1. Introduction
Pearl millet is an important crop cultivated on about 30 million hectares (ha) in more than 30 countries [
1]. In India, pearl millet is mainly cultivated under rain-fed conditions during the rainy season (kharif) and represented
of the yearly cultivated surface during the period 1998–2017. Due to its capacity to use water efficiently and its heat tolerance, pearl millet is adapted to harsh climatic conditions where other crops fail to produce economic yield [
2,
3]. However, despite constant increases across the production area, with an average of 1161 kg/ha over 1998–2017, pearl millet yield remains low compared to other crops in the same regions (e.g., rice: 1967 kg/ha; maize: 1650 kg/ha). Moreover, limited market opportunities and minimal government support are important constraints for farmers to consider pearl millet for cultivation. Thus, pearl millet is mostly used for food and fodder [
4] and it is under the constant pressure of changes in dietary patterns for crops, such as wheat or rice [
5].
The sustainable intensification of complex agronomic systems like pearl millet cultivation in India with a wide range of weather conditions (e.g., low and high precipitation), soil type (e.g., from sandy to black vertisol), and/or socio-economic usages with many possible interactions [
6], requires quantitative understanding of production environments. This quantification set the base to design context specific technologies [
7,
8,
9]. Increasingly, crop simulation modelling (CSM) is used for environmental characterization (envirotyping) and the definition of target population of environments (TPE) [
10,
11,
12,
13,
14]. A TPE is composed of environments (TPEs) with more homogeneous biophysical properties. Their identification and use for breeding help to reduce the GxE effect in selection [
15,
16]. Therefore, the TPE definition is a prerequisite for several important applications, such as the optimization of multi-location breeding trials sampling.
The initial characterization of pearl millet cultivation in India dates to 1979 when the National Agricultural Research Project divided the environment into zones A and B. Later, zone A was split into A1 and A parts [
17,
18] (
Figure 1A). This zonation mostly relies on the yearly rain pattern and is still considered today as a reference in many pearl millet breeding programs. According to this zonation, pearl millet is cultivated under low precipitations (<400 mm/year) in the north of Rajasthan classified as A1 zone, in the neighbouring regions of south Rajasthan, Haryana, Gujarat, and Uttar Pradesh termed as A zone (>400 mm/year), and central Western India referred as B zone (>400 mm/year).
Except for the work by Gupta et al. [
19], no notable revision of the TPE has been proposed recently. That work supported the absence of difference between the A1 and A zones, which was equivalent to the initial zonation. However, this analysis was based on yield data of well-managed improved varieties breeding trials. We suggest that a revision of the Indian pearl millet TPE integrating a wider source of information should better capture the complexity of this system [
6]. This revision is the main objective of the article.
CSM progressively became one of the most important tools for environmental characterization [
20,
21]. CSMs are in silico representations of natural systems captured by a series of equations [
22]. These equations express agronomic traits such as yield as a function of environmental inputs (e.g., temperature, soil water potential), management practices (e.g., irrigation, fertilization) and crop characteristics (e.g., phenology, canopy growth, biomass partitioning) [
23]. CSM reconstruct the soil–plant–atmosphere continuum of a system at the plot level. Those simulation can be scaled up across spatio-temporal scales (e.g., region) to identify TPEs with similar biophysical properties forming the TPE [
16,
24].
Recently, increases in computer power created the opportunity to run large-scale simulations to extensively explore the properties of agronomic systems (e.g., Ronanki et al. [
24]). The second objective of the article is to present an innovative strategy based on large-scale crop model simulations to support the revision of the pearl millet TPE. This strategy helped us to gather extra information about the locations where pearl millet is cultivated, as well as estimating the influence of different parameters on the yield (sensitivity analysis). This article also allowed us to evaluate a new crop model characterized by an improved algorithm to describe pearl millet tillering [
25,
26]. Since tillering is critical for pearl millet adaptation to drought-prone environments, the evaluation of this updated CSM tool in the new TPEs is an important prerequisite to design crops and management strategies for the current and future climates in the target geographies. It will also support pearl millet crop model development, which remains an under-researched area [
27,
28,
29].
4. Discussion
The main objective of this article was to characterize the Indian pearl millet production regions using novel data sources and state-of-the-art methodologies. This analysis was originally requested by the Indian national and international research institutions to provide a deeper understanding and quantitative base for the optimization of the pearl millet production systems. The revised TPE is still largely based on influential biophysical factors, such as rain pattern, temperature, and the soil characteristics, but provides several novel insights, particularly hinting to the changing climate and socio-economic drivers, which should be further investigated. In the following sections, we will first discuss the implications of important causes of changes like the possible rain pattern increase and the socio-economic aspects driving some of the TPE features. Then, we will discuss the pros and cons for the use of crop models for pearl millet envirotyping.
4.1. Kharif Rain Increase and Transition of Pearl Millet Cultivation to Summer Season
In all TPEs, the Nasapower weather data suggest that the amount of precipitation during kharif has increased between 1998 and 2017. The hypothesis of increasing kharif rains is supported by other sources and is expected to continue [
53,
54]. Other authors, like Praveen et al. [
55], used observed data from the Indian meteorological system and reached the conclusion of a general rain decrease since around 1960. The later reference is, however, supported by less data than [
53,
54]. Therefore, based on our data and elements of the literature, we will assume the hypothesis of a kharif rain increase between 1998 and 2017 that could continue in the future. Such a change will strongly influence the future strategy for pearl millet systems improvement and should be closely monitored. We suggest that a part of the yield increase observed over the studied period could be due to those extra precipitations.
In other cases however, the potential extra kharif precipitation could directly or indirectly reduce the area allocated to pearl millet cultivation. A first hypothesis would assume a direct negative effect of rain due to an increase in the episodes of high intensity. The extra variability of rain, particularly strong in the G TPE, could result in short and intense rain episodes causing problems of water logging, stem mechanical break, and/or grain mold impacting the crop. Another hypothesis would assume an indirect effect. Indeed, we could also imagine that the extra available water during the traditional pearl millet growing season motivates the farmers to switch to crops that require more water, such as cotton or maize. Such a hypothesis is supported by the important increase area allocated to those crops in the B and G TPEs, which could explain the strong decline of pearl millet in those TPEs (see
Section 4.2).
This direct and indirect effect of rain could also partly explain the transfer of the pearl millet cultivation from kharif to Summer season observed in the G TPE and to a lesser extent in AE2 (
Figure S9). This transfer was confirmed by several experts and the DLD data. Indeed, in the G TPE, the proportion of area cultivated in Summer has progressively increased to reach more than 50% (
Figure S10). However, according to the DLD data, it does not represent a net area transfer between seasons but a larger propensity to cultivate pearl millet during Summer because, even during this season, the total pearl millet area stagnated or even decreased. In the AE2 TPE too, we could observe an increase in the area cultivated in Summer with up to 20 % of the yearly surface around 2009. However, after this year the proportion decreased. Those observations need to be confirmed over longer periods.
4.2. The Essential Role of Socio-Economic Factors in Pearl Millet Cultivation
The average yield increase of 37 kg/ha/year over all districts is similar to Yadav et al. [
56] findings, but hides important differences between TPEs (A1: 20 kg/ha/year, AE2: 58 kg/ha/year) that let us suppose some differences in the assimilation of new management practices and varieties. Generally, except in the G TPE, the yield increases were correlated with area trends. In TPE B and G, the decline in the pearl millet cultivated area was strongly correlated with the increased cultivation of crops, such as cotton, maize, or castor bean, having a more structured value chain and a higher profitability [
57,
58]. Therefore, the overall system dynamic seems to be closely related to socio-economic factors. If this is true, efforts to improve pearl millet production could be in vain if farmers continue to replace pearl millet by higher value crops. The necessity to strengthen the pearl millet value chain by increasing its profitability should, therefore, be a priority. However, such a task seems complicated because government policies, as well as trends such as urbanization and the changes in dietary patterns, are very difficult to influence [
59].
Another example of the importance of socio-economic factors was the increase in pearl millet cultivation area in the AE2 TPE while the price per quintal was the lowest. According to regional experts, substantial parts of the AE production (especially AE2) might be exported to A1, where pearl millet is a staple food and market is more important when A1 production is possibly not sufficient (OP Yadav personal communication). The increased prices occurring in A1 (around 20%) can motivate the export between regions. This would represent an example of commercially viable pearl millet value chain, but the hypothesis should be confirmed by further investigations. Generally, our study emphasizes the existence of different socio-economic drivers for pearl millet cultivation in TPE such as A1 that seems more oriented toward subsistence and AE2, which looks more market oriented.
4.3. The Use of Crop Model for Pearl Millet Envirotyping and Its Limitations
A second objective of this article was to test the ability of crop models to reflect the dynamics of the pearl millet production systems in India, which is the basis for future system optimization (e.g., Ronanki et al. [
24]), and to incorporate some crop model information into the TPE analysis. Here, we shall emphasize that the APSIM crop models do not simulate important biotic stresses, such as pest and disease attack. For example, it is documented that an important share of pearl millet has suffered from an increasing blast attack [
60]. The combined effect of those stresses with the assumption of representing a whole district by a single simulation can explain an important part of the difference between observation and crop model predictions. Despite those, in the A1, AE1, and G TPEs, which represent the large majority of area (
) and production (
), we could obtain prediction abilities between 0.15 and 0.61
(squared correlation for comparison) that were comparable to the values about regional prediction from other studies Chen et al. [
61] (0.26–0.42) de Wit et al. [
62] (0.24–0.65).
Nevertheless, in regions that potentially experienced rapid changes (B, AE2, and G), the crop model was less effective to capture the production fluctuations. In those TPEs, the prediction ability over the years was strongly reduced, which is likely due to a rapid and large variability in the crop management practices between the seasons. The socio-economic drivers and the potential modification in the rain pattern (see above) could be major drivers of those changes. Therefore, we shall emphasize the interest of combining a large set of data, allowing the joint understanding of biophysical process and socio-economic trends, which seem to have a strong effect on the crop model prediction ability in the future. Reduced prediction abilities in AE2, B, and G could also result from choice of the testing locations to develop the models (Telangana and Rajasthan). The collection of extra data on varieties in AE2, G, and B testing sites during future development would help to improve the crop model prediction ability there.
The evaluation of an updated version of the APSIM model for dynamic tiller simulation was promising. We could show that, despite the general overestimation of the average yield, this new model offered more sensitivity to predict yield trends (+0.18–0.19 correlation) in the A1 and AE1 (core of the pearl millet tract,
area,
production). Therefore, this model could be used for optimization of the Indian pearl millet system. The crop model sensitivity analysis emphasized the importance of the soil parameters for yield. It supports the conclusion from Carberry et al. [
63] that this component should receive extra attention. Finally, the large effect of the variety, especially in AE2, B, and G suggests to further dissect the optimal genotype properties to enhance pearl millet production in those TPEs.
One of the novelties presented in this article was the use of extended simulations to estimate ExM parameters. We showed that this strategy could help to estimate parameters with large influence on yield like soil water content or irrigation, but its capacity to infer less influential parameters, such as the sowing date, was weak. Generally, the ExM parameters estimation tended to determine values that overfit the observed data. Nevertheless, the estimation of ExM parameters could be useful in situations with low levels of information.
5. Conclusions
In this section, we summarize the main characteristics of each TPE.
A1 (2.9 Mha, 1.4 Mton, 499 kg/ha): This TPE is characterized by harsh agronomic conditions due to low precipitation, high temperature, and potentially poor soils (sandy soils, low water content). This scarcity largely explains the reduced number of alternatives to pearl millet in A1, which is an important reason for its stability there. Potential reduced benefits could explain the low level of input (irrigation, fertilization) and the limited effect of improved varieties observed in the A1 TPE. Our data revealed that a proportion of pearl millet might be imported from AE but further investigation is needed to confirm this hypothesis. Increasing the local production to meet this demand could be a driver for pearl millet cultivation in A1. With 8 testing sites out of 76 in the All India Coordinated Research Project (AICRP;
Figure 1B), the A1 TPE is under-represented. Resources from sites falling in none of the TPEs (17/76), from the B TPE (17/76) or the G TPE (14/76) should be re-allocated to A1.
Districts (Rajasthan): Barmer, Bikaner, Churu, Hanumangarh, Jaisalmer, Jalore, Jodhpur, Nagaur, Pali.
AE1 (2.1 Mha, 2.8 Mton, 1397 kg/ha): This TPE is the area where most of the pearl millet was produced between 1998 and 2017 (around 39% of the total production). The increase in production is due to a moderate growth of the cultivated area together with high yield increase that appeared to be associated with good soil properties, technological investments (irrigation, fertilization, improved varieties), and a hypothetical rain increase. We recommend to increase the attention on this TPE, for example by allocating more testing sites there, especially in AE1 southeast part (
Figure 1B).
Districts (Rajasthan): Ajmer, Alwar, Bharatpur, Dausa, Jaipur, Jhunjhunu, Karauli, Sawai Madhopur, Sikar, Tonk;
(Haryana): Bhiwani, Gurgaon, Hisar, Jhajjar, Jind, Mahendragarh, Rewari;
(Uttar Pradesh): Mathura.
AE2 (0.7 Mha, 1.37 Mton, 1859 kg/ha): AE2 is the only TPE where the fraction of area under pearl millet cultivation increased between 1998 and 2017. This success could be linked to suitable soil type (higher water content and depth), and a potential moderate increase in rain. AE2 is a region where pearl millet production benefited from technological investment (fertilization, irrigation, possibly adoption of new hybrid cultivars), which made pearl millet competitive with respect to other crops such as maize. The lowest market prices obtained in AE2 support the hypothesis of an exportation of surplus into the A1 TPE. With 6 testing sites out of 76, this TPE is the least represented among the AICRP testing sites (
Figure 1B). It would be strategic to increase the testing sites coverage to better understand the reasons of AE2 success and the way the pearl millet economy functions in this region.
Districts (Uttar Pradesh): Agra, Aligarh, Auraiya, Budaun, Etah, Etawah, Firozabad, Kasganj, Sambhal;
(Madhya Pradesh): Bhind, Morena;
(Rajasthan): Dholpur.
B (1.4 Mha, 1.1 Mton, 788 kg/ha): The B TPE is characterized by the largest decline of area under pearl millet cultivation. The presence of many alternatives, such as cotton and maize, with more profitable value chains might explain the transition from pearl millet to other crops. This declining interest translated in a low level of technological investment and resulted in very limited yield increases. Therefore, we recommend to investigate the reasons for abandoning pearl millet cultivation in this TPE. Districts (Maharashtra): Ahmednagar, Aurangabad, Beed, Dhule, Jalgaon, Jalna, Nashik, Pune, Sangli, Satara; (Karnataka): Bagalkot, Bijapur, Gulbarga, Koppal, Raichur; (Uttar Pradesh): Allahabad.
G (0.4 Mha, 0.5 Mton, 1208 kg/ha): In Gujarat, the pearl millet production sharply decreased, probably because of the availability of more profitable crops, such as cotton or castor beans. The data suggest that a part of the kharif pearl millet production has been transferred to Summer. This transfer is potentially due to an increase in the rain volume and variability during kharif. Despite a relatively high investment in technology (irrigation, fertilization) and high yield, the reduction in the pearl millet cultivated area was not compensated by the extra yield. The G TPE can be used as a testing site to investigate the influence of rain pattern on pearl millet cultivation, especially the question of potential cultivation transfer between the seasons. Districts (Gujarat): Banas Kantha, Bhavnagar, Kachchh, Kheda, Mahesana, Patan.