H2020 CLIM4CROP Project: Climate Monitoring and Seasonal Forecasting for Global Agricultural Production

Accurate and timely drought information is essential for moving from post-crisis drought risk management to pre-impact drought risk management. Several drought datasets already exist. These cover the last three decades and provide near-real-time data (using different sources), but all are "deterministic" (i.e., from a single realization), and the results partially differ between them. In CLIM4CROP, we first assessed the quality of continuous, long-term climate data for timely meteorological drought monitoring, considering the Standardized Precipitation Index. We subsequently developed a new global gridded land-based dataset, operationally updated every month, for drought monitoring using a probabilistic approach (DROP) inspired by the multi-model approach in weather and climate prediction. We present a monitoring tool that uses an ensemble of observation-based datasets to obtain the best estimate in near real-time with its associated uncertainty. This approach makes the most of the available information and makes it available to end-users. The high-quality, probabilistic information provided by DROP is useful for monitoring applications and can help inform global policy decisions on adaptation priorities to alleviate the impacts of drought, especially in countries where meteorological monitoring remains a challenge. The article describing the DROP dataset is currently under review in the Bulletin of the American Meteorological Society; that is, fulfilling the project's dissemination task.

The links between crops and drivers of heat/water stress have been analyzed to better understand their interactions and develop statistical models. These models will be used to explore the seasonal predictability of crop yields. We used historical yield data for the four major crops worldwide (maize, rice, soybeans, and wheat) from the Global Historical Yield Dataset (Iizumi 2017), a recently developed gridded dataset never before used for this type of study. Importantly, we established a collaboration with the developer of this dataset, Dr. Iizumi, an internationally recognized agrometeorologist.

Historical crop data and potential climate predictors have been used to calibrate parsimonious regression models, providing a computationally cost-effective alternative to process-based models that typically require a variety of fine-scale variables as inputs, where seasonal forecasts are not yet effective. The researcher received training from JRC staff in developing these models, which follow and extend the models developed by Zampieri et al. (2017, 2018, 2019a) and Ceglar et al. (2018).

The novelties of the developed empirical model are mainly based on (i) taking into account the potential effect of antecedent (i.e., within-season) climate conditions on crop yield and (ii) calibrating the model to achieve out-of-sample crop predictions based on knowledge of predictor data outside the period used to train the model, adopting a leave-one-out cross-validation method. That is, the calibration of the crop-climate models and their evaluation are performed using cross-validation to assess the predictions as if they were made operationally. Specifically, merging observational information (for months preceding harvest months) with seasonal forecasts for the rest of the crop growth calendar (e.g., anthesis and harvest stages) is a special feature of our approach that can contribute to increasing crop predictability, making the most of the best information available to users. This is especially useful in areas where the performance of dynamic forecasting systems is still affected by significant errors (e.g., Europe). Preliminary results suggest that lagged relationships between climate and crops contribute substantially to the development of a seasonal forecasting system that enables more efficient crop management.

CLIM4CROP was conceived to offer a unique opportunity to strengthen and expand the fellow's multisectoral experience, connecting their training in meteorology and climate change studies with the theoretical and practical requirements for climate prediction applications in agriculture. Therefore, this MSCA has also addressed the training objectives outlined in the proposal on seasonal prediction, where the expert guidance of the BSC supervisor (FJ Doblas-Reyes) has been essential for the successful implementation of the work plan. Likewise, the training provided by JRC collaborators on crop yield impacts has allowed them to learn about agricultural production and food security and expand their competencies in statistical aspects related to climate impact models. Furthermore, the scientific challenges within the context of CLIM4CROP, such as highly cooperative research, have ensured the bidirectional transfer of new knowledge, strengthening the skills of all participants.