SIMONTO-Pea: Phenological Models to Predict Crop Growth Stages in BBCH of Grain and Green Peas (Pisum sativum) for Temporal Pest Management

Abstract

Many pests damage pea crops, which potentially leads to reduced quality and yield losses. Since pests occur at different phenological growth stages of pea crops, the prediction of growth stages, for example as BBCH stages, is beneficial. In this study, three models have been developed to simulate growth stages of grain and green pea crops, for the latter with early and late sowing dates. All data, such as BBCH stages and air temperature, were collected in Germany in a three-year study under practical farming conditions at 415 sample sites. For the development of each model, a Gompertz regression model based on the observed data was performed. The model validation suggests that each model precisely and reliably predicts pea crop growth stages for spring-sown peas. Amongst others, the RMSE_Index for grain peas was 3.4; for green peas, early and late sowing dates, respectively, they were 3.4 and 4.5. SIMONTO-Pea (SIMulation of ONTOgenesis) is the first model that predicts detailed pea crop growth stages based on the BBCH scale. This innovation is especially beneficial for users such as advisors and farmers dealing with spring-sown pea crops as a decision support system in monitoring and pest management according to pea crop growth stages.

1. Introduction

Since 2015, the Federal Republic of Germany financially supports the cultivation of legumes as ecological focus areas, including peas (Pisum sativum L. (Fabaceae: Faboideae)), to increase the local production of legumes as ecological valuable and high-protein crops [1]. Peas can be used as such a crop and, at the same time, they are a very valuable preceding crop [2]. As a result, in Germany grain peas acreage has doubled from 37,900 ha in 2013 to 87,600 ha in 2016 [3], which led to an increase in pest pressure in intensive growing areas (own study data).

According to Ritchie and Nesmith [4], there is a difference between growth and development in plants. Growth is related to the gain of, for example, biomass or parts of plants in measurable units like length and weight. Development is related to determined phenological stages from germination to maturation. Nevertheless, the term “growth stage” is often used and will be used in this paper to describe phenological stages of plants, referring to Zadoks et al. [5] and Meier et al. [6].

In the past, different numerical scales have been developed to describe phenological growth stages of different monocotyledonous and dicotyledonous crops [5,7]. Later, in Germany, BBCH scales as plant growth stages for several crops were developed based on Zadoks et al. [5], amongst others, for peas [5,6,8,9,10]. BBCH scales consist of numerical codes as detailed phenological growth stages. There are principal stages, such as leaf development and flowering, and in addition, there are secondary stages, which describe the principal stages in detail, like the number of leaves or the percentage of open flowers. Using these BBCH-based phenological growth stages, simulation models of crop phenology for several pest species, their monitoring and their management during the vegetation period are widely applicable.

The timing of plant infestation by different pest species, insects and pathogens, often coincides with a particular crop growth stage. For example, adult pea leaf weevils (Sitona lineatus) prefer to feed on newly emerged young leaves (BBCH 9–19). Their larvae feed on the pea’s nitrogen-fixing root nodules, resulting in lower nitrogen uptake and increased susceptibility to fungal diseases due to wounding [11,12]. The first generation of pea gall midges (Contarinia pisi) harms the buds (BBCH 51–59), resulting in reduced flower development and yield loss [13]. Adult pea moths (Cydia nigricana) enter pea fields attracted by inflorescence and flower odors (BBCH 51–69, [14,15,16]; adult pea weevils (Bruchus pisorum) enter pea fields at flowering time (BBCH 60–69). Their larvae and those of pea moths feed on seeds inside pods (BBCH 71–89), resulting in severe yield and quality loss [15,17].

In Germany, there are well-established phenological crop growth models using BBCH scales, e.g., for wheat by Johnen et al. [18] and for oilseed rape by Böttcher et al. [19]. There are some models SIMulating the ONTOgenesis (SIMONTO), for example, winter cereals [20], which predict BBCH stages and combine them with different models for infection risk prediction of leaf diseases based on weather data (SIG-Getreide = pest infection risk for cereals; [21]). Additionally, a winter oilseed rape SIMONTO is available combining BBCH scales with the prediction of Sclerotinia sclerotiorum [18,20]. Moreover, Racca and Tschöpe [22] developed a BBCH-based crop growth model for blue lupine (Lupinus angustifolius), to regulate anthracnose (Colletotrichum lupine). However, a similar detailed crop growth model based on BBCH scale is missing for peas.

Already existing simulation models predict growth periods of pea plants such as emergence, flowering or maturity using growing degree days, without discriminating the single substages of leaf development, flowering and maturity, respectively [23,24,25,26,27]. These models are mainly based on temperature and growing degree days, respectively.

Nowadays, there are additional techniques available via earth observation [28]. With high-frequency earth observation satellites that fly over an area of interest, it is possible to obtain enough cloud-free images to create timelines. Consequently, a correlation between the images and ground truth data of phenological stages can be calculated. Subsequently, phenophases of cultures can be simulated. Some studies have already tried this, here, regardless of the satellite and analysis systems. For example, Dingle Robertson et al. [28] detected phenophases for canola, lentils, peas and wheat, of which the flowering phase was especially recognizable in lentils and peas. Gao and Zhang [29] studied crop phenology using satellite images of corn and soybean. Both studies showed rather coarse phenophases, which are not as precise as BBCH stages. Nonetheless, this might improve in the future and might become an interesting approach.

In this study, we present SIMONTO-Pea, a pea crop growth model for grain as well as green peas. SIMONTO-pea is part of the decision support system CYDNIGPRO (CYDia NIGricana PROgnosis), which focuses on the prevention and reduction of pea crop damage caused by pea moths and is developed by the Central Institute for Decision Support Systems in Crop Protection (German acronym ZEPP).

2. Materials and Methods

2.1. Data Sets

2.1.1. Sample Sites

Field data, for the development of the pea growth models, were collected in the years 2016–2018 in three representative German pea growing regions, which were located in the federal states Saxony-Anhalt (ST), Saxony (SN) and Hesse (HE). In Saxony-Anhalt, north-east of the Harz Mountains, in the proximity of Quedlinburg (51.93°–51.74° N, 10.98°–11.29° E), the cultivation of grain peas has a long history, especially for seed production. The “Lommatzscher Pflege” (51.40°–51.18° N, 13.20°–13.53° E) in Saxony is a traditional green pea growing area, a major one for frozen green pea products in Germany. The cultivation of peas in Hesse (51.38°–51.14° N, 9.75°–10.12° E) is not as intense and commercial as in Saxony-Anhalt and Saxony; the pea sites are much smaller and most of the peas are forage crops. Additionally, some data were collected in Rhineland-Palatinate (RP; 49.85°–49.70° N, 7.85°–8.12° E), providing a contrasting climatic region of a wine growing area. The sample sites were cultivated by regional farmers or agricultural cooperatives. Field management such as fertilization and pesticide application was according to the farmers’ practices at each sample site.

2.1.2. Grain Pea and Green Pea Data Sets

Testing several groups with different sowing date periods and production types resulted in three models: one for grain peas, one for green peas with early sowing dates and one for green peas with late sowing dates. The difference between a model for grain peas with early and late sowing dates, respectively, was too small to necessitate two distinct models. Accordingly, data sets of green peas were grouped in early and late sowing periods. Early sowing dates were defined as 1 March to 15 April, whereas late sowing dates were defined as 16 April to 31 May. The data sets for each model were divided into two data blocks: one for model development (approx. 70%) and one for validation (approx. 30%;

Table 1. The number of used data sets (equals sample sites) and observations, respectively. The data sets were divided into two blocks: one for model development and one for validation.

	Data Sets (Sites)	Observations	Model Development	Validation
Grain peas	225	2518	1848	670
Green peas, early	141	847	644	203
Green peas, late	49	312	239	73
Sum	415	3677	2731	946

). Each data set comprised all observations per sample site and vegetation period, with an average of eleven observations for grain peas and six observations for green peas per site. Altogether, 2731 observations were used to develop and 946 observations were used to validate the three models. Insufficient data sets were excluded to develop precise pea growth models. Most green pea data sets were from Saxony, whereas grain pea data sets were mostly derived from Hesse, Rhineland-Palatinate and Saxony-Anhalt.

2.2. Data Acquisition

Plant growth and development mainly depends on nutrients, carbon dioxide, water, temperature and light to perform photosynthesis. However, the study is focused on temperature and photoperiod to create pea growth models for practical and area-wide application. Therefore, it is necessary to access easily available input data for practical use. The following parameters were assessed for the study:

• The sowing date (BBCH 0) was chosen as the starting point for the simulation of the pea growth stages. Sowing dates were provided by the farmers.
• Phenological growth stages using the BBCH scale of peas [6] were recorded on a weekly basis. To develop practically applicable models, the data sets for grain peas had the maximum BBCH 89, whereas for green peas the maximum was BBCH 79, because green peas are usually harvested before.
• The hourly air temperature [°C] data were interpolated data from weather stations located nearby the sample sites [30,31]. The weather data are provided by weather stations from the German Weather Service and the federal states of Germany. The interpolations are calculated for each km² throughout Germany.
• The geographical coordinates of a given sample site were used to calculate the photoperiod for the daily possible development rate. They were linked to the Solar Calculator [32] to precisely calculate the site-specific photoperiod.

2.3. Model Development

2.3.1. Development Rate According to Air Temperature and Photoperiod

The daily development rate (DRd), depending on air temperature and photoperiod, was calculated as follows:

DRd = DR(T) × DR(P)

(1)

where:

DRd	=	Daily plant development rate;
DR(T)	=	Development rate as a function of air temperature;
DR(P)	=	Development rate as a function of relative photoperiod.

For the calculation of the development rate, a function was developed using air temperature and photoperiod (Equation (2)). This equation describes the relative plant development rate as a function of photosynthesis depending on air temperature and photoperiod (Equation (1), Figure 1). The data for the parameterization of the function were obtained from the available literature [33,34,35,36]. The crop development rate, depending on air temperature and photoperiod, is based on a Beta-Hau function [22,37] which is described as follows:

$$\mathrm{DR}\left(\mathrm{T},\mathrm{P}\right){=\mathrm{DR}}_{\mathrm{opt}}\left[\frac{\left({\mathrm{Y} - \mathrm{Y}}_{\min }\right)}{{\left({\mathrm{Y} - \mathrm{Y}}_{\min }\right)}^{\mathrm{n}\left(\mathrm{Yopt} - \mathrm{Ymin}\right)/\left(\mathrm{Ymax} - \mathrm{Yopt}\right)}}\right]{\left[\left({\mathrm{Y}}_{\max }- \mathrm{Y}/\left({\mathrm{Y}}_{\max }-{\mathrm{Y}}_{\mathrm{opt}}\right)\right)\right]}^{\mathrm{n}}$$

(2)

where

DR(T,P)	=	Development rate as a function of air temperature (DR(T)) and relative photoperiod (DR(P));
DRopt	=	Optimum development rate;
Y	=	Air temperature or relative photoperiod;
Ymin	=	Minimum air temperature or rel. photoperiod for plant development;
Yopt	=	Optimum air temperature or rel. photoperiod for plant development;
Ymax	=	Maximum air temperature or rel. photoperiod for plant development;
n	=	Equation parameter.

The crop growth stage Equation (3) was modeled as a function of the cumulated daily development rate which was calculated by summing up the daily development rate starting at the sowing date:

BBCH x = (f)∑DRd

(3)

where

BBCH x	=	Crop growth stage;
∑DRd	=	Cumulated daily development rate from sowing date to BBCH x.

2.3.2. BBCH and Index Stages

In this study, the BBCH stages were decoded into continuous index stages to remove the gaps that are prevalent in the BBCH scale of pea. For example, there are no principal pea growth stages 2 and 4 and, within the principal growth stage 5, there are only three secondary growth stages (

Table 2. Index stages corresponding to secondary BBCH stages; the sections marked with lines separate the principal growth stages.

BBCH	Index	BBCH	Index	BBCH	Index	BBCH	Index
0	0	18	15	60	30	76	45
1	1	19	16	61	31	77	46
3	2	30	17	62	32	78	47
5	3	31	18	63	33	79	48
7	4	32	19	64	34	81	49
8	5	33	20	65	35	82	50
9	6	34	21	66	36	83	51
10	7	35	22	67	37	84	52
11	8	36	23	68	38	85	53
12	9	37	24	69	39	86	54
13	10	38	25	71	40	87	55
14	11	39	26	72	41	88	56
15	12	51	27	73	42	89	57
16	13	55	28	74	43
17	14	59	29	75	44

). Therefore, e.g., the index number 57 (Table 2) corresponds to BBCH 89 in this study. This procedure was necessary, especially for the validation of the models. A validation based on observed and calculated/retransformed BBCH stages would corrupt validation results due to the artificial gaps of the pea BBCH scale. In accordance with [9], we added BBCH stages 66 and 68 to the BBCH scale for obtaining a more precise description of the florescence. BBCH stages 97 and 99 are not important for the model and were omitted. The key advantage of the transformation system is that our crop growth models predict existing BBCH stages of peas only.

2.3.3. Gompertz Regression

After several tests with different types of non-linear regressions, a Gompertz function [38] was chosen for modeling index stages derived from field observations and cumulated DR_d from the sowing date onwards. The maximum index stage for grain and green peas was 57 and 48, respectively. The result of the Gompertz regression is a function of the following type with the fitted parameters a and b according to the respective data set.

$$\mathrm{predicted} \mathrm{Index}=\left[\exp \left(-\exp \left(-\left(-\mathrm{a}+\mathrm{b} \times \sum {\mathrm{DR}}_{\mathrm{d}}\right)\right)\right)\right]{\times \mathrm{Index}}_{\max }$$

(4)

where

a	=	Y-axis intercept;
b	=	Slope;
∑DRd	=	Cumulated daily development rate from the sowing date to date x;
Indexmax	=	57 and 48 for grain and green peas, respectively.

Using Equation (4), index stages can now be calculated and subsequently transformed back into BBCH stages according to Table 2.

2.4. Validation

In order to evaluate the models’ quality and reliability, they were validated with independent data sets (Table 1) derived from 2016 to 2018. For the validation, observed index stages were compared to predicted index stages. Furthermore, logistic regressions have an asymptotic end of the curve; therefore, the maximum value cannot be reached. Thus, the validation of the grain pea model must stop at Index 54 and green pea models at Index 45. Since Index 0 is the start of the calculation, it was also not validated. All statistical calculations were computed with XLSTAT 2017.4 by Addinsoft. The following three methods were used for validation.

2.4.1. Linear Regression

The goodness of the models, which means the match between the observed and the predicted data, was tested by using linear regressions.

2.4.2. Praxis Validation

The predicted BBCH stages were compared to the observed BBCH stages by counting the difference in days. When the deviation was less or equal ± 7 days, the predicted BBCH stages were counted as correct, because the recording took place once a week. Hence, deviation < 7 days were rated as too early and >7 days were rated as too late.

2.4.3. Root-Mean-Square Error

The root-mean-square error (RMSE, Equation (5)) is used to describe the difference between the predicted and the observed data. RMSE was calculated in terms of index stages and dates in the form of days. The equation for RMSE is as follows:

$$\mathrm{RMSE}=\sqrt{\frac{\sum {\left({\mathrm{y}\prime }_{\mathrm{i}}{ - \mathrm{y}}_{\mathrm{i}}\right)}^2}{\mathrm{n}}}$$

(5)

where

y′i	=	Predicted index stages or date;
yi	=	Observed index stages or date;
n	=	Number of observations.

3. Results

3.1. Model Development

3.1.1. Index Stages

As described in Section 2.3.3, Gompertz regressions were used to develop the three models. Figure 2a–c displays the outputs of the three different pea model curves and observations. The model parameters, including goodness of fit parameters, for each of the three models are shown in

Table 3. Parameters of the Gompertz regression for each model, with significance for a (intercept) and b (slope) as *** p ≤ 0.001; R²_N = pseudo-R² by Nagelkerke, Pr Chi² = Probability of Chi².

	a	b	SD	R2N	p	Pr Chi2	n
Grain peas	−1.28 ***	0.14 ***	0.001	0.47	<0.0001	<0.0001	1848
Green peas, early	−1.46 ***	0.22 ***	0.002	0.55	<0.0001	<0.0001	644
Green peas, late	−1.44 ***	0.19 ***	0.003	0.48	<0.0001	<0.0001	239

. In particular, the values of pseudo-R²_Nagelkerke around 0.5 are really meaningful.

Figure 3 illustrates that, especially around flowering (Index 30 ~ BBCH 60), the model curves of green peas with early and late sowing dates are apart from each other, whereas at the beginning and end of the development phase they are nearby each other. This suggests that around flowering the development of early and late sown green peas, respectively, differ most. The curves for grain pea and green pea late have approximately the same run until Index 35 which is equivalent to BBCH 65.

3.1.2. BBCH Stages

The index stages were transformed back to BBCH stages as listed in Table 2. The gaps shown in Figure 4a,b represent the non-existent BBCH stages in the BBCH scale for peas.

3.2. Validation

The goodness of the models was tested with independent data sets using three different validation methods.

3.2.1. Linear Regression

Table 4. Statistical parameters of the linear regressions with observed and predicted index stages. Significance for intercept and slope: * p ≤ 0.05; *** p ≤ 0.001; n.s. = not significant (p > 0.05), p = significance of variance analysis.

	Intercept	Slope	p	R2	n
Grain peas	−0.71 *	1.03 ***	<0.0001	0.95	670
Green peas, early	0.09 n.s.	1.02 ***	<0.0001	0.94	203
Green peas, late	1.89 n.s.	0.96 ***	<0.0001	0.86	73

shows the linear regression parameters of observed and predicted index stages. In particular, the high coefficient of determination (R²) of each model indicates a high agreement of the observed and the predicted data. The R² of grain peas and green peas, early sowing, is higher than the R² of green peas, late sowing. Thus, the validation data set of green peas, late sowing, is comparably smaller (Table 3). In Figure 5a–c, index stage differences between observations and predictions of the validation data set are shown.

3.2.2. Praxis Validation

The results of the praxis validation, which is for farmers and advisors the most important validation, shows differences between observed and predicted index stages in days in the categories “too early”, “correct” and “too late” (

Table 5. Results of the praxis validation and a comparison of observed and predicted index stages in days; the correct predictions are between ±7 days.

	Too Early	Correct	Too Late	n
Grain peas	9.1%	77.6%	13.3%	670
Green peas, early	4.4%	85.7%	9.9%	203
Green peas, late	11.0%	82.2%	6.8%	73

). However, at least 77.6% of predicted growth stages are correct. The high percentage (13.3%) of grain peas predictions that are too late are composed of 52% index stages 6–16 (BBCH 9–19) and 27% index stages 49–54 (BBCH 81–86). These are the stages of leaf development and seed maturing. When focused on index stages 27–44 (BBCH-stages 51–75) only, the predictions are even more precise and reach up to 92.4% of correct predictions (data excerpt not shown). The boxplot shows the deviation between the observed and predicted index stages of Figure 6. The median is 0 for all three models and the mean is −1 for grain peas, −0.6 and −0.3 for green peas, early and late, respectively. All boxes are in the correct zone of ±7 days.

3.2.3. Root-Mean-Square Error

The RMSE validation method investigates if there are major deviations between observed validation data and the model outputs. Validation results for the RMSE are presented in

Table 6. RMSE shows differences between observed and predicted values as index stages and days, respectively.

	RMSEIndex	RMSEdays	n
Grain peas	3.4	6.7	670
Green peas, early	3.4	5.3	203
Green peas, late	4.5	6.4	73

, with RMSE values being low.

4. Discussion

In this study, three crop growth models were developed for grain and green peas, the latter with early and late sowing dates. They predict BBCH stages throughout the entire vegetation period from sowing to development of fruit and maturity, respectively (green peas and grain peas), as a function of temperature and photoperiod. The data for model development as well as for model validation were collected from 2016 to 2018 at sample sites in Hesse, Saxony, Saxony-Anhalt and Rhineland-Palatinate in Germany. In Germany, it is common to use BBCH stages [8] to describe crop growth stages. Accordingly, the model output should be BBCH stages for advisors and farmers.

4.1. Model Development

Pre-calculations revealed that the decoding of non-continuous BBCH stages into continuous index stages resulted in more appropriate predictions and, especially, validations. Using the praxis validation, the prediction of correct BBCH stages was improved up to 12%. Therefore, we decided to use index stages for model development. Subsequently, the statistical validation methods were calculated using index stages as well to avoid the gaps in the BBCH scale when comparing observed and predicted crop growth stages. Furthermore, it made sense to develop two different models for green peas considering early and late sowing dates, respectively, because around flowering time especially the two models differed most. Since pea moth adults are attracted by pea flower odors [14,16], it is particularly important for pea moth monitoring to predict bud development and flowering time as precisely as possible.

For model development, a Gompertz regression was calculated, because it had the best fit of the curves according to the input data. The pseudo-R²_Nagelkerke between 0.47 and 0.55 indicate that the quality of the three models is good. Additionally, the probability value of chi² of all three models is <0.0001 and indicates that significant information has been introduced by the variables.

4.2. Validation

The results of the three different validation methods, linear regression, praxis validation and RMSE, indicate that the predictions of the crop growth models are precise. Coefficients of determination (R²) for the index stages of linear regressions indicate that 86–95% (Table 4) of the predicted stages are explained by the observed growth stages. For all three models, the slopes are between 0.96 and 1.03 and highly significant (p < 0.0001). The intercept is significant only for grain peas; for green peas, the intercepts are not significant. The results of the variance analysis of all three models are highly significant with p-values < 0.0001. In summary, the grain pea model has the best validation results; nevertheless, the green pea models are of good quality as well.

Moreover, the results of the praxis validation of the three pea growth models (Table 5) show that the correctly predicted index stages are between 77.6 and 85.7%, which is quite precise. In particular, the predicted crop growth stages between Index 27 and 44 (BBCH 51–75) have an accuracy of up to 92.4%, which are important stages, for example, for pod infestation by pea moths. The reason for the high percentage of too late predicted index stages of the grain pea model might be the asymptotic start and end of the curve. The green pea early sowing model came off best when using the praxis validation method. The higher values that are predicted too late are not that important for infestation of pea moths or pea weevils, because most of them are stages of leaf development and seed maturation.

The RMSE values for index stages and days have small ranges, which means that all three crop growth models perform very well. When calculated with index stages, RMSE_Index reaches values between 3.4 and 4.5 (Table 6), which indicates that the observed and predicted index stages almost match each other. When calculated with days, the results range from 5.3 to 6.7. Compared to other crop growth models that are based on BBCH scales, the RMSE in days ranged, for example, from 3.23 to 6.44 in the BRASNAP-PH model for winter oil seed rape by [19]. Another example is the modified CERES-Wheat model by Johnen et al. [18] with a RMSE_days between 6.6 and 17.4, which means that our models perform as precise or even more precise compared to them. The green pea early sowing model reaches the best values in RMSE_Index as well as in RMSE_days.

In summary, the green pea early sowing model has the best results, whereas the validation results for the other two models are also precise. Consequently, the results of all three validation methods have an acceptable reliability level and can be used for spring-sown pea crops under practical farming conditions in Germany and presumably in regions with similar climatic conditions.

4.3. General Discussion

In the past, numerous pea crop growth models have been developed. Most of these models only predict important growth stages, such as flowering or maturation, or periods, such as sowing to flowering. For example, Ney and Turc [24] developed a pea growth model which mainly focused on the progression rate of flowering and initiation of seed filling, which was based on cumulated degree days. Bourgeois et al. [23] also used cumulated degree days and focused on crop growth periods such as sowing to emergence, emergence to flowering and flowering to maturity. Summerfield and Roberts [39] primarily centered their model on the prediction of flowering based on vernalization, temperature and photoperiod. Furthermore, Roche et al. [40] even compared different models from other authors that predict the beginning of flowering. Likewise, Biarnès et al. [26] used already existing models for each determined stage; for example, the one of Summerfield and Roberts [39] to predict the beginning of flowering of pea crops. Nevertheless, none of these studies is based on a detailed pea crop growth stage system like the BBCH scale, as SIMONTO-Pea is. By having these detailed crop growth stages, it is possible to connect, for example, one single stage with a pest management strategy.

However, the presented study might have some limitations. The sample sites were pea fields of regional farmers and agricultural cooperatives. Thus, they decided on the pea variety and the sowing date, respectively, for each sample site. Although, this represents real practical farming conditions, this procedure resulted in an uneven distribution of pea varieties and sowing dates within the three developed crop growth models. Thereby, it was not possible to develop cultivar-specific models because the numbers of data sets per cultivar were not sufficient. Most of the peas were sown early in the season, which means that most data sets used in this study start with an early sowing date. In general, the three models might become better with increased data input. Consequently, if the results became better, the models could be adjusted. In our data, there might be bias due to different people who recorded BBCH stages at the sample sites. This can cause errors in the model development and, consequently, errors in the validation of the BBCH stages. Kirby et al. [41] had similar problems in their study. In our case, the observation of BBCH stages 30–59 and 67–72 was not as precise as other stages, since peas develop different stages simultaneously. For example, when the last flowers were blooming, the first pods were filled with already big seeds at the same time. The model is not suited for irrigation, fertilization or yield simulation. It can be used for crop protection only.

In summary, the precise crop growth models, namely, SIMONTO-Pea, of the decision support system CYDNIGPRO are available on www.isip.de (accessed on 11 November 2023) [42], a plant protection service provider for advisors and farmers in Germany, with a user-friendly input mask. Site-specific pea crop growth stage predictions are possible for spring-sown grain peas and green peas, for the latter with early and late sowing dates. In addition, there is a decision support system, which predicts emergence time of the first instars of pea moth larvae for temporal pest management. The SIMONTO-Pea is a background calculation to connect the pea ontogenesis with pea moth population dynamics. Pea moth adults are lured by pea flower odor. The eggs are laid on the just-growing pods or leaves, so as soon as larvae hatch, they are close to the pods. The time period to spray is short, because it must be before they enter the pods. As soon as they are inside the pods, they are protected against contact insecticides. Furthermore, the models can be useful for studies on climate change, like other crop growth models which were already used to simulate the future crop growth of winter wheat, sugar beet and winter oilseed rape in the KLIFF project [43]. For the future, there is the possibility to develop another model for fall-sown peas, as soon as more data are available. At the time of publication, SIMONTO-Pea is available throughout Germany and can be calculated for each km², since it is based on Germany’s interpolated weather. If weather data are available, it could be extended to other countries as well.

5. Conclusions

In this paper, we presented the development of three pea crop growth models for grain peas and green peas, for the latter with early and late sowing dates, respectively. They are based on the BBCH scale. The validation methods suggest that these pea crop growth models have a sufficiently high level with up to 85.7% of precision and reliability; consequently, they can be used for spring-sown pea crops under practical farming conditions. The models predict the complete continuous growth of pea crops in detail. The presented pea crop growth models are effective applications for better temporal monitoring and pest management. They predict the phenology of peas with high precision for different German regions and consequently in other regions with similar climatic conditions. For the calculation of the models, easily available input data are needed; therefore, an area-wide application is possible.