Climate variability has posed increasingly large-scale threats that have a significant impact on all agricultural systems worldwide, such as climate change, various uncontrollable variables, shifting precipitation patterns, and extreme temperatures combined with the degradation of fertile soil jeopardize food security in major cities. For this reason, steps have been taken to develop predictive models based on artificial intelligence technology as a sustainable alternative for agricultural management in times of change, through the early detection of agricultural stress conditions before they affect the harvest and to reduce yield losses. This paper proposes a hybrid predictive framework for assessing crop health that integrates time series forecasting models, including Prophet and ARIMA, with machine learning algorithms such as Random Forest and support vector machines. The objective of this framework is to process two main data streams: climatological variables such as temperature, precipitation, and humidity, along with soil indicators including pH, nutrient levels, and soil moisture, by combining the temporal decomposition capabilities of time-series models with the ability of ensemble- and kernel-based classifiers to recognize nonlinear patterns. The proposal’s primary aim is to provide highly robust, multi-horizon predictions of the current crop health status across various agricultural contexts. The method employed is based on a systems thinking approach, which treats the crop-climate-soil system as an interconnected whole, rather than conceptualizing it as isolated components. A conceptual architecture is presented, including an analysis of data requirements, various integration challenges, and indicators of expected yields. The results are expected to prove the complementary strengths of both model families when applied to heterogeneous agricultural datasets. Contributing to sustainable and climate-resilient agriculture by offering a scalable, data-driven decision-support tool that can be replicated across multiple crop types and geographic regions, future work will focus on model validation using open agricultural and meteorological datasets.

Keywords

Artificial Intelligence, Time-series forecasting, Agricultural resilience, Food safety and Climate change adaptation.