Climate change poses a significant threat to economically vital crops, potentially impacting farm financial well-being. However, understanding the precise nature of this impact remains challenging. This study focuses on cherry and peach farmers in Chile and Tunisia, regions where these crops hold significant nutritional and economic value and are vulnerable to climate change. Previous research has largely concentrated on staple crops like wheat, rice, corn, and soybeans in other regions, leaving a gap in knowledge regarding fruit crops in North Africa and South America. Moreover, existing studies often lack comprehensive assessments of the effects on overall farm financial well-being, not just crop yields. This research aims to bridge this gap by investigating the influence of climate change and other factors on the self-reported financial well-being of fruit farmers in Chile and Tunisia. The study's significance lies in its potential to inform policy decisions by identifying the most impactful factors on farm financial well-being, allowing for targeted and effective interventions.

Existing literature demonstrates climate change's detrimental effects on crop yields, particularly through temperature increases and reduced precipitation. Studies have focused primarily on staple crops in regions like Asia, Europe, and North America, neglecting the vulnerability of fruit crops in North Africa and South America. While the impact of climate change on fruit crop quality and yield has been studied, assessing its influence on farm financial well-being remains challenging. Traditional economic models, such as the Ricardian approach, often focus on land value and agricultural revenue but frequently omit crucial socio-economic and adaptive measures. Machine learning offers a promising approach to analyze large datasets and numerous variables simultaneously, reducing the risk of overlooking important factors. While machine learning has shown potential in agriculture, integrating it with traditional statistical analysis can enhance interpretability and address the complexities of agricultural datasets. This study addresses these limitations by using a hybrid approach that combines machine learning with generalized linear models, enabling detailed analysis while preserving the interpretability of results.

This study employed a hybrid approach combining machine learning and generalized linear models to analyze data collected through face-to-face interviews with 801 fruit farmers (401 in Tunisia and 400 in Chile). The farmers were randomly selected from lists provided by the respective Ministries of Agriculture. The sample included farmers who owned, managed, and worked on their farms, deriving over 70% of their income from farming activities. The survey, translated into Tunisian Arabic and Chilean Spanish, included multiple-choice, open-ended, Likert scale, and yes/no questions. Questions focused on climate change impacts (temperature, precipitation, extreme weather, and drought) experienced between 2009 and 2018, as well as farm-related variables grouped into natural, human, social, biophysical/manufactured, and economic assets. The dependent variable measured farmers' self-reported financial well-being. Data analysis involved logistic regression to assess the impact of climate change factors on financial well-being, and sparse group boosting (sgb) to determine the relative importance of various variables and their interactions. The sgb model allowed for the selection of individual or grouped independent variables, offering both predictive power and interpretability. The model's performance was evaluated using 25-fold cross-validation. The researchers compared the sgb model with other machine learning models (e.g., random forest, neural networks) to assess its predictive capability and interpretability trade-offs. The data was split into training (70%) and test (30%) datasets. Model evaluation metrics included accuracy and the area under the receiver operating characteristic curve (AUC). All analyses were performed using the R statistical programming environment.

The study revealed that decreasing rainfall and increasing temperatures were associated with reduced farm financial well-being, with these effects being more pronounced in Chile than Tunisia. Increasing drought frequencies and extreme weather events did not significantly impact farm financial well-being. While climate change factors played a role, particularly for farms already struggling financially, other factors proved more influential in predicting overall farm financial well-being. In Chile, farm location was the primary determinant, with farms in central Chile performing better financially. In Tunisia, the presence of social assets, including access to and trust in information sources, was the strongest predictor. Sparse group boosting analysis highlighted the importance of social and biophysical assets (farm size, water management, crop diversity) for high financial well-being in both countries. In Chile, natural assets (regional differences) were also significant, while in Tunisia, prior family ownership and human assets (education, age, gender) mattered most. For low financial well-being, regional differences and economic assets (debt, reliance on orchard income) were key in both countries. Variable interaction analysis indicated synergistic effects between trust in information sources (newspapers, industry experts) and the use of industry information. Farmers who trusted and used media sources were more likely to do well financially regardless of climate impacts. In Chile, education and media use also showed a positive modifying effect on financial well-being.

The findings challenge the assumption that climate change is the most critical factor determining fruit farm financial well-being. While climate change impacts are noticeable, particularly regarding rainfall and temperature, other factors like social assets, geographical location, and access to information prove more influential. Policymakers should focus on strengthening these factors to enhance farm resilience and financial stability. The synergistic effects observed between trust in information sources and their usage highlight the importance of targeted communication strategies and building farmer trust in credible information. The study's regional findings underscore the need for context-specific policies, recognizing differences in influential factors between Chile and Tunisia. Future research should focus on exploring the long-term implications of climate change on farm financial well-being, especially as temperatures continue to rise and precipitation decreases.

This study demonstrates a hybrid modeling approach combining statistical models and machine learning to predict farm financial well-being. Results show that while climate change impacts exist, social and geographic factors are more critical for determining farm financial well-being. Policymakers should focus on strengthening farmers' access to information, trust in information sources, and enhancing social capital to improve financial resilience, especially in the context of current climate trends.

The study's limitations include the use of self-reported data on financial well-being, which may differ from actual financial performance. The sample size, although substantial, might not fully capture the diversity within each country. Future research should explore the use of objective financial data and larger datasets to enhance generalizability. The focus on cherry and peach farmers in specific regions may limit the generalizability of findings to other fruit crops or regions.