Global population and food demand are projected to rise considerably in the future 1, and it is thus essential to find robust and sustainable farming practices that can feed the world. Double cropping, a farming practice where two crops are harvested on the same piece of land in a single year 2,3, is a common and widely used cropland management technique in tropical and subtropical low-land regions 4. It is reported that double cropping can boost the total annual crop production 5,6, and can significantly increase the profitability and cash flow stability in local farms 2. Double cropping can be implemented in areas that allow for fast crop development and have favorable meteorological conditions throughout a large portion of the year. Additionally, double cropping can contribute to crop diversification, prevent soil erosion by maintaining plant cover 7, increase soil fertility by augmenting the input of root biomass 8,9, and spare land elsewhere for biodiversity 3,10 or other uses due to better land utilization 4. However, double cropping might lead to a depletion in soil nutrients if not managed properly 11,12, especially when the same crops are planted in the same field in the long run. Double cropping also requires more water than growing a single crop per season, which is problematic in areas with limited water resources 13.
Currently, double cropping is not as prevalent in Europe as in warmer regions of the world, but there is a growing interest in double cropping as a strategy for enhancing food security and production in the face of climate change 14. For instance, in recent years, some farmers in Italy and Spain have started to grow winter crops such as wheat or barley followed by summer crops like grain maize, silage maize or sunflowers during the same growing season 15,16. While this approach maximizes land usage and can help farmers increase their yield potential, further research is needed to fully understand the potential benefits of double cropping in the face of climate change in comparison to other farming practices such as single cropping systems, with only one single crop harvest per year. Additionally, there is a need to evaluate the feasibility of implementing double cropping systems in various regions throughout Europe, particularly considering the possible shortening of the crop growing cycle due to the ongoing warming trend 17–19 which increases the possibility to grow two crops in the same season.
To address these knowledge gaps, here we estimated the feasibility of double cropping with two major crops - maize and wheat - across Europe under current (2009–2020) and future (2089–2100) climate conditions and compared the caloric production under double cropping with maize and wheat in rotation versus single cropping with maize and wheat in rotation, a common rotation practice in Europe. To do so, we used a hybrid approach that combined a process-based model with machine learning to accurately simulate maize and wheat phenology and yield across Europe. The detailed cropping schemes are defined and illustrated by Supplementary Fig. 1. The process-based model, ORCHIDEE-CROP 20, is an updated version of ORCHIDEE land surface model 21 which incorporates basic modules from the STICS crop model 22. We first calibrated ORCHIDEE-CROP for Germany based on the crop phenology datasets 23,24 and a reference crop yield dataset 25. Yet, this calibration did not reproduce the spatial patterns of phenology and yield across Europe well, because ORCHIDEE-CROP cannot capture the diversity of cultivars and practices. To improve the performance of ORCHIDEE-CROP across distinct European countries where crop cultivars and practices differ 26, we collected phenology data from other European countries 24 in order to train machine learning models using the maize and wheat phenological stages and yield simulated by ORCHIDEE-CROP as inputs. This procedure allowed us to significantly improve the predictions of harvest dates and yield compared to the ORCHIDEE-CROP simulations. The machine learning models used in our study are reinforced random forest models (RRF) 27, which combine random forest models with generalized linear models and give a good overall accuracy and ability of extrapolation 27. Furthermore, there is an increasing trend along the crop planting year in the historical yield observations of maize and wheat from 1983 to 2016 (Supplementary Fig. 2), mainly due to the technical progress, including the improved crop cultivars 28,29 combined with the application of improved agronomic practices 29. To test for the possible effects of a continuing trend of yield increase in the future, we built two types of RRF models, one incorporating this trend and another that disregards it. The overall performance of our hybrid models for phenology (harvest date) and yield are given by Supplementary Fig. 3. The performance of the models was assessed by R-squared (R2), root mean squared error (RMSE) and relative error \(\left(RE=\frac{RMSE}{observed mean}\right)\). The R2 values for maize and wheat harvest dates are 0.2 and 0.58, and the RMSE values of maize and wheat harvest dates are 12 days and 9 days, respectively, with a relatively low relative error of 4% for both crops. This is especially noteworthy as our model is designed to estimate harvest dates across a broad range of cultivars, planting days, climate, and soil conditions throughout Europe. As for the yield models, without accounting for the yield increasing trend, the R2 and RMSE values for maize yield were 0.77 and 1738kg ha−1 yr−1 (RE = 24%), respectively, while those for wheat yield were 0.87 and 815kg ha−1 yr−1 (RE = 17%), respectively. When taking into account the yield increasing trend, the R2 values were 0.93 for both maize and wheat models. The corresponding RMSE values were 986kg ha−1 yr−1 (RE = 14%) and 587kg ha−1 yr−1 (RE = 12%) for maize and wheat yield models, respectively. Compared to crop modeling studies using well established mechanistic crop models such as WOFOST (ref. 30,31) and APSIM (ref. 32,33), our hybrid modeling approach appears to perform better (higher R2) with regard to reproducing observed maize and wheat yields. Our modeling strategy may thus be of interest for the global community of researchers in agronomy beyond the scope of our study’s objectives. The details of model description, algorithms for model calibration, training, tuning and testing, and setting for projection are explained in the Methods and further detailed in Supplementary Fig. 4 and Supplementary Fig. 5.
Our analysis shows that it is already possible to successfully implement double cropping with maize and wheat (where maize and wheat are harvested the same year) in most of Southern Europe under current climate conditions, with a probability of up to 60%. While in other parts of Europe, it is currently unlikely to implement double cropping (Fig. 1a). Interestingly, our results show that, at the end of the 21st century, the area where a double cropping system could be applied will expand 3.6 times under the high-warming scenario RCP8.5 compared to current level (Fig. 1c). Under this scenario, there is a chance up to 40% successfully implementing double cropping in both Western and Eastern Europe (Fig. 1b).
In order to assess the sensitivity of our results to different levels of warming, we ran additional simulations of double cropping under the strong mitigation scenario RCP 2.6, and the intermediate scenario RCP 6.0. We show that the potential area where a double cropping system could be implemented in Europe will expand 1.2 and 1.6 times under RCP 2.6 and RCP 6.0 scenarios, respectively (Fig. 1c). Until the end of 21st century, regions with a chance of at least 40% of successfully implementing double cropping will expand to the eastern part of Southern Europe under both RCP 2.6 and RCP 6.0 scenarios (Supplementary Fig. 6a, b), and to the southern part of Western and Eastern Europe under RCP 6.0 (Supplementary Fig. 6b). We calculated the calorie production under different levels of warming (Supplementary Fig. 7) and compared the annual caloric yield, averaged over 2089–2100, between RCP 6.0, RCP 8.5 and RCP 2.6 (Supplementary Fig. 8). We showed that the annual caloric yield decreased when the intensity of warmer effects increased (considering climate and CO2 concentration transitioning from RCP 2.6 to RCP 6.0 or RCP 8.5) in most of the regions where the probability of successful implementation of double cropping remains similar under all RCP scenarios. On the other hand, the annual caloric yield increased in the regions where the probability of double cropping increased (Supplementary Fig. 8, Fig. 1, Supplementary Fig. 6), which suggests that an increase of the probability of successfully implementing double cropping could compensate for the yield loss induced by the warming effect.
To estimate how the probability of successful implementation of double cropping could affect the productive performance of double cropping systems compared with single cropping system, we ran the simulation on these two systems in Europe under both current and future (RCP 8.5 scenario) climate conditions. We show that during the years when double cropping is successfully implemented (i.e., both wheat and maize are harvested the same year), the median annual caloric yield of the double cropping system is 70–145% higher than the single cropping system (maize or wheat is harvested each year) under RCP8.5 (2089–2100) (Fig. 2a). We found that the individual yields of maize and wheat in double cropping tend to be lower than the crop yields in single cropping (by -4 to -19% for maize, Supplementary Fig. 9c, and close to zero for wheat, Supplementary Fig. 9d), but these yield losses are compensated by the double harvest, leading to an overall caloric yield gain when the calorie yields of maize and wheat are added together. Thus, when double cropping can be successfully implemented, the increase in median caloric yield achieved through double cropping is between 133 to 394% compared to single wheat harvest (Supplementary Fig. 9b), but it is comparatively lower compared to single maize harvest (-2% to + 69% higher caloric yields in double cropping, Supplementary Fig. 9a). While during the years when double cropping cannot be successfully implemented (when the second season crop cannot be harvested due to a too long growing cycle, see Method and Supplementary Fig. 1b for the detail definition), the median annual caloric yield of double cropping is up to 17% lower than single cropping (Fig. 2b). Over all the years (with both success and failure of double cropping), the overall caloric yield gain resulting from double cropping is + 11 to + 33% when the probability of successful implementation of double cropping reaches 0.2, and this gain becomes stronger when the probability of success increases (Fig. 2c). In contrast, the single cropping system only provides a higher caloric yield when the probability of successful implementation of double cropping remains lower than 0.2 (Fig. 2c).
The gains induced by double cropping are contrasted between geographical regions. Our simulation results show that the highest calorie production levels are expected to occur in Western Europe for both single cropping and double cropping systems under both current and future climates (Supplementary Fig. 10, Supplementary Fig. 11). In this region, single cropping tends to outperform double cropping by 5.5% on average under current conditions, while the double cropping system tends to outperform the single cropping system in Southern Europe (with an overall increase of 15.5%, 13% for maize and 2.5% for wheat) (Fig. 3a). Due to climate change, the annual caloric yield in single cropping will decrease across most of Europe in the future compared with the current level, with decreases reaching 6.5% and 8.5% in Southern Europe and Western Europe, respectively (Fig. 3b), which is consistent with a previous study 34. In the future, double cropping could outperform single cropping, especially in Southern Europe, with an increase of 27.5% on annual caloric yield in this region (Fig. 3d). A large-scale implementation of double systems could thus compensate for the yield decrease in single cropping due to climate change in most of Europe except the northern part of Western Europe (Fig. 3c). When assuming a continuous yield increasing trend in the future from technical progress (Supplementary Fig. 2), the relative difference of caloric yield between double cropping and single cropping is similar to the value obtained without considering any yield increasing trend (Supplementary Fig. 12a, b, Fig. 3a, d), which indicates that the relative effects of double cropping compared to single cropping are insensitive to the assumption we made on future yield trends from the technical progress. When assuming a yield increasing trend, we show that the annual caloric yield in the future increased a lot for both single cropping (1.3–1.8 times) and double cropping (1.32–1.81 times) systems compared with the yield without considering the yield increasing trend (Supplementary Fig. 10, Supplementary Fig. 11, Supplementary Fig. 13), which suggests that technical progress could compensate for the negative impact of climate change on yield.