dairy enterprises. A large proportion of irrigated properties have undergone substantial remodelling of the farm layout and improved land levelling by laser controlled equipment. Typically, the length of the newly levelled fields is 200-500 m, with slopes ranging between 0.1 and 0.04%. In the majority of irrigation districts in southeastern Australia, water is supplied under an arranged limited rate schedule with water charges assessed volumetrically using Dethridge wheel meters. Typical supply discharges range between 80-140 l s-1.
The predominant types of soil in the irrigated areas of southeastern Australia are cracking clay and sandy loam soils commonly called 'duplex soils' because of their characteristic loam to clay upper layer overlying a medium to heavy texture B horizon. The management of irrigation in cracking clay soils poses some special problems including bypass flow in cracks resulting in reduced matrix flow, limited leaching and substantial lateral flow beyond the irrigation units.
FIGURE 1 - Illustration of integrated real-time irrigation scheduling system
Traditionally, irrigation scheduling is considered as a decision-making process used by irrigators to decide when to irrigate their crops and determine the appropriate quantity of water to apply. This concept has proved to be adequate for pressurized irrigation systems in general (spray and drip) but inadequate for surface irrigation where the amount of irrigation is far less controllable.
Although irrigation scheduling has been widely used by 'irrigation experts', farm operators (intended users of the systems) do not regularly use them. Pleban and Israeli (1989) looked at why irrigation scheduling programs have not been widely used. They found two possible reasons were that (i) the programs were oriented toward research and for use by professionals; and (ii) that they were looking at the scheduling problem from the point of view of the crop and the academic researcher and not from the farmer's point of view. The authors suggest that in addition to these, water pricing and environmental policies often are not adequate to provide sufficient incentive or regulation to improve the management of water on the farm.
Real-time irrigation scheduling is a development which has made significant inroads in the field of pressurized irrigation, especially drip irrigation, where the irrigation scheduling system determines the required input for the systems control algorithms. In this type of system, the irrigation scheduling algorithm is an integral part of the system's control algorithm and hardware.
In southeastern Australia the problem of irrigation scheduling for surface irrigation systems supplied by open canal delivery networks is confounded by: a canal supply schedule in which irrigators are usually required to provide four-days' notice of their water ordering; the uncertainty of weather conditions during this period; and the management practices of the crop and the dairy stock. Mechanistic models are inadequate to cope with the complexity of this type of problem. Decision support systems which do not attempt to generate optimal guidelines but rather generate the best guidelines that can be reasonably developed given constraints on available data, actual field conditions and other farming practices, are better suited to assist in this type of task. Such knowledge-based systems and concepts have advantages over conventional techniques as they allow for detailed explanation of reasoning processes, utilization of incomplete and uncertain data, and experimental and expert knowledge (Waterman, 1986).
The traditional irrigation scheduling concept addresses only two variables: (a) timing of irrigation; and (b) amount of irrigation. While the amount of required irrigation can be easily ascertained from routine soil moisture monitoring, the ability of the irrigator to apply the required irrigation amount is far less controllable or understood. Shaffique et al. (1983) concluded that the successful implementation of an irrigation scheduling program directed at surface irrigation requires two additional variables, namely: discharge (Q0), and time of cutoff (tco). Because these variables must be selected on the basis of soil infiltration at the time of irrigation, real-time determination of the soil infiltration parameters is required to predict the outcome of the current event and provide advice on the management adjustments necessary to improve performance.
The main aims of this paper are to discuss the integration of the hardware and software elements of a real-time irrigation scheduling system and to discuss the current state-of-the-art in the assessment of soil infiltration characteristics in real time and their use as part of an integrated real-time surface irrigation monitoring and control system for the prevailing conditions of irrigated pasture for dairy production in southeastern Australia.
IRRIGATION SCHEDULING
Overview of the real-time irrigation scheduling system
Conventional irrigation scheduling is carried out by assessing the water content of the soil profile using water balance models, measuring the soil moisture content of the soil, or a combination of both. Soil moisture measurement is commonly done by using neutron probe meters, tensiometers or gravimetric methods on a recurrent basis. The main feature of such an approach is that evaluation of the soil moisture status is made at certain intervals and requires the irrigator or irrigation consultant to attend to this task personally. A real-time system is one in which the soil moisture monitoring task is carried out continually and automatically.
Figure 1 provides an illustration of the integrated real-time system. In the context of this paper a real-time irrigation scheduling comprises three main elements, namely: (i) a soil moisture monitoring device (SMMD) capable of measuring soil moisture on a continued basis; (ii) the availability of a short-term weather forecast capability; and (iii) a decision support system which relies on field moisture status, weather forecast and crop cultural practices to select the most appropriate course of action in scheduling irrigations.
The system was initially installed at the dairy farm of the Dookie Agricultural College in northern Victoria and has now been in operation for two irrigation seasons. The dairy enterprise was chosen because it is operated and managed on a commercial basis similar to typical dairy enterprises in the region. The test area is sown with a typical summer pasture mix of rye-grass, white clover and paspalum.
Soil moisture monitoring devices
At the start of the study, an extensive survey of available soil moisture monitoring devices was carried out, from which a group of SMMDs was chosen to represent most of the different principles of operation available on the market. The rationale behind the choice of the devices was that they be loggable in their standard configuration, not subject to substantial drift, and adequately priced for adoption by typical dairy farmers in the region. Other devices, such as Time Domain Reflectrometry (TDR) and neutron probe which did not meet these criteria were used for comparative purposes. These devices, however, remain outside the range of affordable equipment by typical growers in the region or do not have the capacity to log soil moisture status continuously (Tyndale-Biscoe and Malano, 1995). The devices selected are:
The Aquaflex SMM:
The Aquaflex SMM was developed at the Agricultural Engineering Institute at Lincoln University, New Zealand and is not yet available commercially. It works on the principle that a wave propagates down a line (in this case a five-metre flexible ribbon) and is reflected back from the end. The speed of propagation of the wave is proportional to, amongst other things, the dielectric constant of the surrounding material which in turn relates to the moisture content (water has a dielectric constant approximately 20 times greater than soil particles). The device measures the delay between transmission and reception of the wave and generates a frequency which is proportional to this delay. The Aquaflex proved to be highly reliable with low downtime during the experiment. Soil moisture monitoring readings from this probe were used in making irrigation scheduling decisions.
The Microlifik1 SMM: This device works on the heat pulse principle. A pulse of heat of known energy is emitted from a sensor buried and in intimate contact with the surrounding soil. The time taken for the heat pulse to dissipate into the surrounding soil is proportional to the moisture content and so can be calibrated accordingly. In its current configuration, soil moisture can be read directly from the visual display but it is not retrievable by computer. Several problems hampered the performance of this probe which rendered it ineffectual for making irrigation scheduling decisions.
1
Microlink is a trademark of DRW Engineering, South Australia.
The Enviroscan2 SMM: This device uses a capacitive technique to measure the soil moisture. The probe consists of copper plates which are lowered in a permanently installed PVC access tube. The device measures the capacitance which is related to the dielectric constant of the material between the plates and its moisture content. This system assumes an exponential relationship between soil moisture and capacitance. Together with the Aquaflex, this probe performed satisfactorily over the entire length of the experiment.
2
Enviroscan is a trademark of Sentek Pty Ltd., South Australia.
The three selected devices were tested during the first part of the first irrigation season in a soil tank with sandy loam topsoil commonly used for gardening. The purpose of the experiment was to become familiar with the equipment, iron out possible flaws and allow the comparison of the SMMDs with each other under controlled conditions. Later in the season, two sets of SMMDs were installed in an irrigation bay at one-third and two-thirds of the bay length where they remained for the entire length of the trial. The selected bay was 300 m long by 40 m wide with a slope of 0.2%. The soil type is Shepparton Fine Sandy Loam and uniform throughout the bay. Measuring flumes were installed to monitor the amount of water applied in each irrigation event.
The probes were tested under field conditions for two years and rated according to their reliability and accuracy. The reliability criterion was judged on the two main factors which are likely to determine the success of real-time scheduling and farmers' trust in the technology, namely: length of downtime, and manufacturer's support.
Accuracy of measurement of soil moisture content can be expressed in absolute and relative terms. Absolute accuracy refers to the ability of the device to produce readings of the actual moisture content of the soil. Relative accuracy is the ability to reflect changes in soil moisture content accurately although the absolute reading may not represent the actual water content. The ability of the device to reflect actual absolute accuracy depends on the adequacy of the calibration procedure.
Short-term weather forecast
Most of the irrigation schemes in southeastern Australia supply water through a scheduled arrangement whereby irrigators must often provide four or more days of notice to the irrigation authorities. Under typical farming management conditions, there are several elements that the farmer must take into account in making the decision of when to irrigate, e.g., the likely weather conditions during the notice period, the timetable for stock rotation, and other crop management operations.
Four-day weather forecasts are currently available from the Australian Bureau of Meteorology on a regional basis. The forecast variables are wind (km/hr), maximum and minimum temperature (°C), rain probability and frost risk (Malano and Wood, 1995). This forecast was not designed to meet the specific requirements of farm irrigation scheduling, as it does not lend itself for the calculation of evapotranspiration on a daily basis using the Penman-Montieth model. An expanded weather forecast service including solar radiation, sunshine hours and relative humidity is due to be implemented by the Bureau of Meteorology. During the field trial, irrigation scheduling decisions did not include forecast evapotranspiration rates.
The accuracy of the weather forecast was tested over a period comprising the entire 1993-94 growing season (150 days) by comparing the forecast with observed weather data from an on-site automatic weather station. Table 1 shows the comparison between forecast and observed maximum and minimum temperature. Observation of the R2 reveals that as expected the accuracy of weather prediction diminishes as the period of prediction becomes longer. In addition, the maximum temperature can be predicted with a greater degree of accuracy than can the minimum temperature. Wind speed and rainfall predictions was also ascertained for the entire test period.
TABLE 1 - Values of coefficient of determination (R2) and coefficient of efficiency (E) for comparison of forecast and actual temperatures
|
Forecast Day |
Temp |
E |
R2 |
|
Day 1 |
Max |
0.88 |
0.91 |
|
|
Min |
0.65 |
0.65 |
|
Day 2 |
Max |
0.73 |
0.81 |
|
|
Min |
0.65 |
0.66 |
|
Day 3 |
Max |
0.64 |
0.74 |
|
|
Min |
0.61 |
0.63 |
|
Day 4 |
Max |
0.53 |
0.69 |
|
|
Min |
0.55 |
0.69 |
ARCHITECTURE OF THE SURFACE IRRIGATION MANAGEMENT DECISION SUPPORT SYSTEM (SIMDSS)
The aim of expert systems is to step through decision-making processes considering all the variables involved and evaluate the most appropriate solution as an expert in the field would. There are a wide range of systems from simple rule based systems which rely only on heuristic rules to handle the problem of decision support to integrated expert systems (Stone et al., 1986). The design of the Surface Irrigation Management Decision Support System (SIMDSS) falls into the latter category.
FIGURE 2 - Structure of the SIMDSS program
The key elements of the SIMDSS architecture are:
· Decision rules: generalizations or heuristic rules (rules of thumb) which describe how a human expert would make irrigation scheduling decisions.· The database: contains lists of products, objects (e.g., crop types), conditions or procedures and their characteristics for a particular region, for example, soil water retention characteristics.
· The inference engine: contains the control structure of the expert system and determines what information will be used and how it will be used. It accesses the rules in the knowledge base, executes the program, and determines when the software has produced the best answer to the user's problem.
· The user interface: Allows the user to communicate with the computer.
The two principal aims that guided the development of the system were:
· to provide irrigation scheduling management decision support in real time; and· to enable the user to carry out retrospective diagnostic analysis of irrigation scheduling performance during the irrigation season.
The general philosophy which guided architecture of the system software was based on the following premises:
· The software program should have an interface that is simple to use by irrigators with access to a personal computer but who have no professional qualification and require minimum training.· The program should be based on a modular design to permit future expansion of the software to incorporate other aspects of water and crop management.
The computer implementation of the SIMDSS was developed using KnowledgePro-(Win)1 expert system programming shell. The structure of SIMDSS Version 1.0 is illustrated in Figure 2 and consists of four main modules, namely:
KnowledgePro is a trade mark of Knowledge Garden Inc.
1
· Farm setup: This module contains all the information concerning the physical characteristics of the farm including layout, soil types, types of pasture, and instrumentation. The information contained in this module is loaded at the time of installation of the system and will remain unchanged unless the layout or cropping pattern of the farm changes.
· Data input: This sub-menu enables the user to input the information required for the program to assist in making irrigation decisions. This includes the daily weather forecast and recent weather history, time constraints to carry out waterings, etc. This sub-menu also allows the user to input performance monitoring information such as water use, etc.
· Irrigation schedule: This sub-menu is used to present the output from the program execution. This contains the irrigation schedule for each field as well as the supporting information used in decision making.
· Background information: Contains background information which can be accessed via hypertext to assist the user in understanding the decision making rationale.
Field implementation of real-time irrigation scheduling
The integrated system was trialled during the 1994/95 irrigation season to assist in the scheduling of irrigations for a set of four irrigation bays including the test bay described above. During this season, irrigation scheduling was carried out without the aid of SIMDSS, which was still in the process of development. The knowledge gathered during this season was used to build the knowledge base and decision rules of the system.
The retrospective analysis of the irrigation performance shows that the main benefit of the system accrued both in the early and late parts of the irrigation season when decisions on when to irrigate are more complicated due to the higher proportion of crop water requirement met by temporally variable precipitation.
Figure 3 shows soil moisture, rainfall and irrigation events during the first part of the 1994/95 irrigation season. Two main shortcomings in the timeliness of irrigation can be observed, namely:
· Irrigation was applied at an inadequate soil moisture level, either too high or too low.· Irrigation was sometimes immediately followed by a precipitation event leading to poor effective use of rainfall.
The shortcomings were due in part to the lack of familiarity with the system on the part of the farm manager. Because this is a typical situation likely to be encountered in future with the adoption of the system by farmers, it was used to develop the SIMDSS inference engine and to identify the key aspects that are to be emphasized during the training of new users.
FIELD APPLICATION
The application of water onto the cropland by surface irrigation must rely on the soil surface as a conveyance medium; one which exhibits constantly changing hydraulic characteristics such as infiltration and surface roughness. The task of managing the application of water is centred on the selection of the appropriate combination of inflow discharge and time of cutoff with both being very sensitive to infiltration and roughness characteristics.
This discussion attempts to analyse the analytical framework and hardware requirement for the future addition of the field water application component of SIMDSS.
Infiltration is the most crucial variable governing the performance of surface irrigation. Traditionally, infiltration parameters used in design and management have been determined from past irrigation events using either point methods such as ring infiltrometers, blocked furrow tests, etc.; or by analytical methods which indirectly determine the infiltration parameters from measurable hydraulic variables such as rate of advance and flow depth. Infiltration, however, is a highly variable soil property both temporally and spatially which often means that infiltration parameters determined in previous irrigations are largely inapplicable to the current event. An additional complication arises in border irrigation sown to pasture on cracking soils, in that hydraulic roughness is also highly variable, depending on crop type, stage of growth, extent of cutting or grazing, plus the health and uniformity of stand. In practice, it assumes an equal importance to infiltration characteristics and must also be determined locally at the time of irrigation for optimum real-time management.
Parameter estimation by inverse solution on cracking clay soils
Extensive testing of inverse solutions for infiltration and roughness parameters was undertaken as part of a research programme into the performance of border irrigation on cracking clay soils (Turral, 1993). A fundamental assumption of this work is that it is not realistic to assume that hydraulic roughness (as described by Manning's 'n') is known or can reasonably be 'guesstimated' for border irrigation. It is, therefore, one of the parameters that must be optimized if real-time control of border irrigation is to be attempted.
A Zero Inertia model (BRDRFLW, Strelkoff, 1986) was coupled to a constrained Simplex optimization algorithm. The ZI model proceeds with the simulation of advance until it reaches the user-input time limit, after which it calculates the advance curve and depth profile at that time, and returns these to the interpolation and curve matching module. The goodness of fit between observed profiles (field data) and the modelled results is estimated and if the tolerance is less than the predetermined threshold, the optimal parameters have been estimated. Otherwise iterations continue until the tolerance criteria are satisfied. Field measured advance distance and time pairs, or distance and depth pairs, are supplied to the model by the input file and the corresponding profiles determined by ZI simulation are adjusted to the x co-ordinates of the field data by linear (advance) and quadratic interpolation (depth).
Choice of optimization algorithm
A number of considerations arise in the choice of the optimization algorithm:
· rapid convergence is desirable;
· minimum computational load;
· must avoid local minima, e.g., reliably determine the global optimum;
· stable convergence.
The relative merits of the different approaches mean less today given the computer power available on the average desktop and (except where many parameters are to be optimized in complex models) it probably does not matter which technique is used. The Simplex algorithm was chosen for its robustness and simplicity and well-documented reliability (Press et al., 1989) and because the flexible tolerance algorithm can be applied easily to define constraints to limit parameter values within realistic and non-negative values.
Choice of objective functions
Least square curve matching was implemented using three objective functions (OF):
i. Advance curve matching (ordinary least squares):
ii. Depth profile matching (weighted least squares):
iii. Compound function of depth and advance:
where:
T0 = observed advance time
Ti = simulated advance time
x = distance down bay to each observation
L = distance to point of maximum advance in the series
Y0 = observed flow depth
Yi = simulated flow depth
The depth profile objective function was weighted by the distance down the bay of each estimate ((o-i)2.x) because the critical portion of the depth profile for determination of infiltration is the strongly curvilinear part behind the wetting front.
Testing schedule
Field data had been collected over two years to evaluate surge flow in comparison to conventional application on a range of soil types, but predominantly cracking clays. The experiments were conceived with a view to model calibration and possible use in estimating surge flow infiltration parameters by inverse solution. Complete inflow and outflow hydrographs were available, in addition to hydrographs for up to 60 individual stations within one field. This surface flow depth data was collected using specially developed digital depth sensors, which were sampled continuously throughout the duration of an irrigation and logged to disk using a portable computer. Advance, recession and flow depth profiles could be constructed for any selected time during an irrigation event. At two sites, moisture samples were taken using a neutron probe at up to 20 locations in a field to enable quantification of antecedent moisture contents and estimate net infiltration and infiltration uniformity.
A systematic investigation of performance of the inverse solution was conducted using three infiltration equations:
Simple Kostiakov
z=kT
a
Modified Kostiakov
z=kT
a+bT
Linear Model
z=C+bT
where
z = cumulative infiltration in millimetres
T = opportunity time for infiltration in minutes
k = Kostiakov constant millimetres/minutea
C = crack fill constant in millimetres
b = steady state infiltration rate in millimetres/minute
a = exponent
These equations were selected on the basis of field tests using a recycling infiltrometer and for compatibility with the Zero Inertia model. Fitted infiltration functions indicated that a Simple Kostiakov expression, with a high value of constant k and a small value of exponent a represented the data well, as there was considerable evidence to show that final infiltration rates approach zero at long opportunity times on cracking soils. The model incorporated the ability to fix any of the infiltration and roughness parameters, to investigate the effect of characteristic values of parameters, such as b which can conceivably be measured using ring infiltrometers, or by measuring vertical recession of ponded water.
Initially screening tests were performed on 48 combinations of OF, infiltration function and fixed parameters for one continuous event at one time of optimization. The investigation was systematically expanded to cover other events, as follows:
· establish robustness of inverse solution, including: (a) radius of convergence; (b) consistency of solution following restarts; (c) time of optimization, corresponding to advance distances of 100, 200 and 250-300 m on 400-m long bays;· establish performance of optimized parameters by comparing modelled irrigation performance (advance, recession, net infiltration, infiltration uniformity, runoff volume) with field data;
· establish minimum number of data points for acceptable simulation of field irrigation using optimized parameters. The practical interest of this work is to minimize sensor number and therefore cost of a real-time management system;
· generation of the response surface to investigate parameter interdependence and the robustness of objective functions.
General results
· The compound objective function was shown to give the most robust performance, with indications that four parameters could be reliably estimated (Modified Kostiakov equation plus roughness). Three parameters could consistently be determined using the depth objective function with the Simple Kostiakov expression (two infiltration and one roughness coefficient). In both cases, advance, net infiltration and runoff were reliably modelled, although recession was in general in poor agreement.
· The infiltration models do not exactly describe infiltration on a cracking soil, and therefore the optimized values of the variables may not have any physical significance. Similarly optimized values of roughness tend to be less than those determined from measured flow profiles, partly as cracks actually contribute to effective roughness in the field. This contributes to the shorter recession times generated by the model, as roughness at low flow depths is underestimated by the single variable. Manning 'n'.
· The simple linear infiltration function is conditionally observable but consistently underestimates the net infiltration in the field and is not a useful model for inverse solution, as it also exhibited the greatest reluctance to converge to an optimum result. Response surface tests showed that if the steady state infiltration rate is less than six millimetres per hour (as is the case with most cracking soils in Australia), the Zero Inertia model is too insensitive to small values of b to allow an optimum value to be determined.
· The calculation of a 'global optimum' is ensured by repeating the optimization using the output from the first run as starting parameters for the second one (a 'restart'). This means that the choice of starting values is not restricted to those close to the global optimum (expected values).
· The minimum number of depth points required for reliable convergence of the depth and compound objective functions was determined to be four, implying significant instrumentation costs, although a considerable improvement on Katapodes' (1990) requirement for ten points on a depth profile.
· Parameters optimized at different times in the advance phase are not consistent due to differences in infiltration and roughness conditions over successive portions of the field.
· The most representative point at which to determine the 'average condition' is location specific and must be predetermined by knowledge of vegetation and drainage conditions. The preferred location for well-drained fields is after the field mid-point and before advance reaches TCO.
· Improvements in the reliability of convergence were found when the average inflow to the time of optimization was used in place of the average inflow for the complete event, which accords with other researchers' findings that the ZI model is quite sensitive to small changes in inflow rate (Maheswari, 1988)
· Gross variations in antecedent moisture condition are typical in a field situation and can be any one (or worse, a combination) of the following:
· Dry bottom of bay due to incomplete advance in previous watering.
· Wet head of bay due to surface or subsurface seepage from supply channel.
· Wet tail of bay due to restricted drainage and/or generous prior irrigation.
· Variations in crop water use due to strip/paddock grazing, or cutting (hay).
· Variation in crop water use due to crop type (e.g., lucerne and grass mixes).
Field instrumentation for real-time monitoring
The minimum data and hardware requirement for real-time management using an inverse solution comprises:
· Field inflow rate, cumulative inflow and field width are needed to enable specification of average unit discharge to the point and time of optimization. It is possible to instrument the Dethridge wheel meter at the head of the farm channel, but the accuracy of individual field inflow is likely to be compromised by leakage from the channel or by increasing in-channel storage, or possibly both.· Field slope
· Three advance points, using soil moisture sensors; or four depth and advance points using purpose-built depth sensors, for increased reliability and flexibility using the compound or depth objective functions.
· Telemetry or a hard-wired link is required to communicate the advance/depth and discharge information to a computer or similarly powerful dedicated controller, which will also be used to control the closure and opening of the field inlets.
· Automatic gate actuation. Hydraulically actuated gates developed in Australia look to be the most promising option, and have been evaluated for channel and farm-gate operation by the water authority.
CONCLUSIONS
A surface irrigation management strategy under the conditions of southeastern Australia must rely on two basic elements, namely: (a) irrigation scheduling; and (b) field water application.
Irrigation scheduling
The irrigation scheduling strategy for surface includes three basic component, namely: (a) continued soil moisture monitoring; (b) short-term weather forecast; and (c) a decision support system that can use the monitored variables to assist the irrigator to make decisions on irrigation scheduling and water ordering.
The program has been in place for two seasons in which the SMMDs were tested for accuracy and reliability. Two probes were selected from a set of three for their performance. The probes were employed to assist in making irrigation scheduling decisions during the 1994/95 irrigation season without the use of weather forecasts. Although better overall irrigation timing was achieved, some shortcomings still occur due to untimely irrigations or the inability to consider the possibility of rainfall occurring immediately after an irrigation.
The entire system is intended to be operated on a network of five irrigation farms during the 1995/96 irrigation season where detailed monitoring will be carried out and their irrigation performance compared with the current baseline levels.
Field application
The analytical framework for real-time determination of hydraulic variables (infiltration and surface roughness) and the hardware requirement for the field application component of SIMDSS were analysed. An inverse solution consisting of the Zero Inertia model coupled with a constrained Simplex optimization algorithm was used and tested on a range of soil conditions but predominantly cracking clays in southeastern Australia. Three objective functions based on curve matching, depth profile and a combination of both were examined together with three forms of the Kostiakov equation. The compound objective function gives the most robust performance being capable of estimating up to four parameters (three infiltration + roughness), but the instrumentation requirement makes it impractical for field implementation. Two parameters (roughness and infiltration) are conditionally observable using the advance rate and the Simple Kostiakov equation. This approach was further analysed by fixing the value of the Kostiakov exponent (a) as previously determined by infiltrometer tests; and (b) varying the distance and number of measuring points down the border. A three-point system including the inlet showed similar performance to the full data set.
REFERENCES
Katapodes, N.D. 1990. Observability of surface irrigation advance. J. Irrig. and Drain. Div. ASCE 116 (5): 656-675.
Maheshwari, B.L. 1988. Evaluation of Border Irrigation Models for South-East Australia. Ph.D. Thesis, Univ. Melbourne, Dept. Civ. and Agric. Engrg. 291 p.
Malano, H.M. and Wood, M. 1995. Milestone Progress Report 3. Report submitted to the Land and Water Resources Research Corporation, University of Melbourne. 47 p.
Pleban, S. and Israeli, I. 1989. Improved approach to irrigation scheduling programs. J. Irrig. and Drain. Eng. ASCE 115 (4): 577-587.
Press, W.H., Flannery, B.P., Teukolsky, S.A. and Vetterling W.T. 1989. Numerical Recipes, The Art of Scientific Computing. Cambridge Univ. Press, New York. 702 p.
Shafique M.S., Skogerboe, G., Loftis, J., Scheitlin, S.M., Rude, P., Hatami, B. and Bargsten, G. 1983. Monitoring and Evaluation of On-Farm Improvements in the Grand Valley Salinity Control Project during 1981 and 1982 Irrigation Season. Department of Agricultural and Chemical Engineering Colorado State University. 130 p.
Stone, N.D., Coulson, R.N., Frisbie, R.E. and Loh, D.K. 1986. Expert systems in entomology: Three approaches to problem solving. Bull. Entomological Society of America 32: 161-166.
Strelkoff, T. 1985. BRDRFLW: A Mathematical Model of Border Irrigation. USDA, ARS Publication 29. 104 p.
Turral, H. 1993. The Evaluation of Surge Flow in Border Irrigation with Particular Reference to Cracking Soils. PhD thesis, University of Melbourne, Parkville, Australia.
Tyndale-Biscoe, P. and Malano, H.M. 1995. A laboratory comparison of some currently available soil moisture monitoring devices. Agricultural Engineering Australia 1: 18-24.
Waterman, D. 1986. A Guide to Expert Systems. Addison-Wesley Publishing, Reading, MA, USA.
