A new study comparing crop yield prediction models reveals insights into the effectiveness of different modeling in the face of climate change.

Model-assisted decision-making can minimize farmers’ vulnerability to climate change by predicting crop yield and the variables that influence it. With this information farmers can anticipate changes in crop yields, adjust their farming practices, explore crop diversification options, and plan for adaptation strategies to reduce vulnerability to climate-related risks.

A new study published in in silico Plants compares the yield predictive capacity of multiple modeling approaches.

Achraf Mamassi, a former PhD student at University of Liège in Belgium and Mohammed VI Polytechnic University in Morocco (now a Postdoctoral fellow at INRAE in France) led a study that compared the ability of process-based mechanistic models and empirical models to predict wheat yield across the rainfed areas of Morocco.

Morocco has experienced considerable warming, increasingly erratic rainfall, and an overall decline in precipitation over the last few decades, resulting in highly variable crop yields and large yield gaps. An increase in the frequency and intensity of extreme events is expected under climate change, putting the food and economic balances of the country at risk.

Rainfed parts of Morocco are responsible for 80% of the total cereal production of the country. These areas are particularly vulnerable to climate change.

The research focused on two distinct types of models: mechanistic and empirical models. These models diverge in their fundamental principles and levels of complexity.

Mechanistic (or process-based) models are based on a deep understanding of the biological and physiological processes underlying crop growth. They are developed using scientific principles and mathematical equations that represent the physical, chemical, and biological processes occurring in crops. Mechanistic models simulate the interactions between environmental factors (such as temperature, rainfall, and radiation), soil conditions, and crop characteristics to predict crop responses and yield. The APSIM-Wheat model was chosen to represent this model type.

Empirical (or statistical) models, on the other hand, are based on observed relationships between input variables (such as weather conditions, soil properties, and management practices) and crop yield. These models do not explicitly incorporate the underlying biological mechanisms of crop growth. Instead, they rely on statistical methods and historical data to establish patterns and correlations between inputs and outputs.  Two empirical models, multiple regression (MR) and random forest (RF) models, were considered in this study. MR are based on simple linear relationships between predictors and response variables while RF is based on complex statistical algorithms.

The study incorporated a dataset spanning three years, encompassing data from 125 wheat fields located in the rainfed regions of Morocco. The dataset comprised various parameters such as phenology, temperature, rainfall, soil chemistry, and crop management practices. The fields were assigned to regions based on annual rainfall (i.e. favorable, intermediate, and unfavorable).

The performance of the three models was evaluated by assessing the precision and accuracy of the simulated yields in comparison to the measured yield.

The results showed that APSIM-wheat’s predictive capacity was higher than the empirical models. Both empirical models were able to make accurate, but not necessarily precise, predictions. However, the random forest approach was unable to fit a model and identify yield predictors in unfavorable regions. The APSIM model was consistently more precise than the empirical models although it was less accurate for intermediate and unfavorable regions.

Nonetheless, mechanistic models are useful because they can identify the critical variables that contribute to predicting wheat yield. This is because they are built based on the biological and physiological processes that drive crop growth. These models revealed several significant variables that affect yield, including leaf density and distribution during the heading stage, climate factors such as maximum temperatures at emergence and tillering, and the amount of fertilizer applied during the heading stage.

Rather than clearly supporting the superiority of one type of model over another, the authors advocated for complementary use of all approaches depending on data availability and targeted time horizon for yield simulations (one-year vs. decades) and the modelling objectives.

Empirical models are preferred for long-term predictions due to their precision and ability to capture the effects of climate variability. In contrast, Mechanistic models like APSIM are more suitable for making agronomic recommendations that save time and resources by focusing on those with the most significant impact. They have the advantage of being able to predict yield as soon as essential parameter values become available.

Mamassi explained the impact of this work on Moroccan farmers.

“The adoption of a fundamental pillar of Precision Agriculture, represented by mechanistic and empirical modeling, in Moroccan agriculture could empower farmers in rainfed regions. Firstly, it contributes to assessing the potential yield of these regions and highlighting factors behind crop yield gaps. Secondly, it offers precise long-term predictions and efficient recommendations customized to specific agro-pedo-climatic conditions, respectively. This could lead to better resilience against climate variability, optimized resource allocation, and improved yields, ultimately enhancing the sustainability and productivity of Moroccan farming practices.”

READ THE ARTICLE:

Achraf Mamassi, Marie Lang, Bernard Tychon, Mouanis Lahlou, Joost Wellens, Mohamed El Gharous, Hélène Marrou, A comparison of empirical and mechanistic models for wheat yield prediction at field level in Moroccan rainfed areas, in silico Plants, Volume 6, Issue 1, 2024, diad020, https://doi.org/10.1093/insilicoplants/diad020

Laisser un commentaire

Votre adresse e-mail ne sera pas publiée. Les champs obligatoires sont indiqués avec *