Preseason maize and wheat yield forecasts for early warning of crop failure

Effective humanitarian responses to food crises hinge on accurate and timely early warning systems. Many anticipatory action and disaster response programs, such as distributing farming inputs, securing funds, procuring food aid, and scaling up nutrition assistance, require months of lead time. The need for early warning information at lead times of six to twelve months is crucial given these logistical challenges. Currently, routine crop yield forecasts at such lead times are virtually nonexistent, with research primarily focused on within-season forecasts. This study addresses this gap by exploring the potential of preseason crop yield forecasts, building upon recent advancements in multiyear climate forecasting, particularly concerning the El Niño Southern Oscillation (ENSO). The widespread influence of ENSO on crop yields globally, coupled with improved multiyear ENSO forecasting capabilities, offers a promising avenue for developing skillful long-lead crop yield forecasts. The aim is to demonstrate the feasibility and accuracy of these forecasts, highlighting their potential to significantly enhance food security early warning systems.

Previous research on crop yield forecasting has predominantly concentrated on within-season forecasts issued a few months before harvest. While some preseason crop yield forecast systems exist for specific crops and regions (e.g., sugarcane in South Africa, maize in Zimbabwe, wheat and sugarcane in Australia, rice in the Philippines, and various crops in Europe and the US), these generally focus on shorter lead times (around six months before harvest) due to limitations in climate forecast availability. The authors acknowledge these past efforts but highlight the limitations in lead time. The advancement in multiyear ENSO forecasting provides a critical opportunity to extend these lead times. The use of ENSO in past preseason crop yield forecast systems, combined with its known impact on global crop yields, makes this a relevant area of investigation.

The authors propose a crop yield forecast system based on multiyear ENSO forecasts. The system combines historical probabilities of below-normal (bottom tercile) crop yield anomalies conditioned on ENSO phases (El Niño, La Niña, neutral) with model-analog based ENSO forecasts during the crop growing season. The historical relationship between ENSO phases and yield anomalies is established, and these probabilities are then linearly combined using the forecast probabilities of each ENSO phase for the target season. Forecasts are issued only before the start of the vegetative season. Data sources include FAOSTAT for global crop yield data, supplemented by subnational data for the US, China, and Australia. The Oceanic Niño Index (ONI) from ERSSTv5 is used to represent ENSO, with forecasts from NOAA/PSL and University of Colorado/CIRES model-analogs. Data quality control steps are implemented to address imputed values and low variance issues in the FAO data. The authors combine data from multiple countries to account for historical boundary changes. Crop growing seasons are aligned with the ENSO forecasts using GEOGLAM crop calendars. Forecast skill is evaluated using a hold-one-out cross-validation approach, employing the ROC score and reliability diagrams. Sensitivity experiments are conducted to explore the impact of selecting a single target season within the growing season and the effect of using perfect ENSO forecasts. Finally, a hindcast for the 1982 El Niño event is presented to illustrate the system's performance in a real-world scenario.

The study demonstrates that skillful preseason forecasts of maize and wheat yields are possible at lead times up to a year ahead of harvest. For forecasts issued 10–13 months before harvest, skill was observed over 15% of global maize cropland and 30% of global wheat cropland. Wheat forecasts consistently showed skill over a larger portion of global cropland than maize forecasts across all lead times. Regions with the highest forecast skill included Southeast Africa and Southeast Asia for maize, and parts of South and Central Asia, Australia, and Southeast South America for wheat. The percentage of global harvested area with skillful forecasts (ROC > 0.6) at lead times of 6-9 months, 10-13 months, and 14-17 months was greater for wheat than maize. Skillful wheat forecasts were observed for ~20% of harvested areas even at lead times of 18–21 months before harvest. Analysis showed that forecast skill varies depending on the country, crop, and month the forecast was issued. The study found a relationship between ENSO forecast skill and crop yield forecast skill. For example, forecasts issued prior to the spring predictability barrier had lower skill, reflecting ENSO predictability challenges. The skill of longer-lead forecasts appears to be linked to the enhanced year-two predictability of La Niña events following strong El Niño events. Analyzing forecast skill for individual 3-month seasons within the growing season revealed that targeting specific climate-sensitive periods could further enhance forecast accuracy. Sensitivity experiments showed that selecting a single target season significantly improved forecast skill, approaching levels achieved with perfect ENSO information. A hindcast for the 1982 El Niño event demonstrated both the system's ability to predict below-normal yields in some regions and its limitations in capturing all major yield failures, highlighting the influence of factors beyond ENSO on crop yields.

The findings confirm the potential of using multiyear ENSO forecasts to produce skillful long-lead crop yield predictions. The spatial pattern of forecast skill is consistent with the known influence of ENSO on crop yields in different regions. The improved forecast skill achieved by focusing on specific climate-sensitive periods within the growing season suggests a pathway for enhancing prediction accuracy. The sensitivity experiments indicate that even with perfect ENSO information, the model's skill is limited, implying the need to incorporate other climate variables and factors to improve forecasts beyond ~40% of global harvested areas. The 1982 El Niño case study highlighted the importance of considering factors beyond ENSO when interpreting forecast results. These results highlight the importance of both improving year-two ENSO forecasts and the incorporation of additional climate variability modes in preseason crop yield forecast systems.

This study demonstrates the feasibility and skill of producing preseason crop yield forecasts for maize and wheat, using multiyear ENSO forecasts. While skill is limited by the current ENSO forecast capabilities and the influence of factors beyond ENSO, the findings underscore the value of leveraging these forecasts for improved food security early warning systems. Future research should focus on incorporating other climate variables and improving ENSO year-two forecast skill to extend the geographic coverage and accuracy of long-lead crop yield predictions.

The study's reliance on ENSO as the primary predictor limits the forecast skill to regions strongly influenced by ENSO teleconnections. The accuracy of the forecasts is dependent on the quality and availability of historical crop yield data, which can be unreliable or incomplete in some regions. The model assumes a linear relationship between ENSO and crop yields, which might not always hold true. The study does not fully account for the impacts of other factors, such as pest outbreaks, disease, and agricultural practices on crop yields. The methodology relies on historical data which may not be fully representative of future conditions.