|
M. Cabelguenne, Ingénieur de Recherches, INRA Centre de Recherches de Toulouse, Castanet Tolosan, France; J. Puech, Ph. Debaeke, N. Bosc, A. Hilaire (technical collaboration O. Girotto, P. Petibon, P. Rouet) |
SUMMARYThe real time EPIC-phase model developed at the Toulouse Auzeville INRA Station, was tested in an experiment on maize. The main objective of the trial was to evaluate the potential of the model in real time tactical irrigation management based on model predictions every five days. The trial compared the results of farmer irrigation management based on simple decision-making rules with those resulting from the use of the model.
Compared with the standard version of EPIC, which is capable of simulating the evolution of soil water and nitrogen content, the option of 'real time' enables integration of intermediate states and realization of simulations step by step. The trial reported here has demonstrated that it is possible to simulate different irrigation tactics and, hence, different irrigation amounts with weather forecasts.
Results from the study showed that discrepencies between actual and forecast weather led to differences in tactical irrigation management.
Simulations carried out with observed data demonstrated the capabilities of the model as an irrigation management decision support tool. Compared with farmer management, use of the model led to increases in yield and reductions in the irrigation water applied. Use of the model also reduced the risk of overirrigation and associated nitrogen leaching.
The effect of economic constraints linked to the new Common Agriculture Policy (CAP reform), variations in availability of water and in climate, and the environmental risk of excessive irrigation give rise to a need to manage irrigation more and more rigourously with regard both to the adjustment of the resource to the yield objectives and to the leaching of nitrates when water supply is not a limiting factor.
Every year the list of models for irrigation management and drainage lengthens (Bonnet, 1994): among 140 models noted the majority make use of the Doorenbos and Kassam water production functions (1979) or climate formulae.
These models are unable, however, to forecast correctly the effect of water constraints on the growth of the plant because they take no account of dynamic functions. On the other hand, mechanistic agronomic models which are day based, such as CERES-MAIZE (Jones and Kiniry, 1986), EPIC (Williams et al., 1989), CROPSYST (Stockle et al., 1994). have the advantage, if correctly calibrated, of simulating the effect of water depletion during the growth cycle and so of being usable as tools to forecast the water state of the soil and the crop. It then becomes possible to manage water resources on the basis of simulations of function indicators of both soil and plant.
At a decision-making level, the tactic may be defined as a series of short term decisions made on the basis of knowledge or forecast of the soil moisture, of the plant and of climate conditions. In such a case, it is important that the farmer is able to anticipate the evolution of the function indicators of the soil and the plant in order to simulate their effects on plant development and yield.
New models have been developed with this in view either for a single crop (Texier et al., 1993), or for several species (Plauborg et al., 1991). The INRA agronomic centre in Toulouse has chosen to produce a version of EPIC functioning in real time. This interactive version was tested on maize in order to compare irrigation tactics based on farmer decision rules with those produced by the model.
This article aims, on the one hand, to present the results of the test, to assess the value of simulating soil and plant function indicators to manage irrigation tactics and gauge the accuracy of forecast climate data, and, on the other hand, to identify the obstacles which, for the moment, prevent farmers from making practical use of this model.
MATERIAL AND METHOD
The model
As described by Cabelguenne et al. (1994a, b), EPIC-Phase real time is the result of successive modifications of the standard American version of EPIC with a view to producing a model better adapted to the water management of agricultural crops.
A first modification consisted in developing a new growth module for plants taking into account several phases based on sums of temperature, in modulating by crop the ability to extract water from the soil by roots, and in varying the harvest index as a function of the water stress suffered by the crop at different stages.
The second modification was to transform EPIC-Phase into an interactive program which would allow the user:
· to stop the simulation at any day and visualize the soil and plant variables;· to modify the values of these variables from measurements and resume the simulation;
· to load short-term weather forecasts in order to anticipate the evolution of soil-water depletion and stress, and then to simulate their effects on crop development (leaf area index, biomass) and yield.
Different interactive menus allow the user to select the stop day for the simulation, to go backwards if necessary and visualize indicator states on a daily basis. The ease of accessing these states, which fluctuate and are difficult to measure, and of gauging their effects on yield, makes this model a useful tool for decision making.
The experimental site, the decision rules
Located at Auzeville on deep sandy clay soil, the experiment consisted of two irrigated blocks with a variable water resource (>2 500 m3.ha-1 with a yield objective of 11 t.ha-1, and limited to 1 500 m3.ha-1 with a yield objective of 8-9 t.ha-1 variable according to the climatic constraints of the year). Each block was divided in two plots, one was managed using simple decision-making rules, so-called farmer management, the other with the model. The 300 m2 plots were irrigated by sprinklers. The maize was planted on 2 May at 78 000 plants.ha-1 and received 80 units of ammonium-N and 130 units of nitrate-N.
Irrigation of the farmer management plots was under non-limited water supply with a first application before flowering whenever soil water depletion reached 50 mm, followed by 35-mm applications every eight days, delayed n days according to n= rain (mm)/5. The end of irrigation season was based on a temperature sum of 1 650 degree days (base 6). Under limited water supply, a first irrigation of 30 mm was applied at the end of June, if no rainfall > 10 mm was observed for ten days, followed by four 30-mm applications every ten days, delayed according to n = rain (mm)/6.
The management with the model was based on the evolution of soil water depletion and water stress intensity (0 = maximum stress, 1 = no stress), predicted using weather forecasts provided by Meteo-france at Toulouse/Blagnac. Concretely, historical weather data were replaced every five days by weather forecast for the next five days, and then by observed weather at the end of the five-day cycle. The decision rules consisted of irrigating when the accumulated simulated stress for the period reduced the yield by 5% of the yield objective. As it was possible to anticipate the effects of water depletion and stress on yield, it was possible to adjust irrigation according to the available water.
RESULTS AND DISCUSSION
Calibration of the model: comparison of measured and simulated values for leaf area index, biomass and grain yield
Before evaluating the model in terms of its usefulness in assisting decision making, it was important to evaluate its accuracy with regard to the simulation of biomass, leaf area index and yield compared with actual measurements. Ten plants were sampled on sub-plots of 0.2 m2 at eight growth stages from emergence to harvest. Figure 1 shows a very good agreement, under non-limited water supply, between measured and simulated leaf area index, throughout the cycle. The evolution of the biomass and grain yield were very satisfactory with a maximum difference of 10% at harvest compared to the measured data. This is a smaller difference than the variation coefficient observed during experiments (Cabelguenne et al., 1986). Under limited water supply (Table 1), the low value for the measured biomass (M) on the model management plot, compared to the simulated value (S), is due to an attack of Zyginidia scutellaris not simulated with the model. However, this attack had no effect on the yield and thus the measurement result here was very close to the simulated value. The relative high yields obtained under limited water supply were higher than the yield objective and unusual for the region. This was due to the exceptional rainfall received in June (70 mm), which maintained the soil at field capacity till the end of June and met the crop requirements up to the beginning of flowering phase, so delaying water depletion at the grain filling phase and being less penalizing to the yield.
Results of measured (M) and simulated (S) data under limited water supply to 1 500 m3 ha-1
|
|
Biomass (t/ha) |
Grain yield (t/ha) |
||
|
M |
S |
M |
S |
|
|
Farmer management |
19.2 |
18.9 |
10.2 |
10.4 |
|
Model management |
15.7* |
18.3 |
8.8 |
9.1 |
* Diseases on leaves, not simulated with the model.
Once the accuracy of the model had been verified, it became possible to evaluate its usefulness as a decision-making tool in the tactical management of irrigation.
The influence of weather forecast on the choice of irrigation tactics
Figure 2 compares forecast data for rainfall, radiation (cumulative value) and windspeed to those actually observed at the INRA Auzeville weather station. Considerable differences are observable; as far as rainfall is concerned, there was an overestimation apart from late June when it was underestimated due to heavy very localized storms. For the June to August period the estimate was for 200 mm whereas the measured amount was only 100 mm. Similarly the radiation, expressed in cal/cm2, and converted into Megajoules/cm2 by the model was overestimated for the whole period. This overestimation had an important effect on the calculation of the potential evapotranspiration (PET) as the model used the Penman Monteith equation. Windspeed was also overestimated, sometimes quite significantly. Only maximum and minimum temperatures and relative humidity were correctly estimated (not shown in the figure).
The poor forecasts led to irrigation tactics being different from those which would have been applied had the forecast been fulfilled or if they had correctly anticipated actual observations. Table 2 details the effects of two scenarios on the tactics compared to the actual case, taking as an example the plot managed on the basis of a non-limited supply: if the forecasts had been correct (scenario 1), the total irrigation would have been concentrated more on the period of flowering and the start of grain filling, given a greater forecast demand (effect of radiation and windspeed) reflected in higher PET and ET (+19 and +16%). Under these conditions the yields were greater (+11%). On the other hand, had the actual observations been correctly anticipated (scenario 2), the tactic would have consisted in delaying the first irrigation until 5 July because of the late June storms, and ending the irrigation on 10 August. The yield would have been only marginally higher than the actual case, but with overall irrigation reduced from 235 mm to 210 mm. These simulations are revealing about the effect of poor forecasts on tactics, irrigation amount and yields.
FIGURE 2 - Comparison between weather forecast and observed data
TABLE 2 Effects of two climate scenarios on the irrigation tactic: comparison with the actual case
|
ACTUAL CASE |
SCENARIO 1 |
SCENARIO 2 |
|
With weather forecast updated every 5 days |
If weather forecast well predicted |
If observed weather well predicted |
|
Actual tactic applied: |
Simulated tactic: |
Simulated tactic: |
|
DATE mm |
DATE mm |
DATE mm |
|
6/23 30 |
6/30 30 |
7/05 30 |
|
7/11 30 |
7/04 30 |
7/11 35 |
|
7/20 35 |
7/11 35 |
7/20 35 |
|
7/26 35 |
7/15 35 |
8/01 35 |
|
8/04 35 |
7/20 35 |
8/06 35 |
|
8/11 35 |
7/26 35 |
8/10 40 |
|
8/17 35 |
8/04 35 |
|
|
TOTAL: 235 mm |
TOTAL: 235 mm |
TOTAL: 210 mm |
|
Results |
Results |
Results |
|
Biom: 19.6 t/ha |
22.2 (+12%) |
19.9 |
|
Yield: 10.8 |
12.2 (+11%) |
11.2 |
|
PET: 661 mm |
820 (+19%) |
662 |
|
ET: 469 |
557 (+16%) |
467 |
Importance of simulating soil water depletion and water stress in real time on tactical choices
Knowledge of soil water depletion alone is an insufficient indicator for tactical management. Since available water varies according to soil characteristics, the rooting depth and the crop water extraction capacity (Maertens and Cabelguenne, 1971), the same levels of water depletion may generate different stress intensities. It is thus important to know both the soil water depletions, the water stress intensity, and to anticipate their effects on crop growth.
Figure 3 demonstrates the changes in water depletion arising from the irrigation tactics actually applied and those retrospectively simulated on the basis of observed data. In the case of model management, under limited water supply and with weather forecasts (Figure 3a), the actual tactic caused water depletion and water stress at a relatively high level (late July and early August) as well as drainage (PRK) and nitrate leaching (PRKN). The yield was 9.1 t.ha-1. By contrast, the retrospective simulated tactic using observed weather data would have reduced the water depletions and stress levels, avoided drainage and leaching, and significantly increased the yield (10.3 t.ha-1). In late September the water depletion would have been similar to the real case, showing that for the same amount of irrigation the additional yield is due to a better use of soil water during the cycle.
In the case of the farmer management (Figure 3b), the comparison also comes down in favour of the simulated tactic, with a slight increase in yield and total elimination of drainage and leaching.
Under non-limited water supply and with the farmer management (Figure 3c), the simulated tactic shows that it would have been possible to save 63 mm by avoiding the applications of 16 June (because of the late June storms not predicted) and of 18 August (because of the low soil water depletion level). Yields for both tactics are comparable but, in the case of the retrospective simulation, the drainage and nitrogen leaching would have been reduced and the soil water depletion at harvest would have been higher, which would have delayed the risk of drainage and nitrogen leaching during the recharge period (Puech et al., 1978).
As for the model management (Figure 3d), the simulated tactic demonstrates that irrigations on 5 July and 10 August would have reduced the soil water depletion at periods of high water sensitivity. This tactic which better matches the climate constraints of the year, would also have enabled a more efficient use of the soil water reserve, since at harvest there is an identical water depletion level for both strategies despite lower total irrigation (210 mm as opposed to 235 mm).
On the other hand, examination of actual tactics as between farmer management and model management in both limited (Figures 3a and 3b) and non-limited (Figures 3c and 3d) water supply merit comment as lower yields are observable with the model, particularly in the limited available water. This may be explainable by the fact that, in the light of the high precision of the plant data, the underestimation of the yields is due to differences in soil depth between the plots and, thus, in water reserve, or because the poor weather forecast which led to a management not adapted to the actual climatic constraints. In this case, the apparent greater efficiency of the farmer management is due to the combined effect of the year's favourable climate conditions and the high soil water reserves which, in a limited water supply context, allowed above average yields to the achieved. These effects served to conceal the possible consequences of poor farmer management irrigation. Conversely, the year's climate conditions did not enable the model to demonstrate its ability to select more efficient management tactics. In conclusion, both forms of management were equally efficient and the yields obtained are higher for an irrigation input of 150 mm, given the slight difference observable in a non-limited water context.
CONCLUSION
At a methodological level this experimental test represents the first practical trial of the EPIC-Phase real time model for helping decision making in the tactical management of irrigation. If this model is a potential step forward in water management, its practical application to an irrigated area or a whole farm would necessitate modifications and could only emerge from close collaboration between researchers and extension services to develop a version especially designed for farmer use.
Among the possible modifications, it would be profitable to remove the set of modules which have no direct link with irrigation management and to increase its interactivity by creating, for instance, files for predetermined soil types with different textures, water reserves etc., and crop files representing the principal cultivars in current use.
Concerning the use of weather forecast data, which are clearly of interest for anticipating soil water depletion and improving water management in space and time given the available resources, the problem is more delicate. Since this experimental test has shown that it is risky to base oneself on a five-day forecast, it is imperative to find other solutions. For example, it would be preferable to update the data on a daily basis as the forecasts are defined down.
REFERENCES
Bonnet, G. 1994. Base de données LOGID de la CIID sur les logiciels. Association Française pour l'Etude des Irrigations et du Drainage (AFEID). January 1994.
Cabelguenne, M., Charpenteau, J.I., Jones, C.A., Marty, J.R. and Rellier, J.P. 1986. Conduite des systèmes de grande culture et prévision des rendements: Tentative de modélisation. II- Etalonnage du modèle: résultats et perspectives. CR. Acad. Fr. 72(2): 125-132.
Cabelguenne, M., Debaeke, Ph. and Puech, J. 1994a. Simulation de stratégies et de tactiques d'irrigation en conditions de ressources en eau limitées. 17ème Conférence Régionale Européenne sur les Irrigations et le Drainage. ICID/CIID Varna, 1994. pp. 39-48.
Cabelguenne, M. and Debaeke, Ph. 1994b. Simulation of short term tactical irrigation under limited water resources. Proc. 3rd ESA Congress, Abano-Padova. pp. 328-329.
Doorenbos, J. and Kassam, A.H. 1979. Yield response to water. FAO Irrigation and Drainage Paper 33. FAO, Rome. 193 p.
Jones, C.A. and Kinity, J.R. 1986. CERES-MAIZE, a Simulation Model of Maize Growth and Development. Texas A&M University Press. 194 p.
Maertens, C. and Cabelguenne, M. 1971. Influence de l'irrigation sur les modalités d'utilisation de l'eau du sol par différentes cultures annuelles et pluriannuelles. Cr. Acad Agr. Fr. 56: 926-936.
Plauborg, F. and Olesen, J.E. 1991 Development and validation of the model. MARKVAND for irrigation scheduling in agriculture. Tidskrift for Planteavl. Beretning S 2113. Danish Institute of Plant and Soil Science, Copenhagen. 103 p.
Puech, J., Marty, J.R. and Hernandez, M. 1978. Bilans hydriques pluriannuels sous diverses rotations culturales irriguées ou non. Société hydrotechnique de France. XV° journées de l'hydraulique Toulouse 5, 6, 7 sept. 78.
Stockle, C.O., Martin, S.A. and Campbell, G.S. 1994. CROPSYST, a cropping systems simulation model: water/nitrogen budgets and crop yield. Agric. Syst. 46: 335-359.
Texier, V., Blanchet, R. and Bouniols, A. 1993. Influence of pluriannual weather conditions on parameters of sunflower growth modeling in Southwestern France. Proc. 13th int. Sunflower Conf., Pisa pp. 434-440.
Williams, J.R., Jones, C.A., Kiniry, J.R. and Spanel, D.A. 1989. The EPIC crop growth model. Trans. ASAE 32: 497-511.
