Poluektov and Topaj: Present and Future of Crop Modeling

T appearance of dynamic models in agroecology has led to a new understanding of the processes During the last two decades, computer simulation models have become powerful tools for investigating agricultural crop dynamics taking place in the soil–plant–atmosphere system and and solving practical problems. Many models have been developed to formation of so-called dynamic thinking. The Russian in various countries, which permits exploration of the influence of school of crop simulation recognizes Monsi and Saeki weather conditions and agricultural strategies on the fate of a crop. (1953) as the originators of this new branch of science However, some fundamental problems related to the description of (this may deviate from the traditional Western point of agricultural plant growth and development remain unsolved. These view) (Sinclair and Seligman, 1996). Monsi and Saeki’s primarily concern the totality of biological processes such as ontogeapproach was further developed by many authors in the netic development and morphogenesis due partly to the lack of knowlformer Soviet Union and, in particular, by a group of edge in plant physiology and the absence of realistic ideas about the scientists at the Agrophysical Research Institute in St. origin of plant life. These circumstances have forced modelers to use Petersburg. Unfortunately, contacts between eastern quite sophisticated heuristic approaches rather than biologically sound European and Western scientists were rare for a long descriptions. This paper represents the authors’ vision of this situation. time, and agroecological simulation developed along separate courses. Agrophysical Res. Inst., Russian Acad. of Agric. Sci., 14 Grazhdansky Budagovsky and Ross (1966) proposed a theoretical prosp., 195220, St. Petersburg, Russia. Received 24 Oct. 2000. *Correapproach to the quantitative description of crop photosponding author ([email protected]) synthetic activity. It was probably the first publication in Russian concerned with crop simulation. Later, BikPublished in Agron. J. 93:653–659 (2001). 654 AGRONOMY JOURNAL, VOL. 93, MAY–JUNE 2001 hele et al. (1980) developed the model describing the Models based on an empirical approach can be considered as a set of heuristic equations describing crop processes of photosynthesis and transpiration under soil water deficit conditions. Sirotenko and Boyko (1985) growth and development. Each of these equations is usually a static description of a relation between the rate created a complex set of differential equations for the simulation of energy and mass transfer in a crop. Our of the considered process and environmental conditions. Input parameters for these equations must be identified efforts (Poluektov et al., 1979; Poluektov, 1991; Poluektov and Vasilenko, 1993; Poluekotov and Topaj, 1996; using standard or specially planned field experiments. Insertion of these equations into simple dynamic algoPoluektov and Zakharova, 2000) were directed toward the development of theoretical as well as applied modrithms yields an empirical model. It is easy to see the shortcomings of such a simple approach. Firstly, such els. Various complex problems including such specific tasks as wheat (Triticum aestivum L.) wintering or acmodels represent a return to the concept of regression analysis although on a new qualitative level. Complicacount of soil moisture excess were addressed. As a result, a family of models was developed starting from tions arise if one increases the number of defining relations, resulting in additional difficulties for parameter simple constructions and finishing with very complex and detailed structures. Now we have a set of models identification. An empirical model is not versatile, and in reality, adapted to several crops [winter and spring wheat, barley (Hordeum vulgare L.), maize (Zea mays L.), alfalfa can require too much time to identify model parameters for each specific set of crop, soil, and environmental (Medicago sativa L.), and others] grown in a number of different regions of Russia (Krasnodar, Saratov, Altaj, conditions. In many cases, we can obtain excellent correspondence between measured and simulated data. HowLeningrad, and Kaliningrad). It is possible to assert that these and other existing models together constitute the ever, one will never be certain that the developed model will be useful for the description of a different crop exbase of crop simulation knowledge, and it should seem that all of the principal problems in this field (especially posed to different soil and weather (Fig. 1). When these conditions change, the parameter values (this problem since the appearance of modern powerful personal computers) are finally solved. But . . . can be solved by reidentification) as well as the qualitative type of equations may be wrong. The most impressive argument for the insufficiency SCIENCE VERSUS UTILITY of curve fitting methodology in crop modeling is the IN CROP MODELING comparison of results from different models applied to Computer crop modeling is now a power industry in the same data set. Such comparisons have taken place itself with its own tasks, methods, and fields of applicawithin the framework of various workshops or projects tion. So, it is useful to turn back and sum up the results but almost always with the same consequence (Dieckfrom more than 40 yr of history. A careful observer kruger et al., 1995; Poluektov et al., 1999). Divergence may notice that there are two principal simulation phiin the results of production process simulation (for its losophies corresponding to alternative approaches of various components, e.g., crop yield, soil water content, algorithmic representation of the physical, chemical, and phenology) is very high, sometimes reaching several and biological processes taking place in real agricultural hundred percent! So, which models can one trust? ecosystems. The first approach is often called theoretical Any attempt to extend the scope of an empirical and the second one empirical, but in other works, one model beyond the events or conditions for which it can find the terms mechanistic and functional or biowas developed and tested is not simulation but rather physical and heuristic. What is the main difference bespeculation. Therefore, the empirical approach to crop tween the two methodologies in crop simulation? simulation cannot be used with confidence as a method of scientific investigation. A common example of its inapplicability is the currently popular modeling task connected with studies of the possible influence of global climate change on the ecological stability of agricultural systems. Probably the lone essential advantage of the empirical approach is that these models are available and can be successfully used for decision making in agriculture. As will be shown below, the theoretical models do not have this important property. Theoretical approach means an honest description of crop and environmental dynamics and entails development of a mathematical model according to the physical, chemical, or biological principles underlying all of the processes included. A pure theoretical model consists of physically interpreted relations (unlike the logically interpreted ones in the empirical models). As a rule, they are the differential equations of mathematical physics, which follow from the consideration of energy Fig. 1. Validity of empirical approach in agroecosystem simulation. and matter balance for selected spatial or functional POLUEKTOV AND TOPAJ: PRESENT AND FUTURE OF CROP MODELING 655 compartments. Certainly, such a model could be used lar space from the atmosphere, chlorophyll excitement upon light absorption, and dynamics of biochemical reas a tool for scientific research. Its algorithmic content actions in the Calvin cycle are formulated as a set of is not connected with the conditions of its adjustment differential equations. The model of leaf photosynthesis and validation, and one can be sure that the laws of is extrapolated to the crop scale to produce a theoretical nature are more universal than human fantasies. There model of crop photosynthesis where each macroparamis only one difficulty (but it is global) in applying the eter has a concrete physical meaning. theoretical approach to mathematical simulation of agroComparison of model run calculations using the same ecosystems. The honest description of all of the proinput data clearly demonstrates the advantages and discesses included in the model, and especially their inteadvantages of each approach. Note that there were no gration into the complex scheme with the same level specific calibrations of either model before comparison. of accuracy, is an extremely difficult problem. Some There are some conditions where the theoretical model phenomena (mainly of a biological nature) have not produces results that are quantitatively and qualitatively yet been studied in sufficient detail. Theoretical models similar to the empirical model (Fig. 2). It is more interrequire developers to be skilled specialists in various esting, however, to consider the circumstances where branches of science. It makes the development of the they vary. Some of these cases suggest an advantage to mechanistic model so difficult that there still is no comthe empirical approach (Fig. 3). For instance, results of plex agroecosystem model that can truly be called theothe AGROSIM model show the well-known unimodal retical. Probably the main success has been achieved in dependency of photosynthesis rate on phototemperacreating single units or submodels of separate processes ture while the theoretical model does not take this feain the soil–plant–atmosphere system. Their integration ture into account (Fig. 3A). We can review our theoretiinto a comprehensive model now seems to be a utocal model to correct this mistake, but we can never be pian dream. certain to avoid another absurd result. However, the Let us compare the two approaches, using as an examtheoretical model has an important advantage—it can ple the photosynthetic units (blocks) of corresponding give new, unexpected, and scientifically valuable results models. For the empirical case, the winter wheat model as shown in Fig. 4 where the joint influence of two AGROSIM-WW [AGROecosystem SIMulation—winter factors on photosynthesis rate has been investigated: wheat (WW)] has been chosen. It was developed by atmospheric CO2 concentration and water stress. For specialists at the Institute of Landscape Modeling (Muenthe empirical model, one can see that the C curves of cheberg, Germany) and is a part of the AGROSIM model photosynthesis are qualitatively similar under various family (Wenkel and Mirschel, 1995). This model follows levels of water availability (Fig. 4B). The results are the typical pattern of empirical methodology. The main different with the theoretical model. Under unstressed relation for calculating daily photosynthetic rate is given conditions, CO2 concentration in the atmosphere is by the following multiplicative equation of partial stress nearly at the saturation point and further increase has functions: little effect on the photosynthesis rate. However, under FD 5 FM 3 B 3 f1(QP) 3 f2(TP) 3 f3(BM) 3 f4(WS) conditions of strong drought, the dependency of photosynthesis on CO2 concentration is practically linear. This 3 f5(WL) 3 f6(NF) 3 f7(CO2) 3 f8(DL) result is correct; it has been confirmed by experimental where FD is actual daily photosynthesis rate (kg ha21 results from greenhouse experiments. This mechanism d21 ), FM is biological maximum of assimilation per unit was not explicitly included in the model during its develof green leaf biomass under optimal conditions, and B opment, so we have used the model as a tool of scientific is total green biomass. The f functions are partial stress research and gained new knowledge as a result. functions describing the relative decrease in primary During the past 30 yr, and especially for the past few assimilation under nonoptimal values of each of the years, we have had to maneuver between scientific and following factors: QP, incoming solar irradiance; TP, utilitarian thinking in crop simulation. Indeed, progress phototemperature; BM, total biomass; WS, current soil in the development of modern hardware has removed water contents; WL, past soil water contents; NF, N many of the previous restrictions concerning computer content; CO2, atmospheric CO2 concentration; and DL, resources. This provides a basis for improving empirical daylength. Each of these stress functions is a purely models and closing the gap between empirical and theoempirical dependency with the parameters identified retical approaches. Some efforts have been made at the from field tests. So, the main relation embodied in the Agrophysical Research Institute to fulfill this promise. equation can be easily interpreted but has no physical As an example, two proposed methods for describing basis. physiological processes in the plant canopy are preAs an alternative, the model developed at the Agrosented below. physical Research Institute for calculating daily photoThe first is a new method for simulating actual values synthesis rate (Poluektov, 1991) has been selected. It of plant transpiration and soil evaporation (Poluektov et purports to meet the requirements of the theoretical al., 1997). The well-known Penman–Monteith approach approach. The algorithms used in the model consider was used as a basis for our method (Penman, 1948; the main physical and biochemical phenomena conMonteith, 1981). However, our approach differs in two nected with photosynthesis and gas exchange in green respects. Only the radiation absorbed by phytoelements is included in the heat balance equation, and the depenleaves. The processes of CO2 diffusion to the intercellu656 AGRONOMY JOURNAL, VOL. 93, MAY–JUNE 2001 Fig. 2. Function of daily photosynthesis rate vs. global radiation and CO2 concentration for (A ) a theoretical model and (B ) an empirical model. dence of stomatal resistance on leaf water potential is soil water content exceeds field capacity. A similar approach has been proposed for the calculation of evapoused to calculate the actual value of plant transpiration as influenced by weather conditions and leaf water poration from the upper soil layers. The second method is a new algorithm for the simulatential. The method accounts for the physical processes in the soil and atmosphere as well as for the physiologition of dry matter distribution between shoot and roots. It can be called an adaptive distribution key. A fixed cal characteristics of water transport in plants. It takes into account the water deficit and its impact on crop distribution key is usually used in models of annual crops (Penning de Vries et al., 1989). It does not allow dynamics. In addition, a new technique has been proposed for description of the opposite situation—the indescription of plant reactions to environmental conditions such as soil water and N regimes. However, some fluence of water excess (and consequently, soil O2 stress). The main idea was to include a unit in the model processes are affected by the C and N content of the plant organs, for example, CO2 assimilation by green to describe the exhaustion of internal plant energy resources under saturated conditions, i.e., in the case when parts of the plant and N uptake by roots. Consequently, POLUEKTOV AND TOPAJ: PRESENT AND FUTURE OF CROP MODELING 657 Fig. 3. Function of daily photosynthesis rate vs. daily average temperature and CO2 concentration for (A ) a theoretical model and (B ) an empirical model. C 3 N interaction should be included in the model to N uptake by roots on assimilation of CO2 by leaves for both annual and perennial crops. Let us consider the adequately describe plant growth and development. The method that has been developed was based on the estiexample of alfalfa in the second or third year of vegetation. In this case, there is a large amount of roots and mation of a fraction of daily assimilates, which are either translocated into roots or remain in the leaf according a small amount of aboveground dry matter during vegetation renewal in spring or after a recurrent cut. The to N demand and availability (Poluektov and Zakharova, 2000). high N availability leads to the primary growth of green plant organs so that accumulated carbohydrates limit Our goal was to describe an alternative mechanism for dry matter partitioning that reflects adaptive crop plant growth. Root dry matter decreases due to respiration. When shoot dry matter reaches the compensation reactions to ambient conditions, especially the effect of 658 AGRONOMY JOURNAL, VOL. 93, MAY–JUNE 2001 Fig. 4. Function of daily photosynthesis rate vs. water stress index (stomatal resistance) and CO2 concentration for (A ) a theoretical model and (B ) an empirical model. point and excess carbohydrates are produced, a fraction reaction to environmental conditions, thus the name, adaptive distribution key. is translocated into roots so that dry matter rises again. The whole picture is clarified in Fig. 5. It demonstrates the sawtooth time course of the aboveground biomass DISCUSSION AND CONCLUSIONS and the oscillatory nature of root biomass dynamics. Summarizing our vision, we can state that, in general, Because the daily amount of assimilates produced by empirical models produce results that are usually reaphotosynthetic organs depends on the current weather conditions, this distribution key reflects adaptive plant sonable, not always correct, and never scientifically valPOLUEKTOV AND TOPAJ: PRESENT AND FUTURE OF CROP MODELING 659 the scope of application of both theoretical and empirical models. Certainly, the number of holes in the existing models is still too high to dream about the appearance of a good complex model in the near future. But we must note that success comes to the patient researcher only. It is difficult to say which of these opinions about future development is more correct. Most likely, the truth is somewhere in the middle.