Development of a system for monitoring and managing climate-dependent process risks based on hidden Markov models (using grain crop yields as an example)

Abstract

This study focuses on climate-dependent production processes, in particular grain crop yields, under conditions of climatic variability and uncertainty in Northern Kazakhstan. The problem addressed is the low effectiveness of deterministic risk monitoring approaches due to limited predictive power and the lack of formalized risk criteria, which leads to unreliable decision-making under uncertainty. The results include the development of a three-state hidden Markov model (S0–S2) and a TO-BE architecture for continuous risk monitoring and decision support. The model enabled the identification of latent climatic regimes and probabilistic assessment of risk states for 2025. The highest probability of an unfavorable regime was observed in Korgalzhyn (61.2%) and Ereymentau (58.8%), while Arshaly (42.9%) and Zhaksy (38.1%) showed moderate risk levels. The Brier score ranged from 0.106 to 0.199, confirming acceptable calibration of probabilistic estimates. The key feature of the approach is the representation of climate-dependent processes as transitions between latent probabilistic states, allowing the capture of temporal dependencies (climate memory) and the persistence of unfavorable conditions. Unlike deterministic models, the proposed framework enables dynamic risk tracking through continuously updated probability estimates integrated into a monitoring loop. The advantage of the approach lies in combining probabilistic modelling with an operational architecture, where risk probabilities serve as formalized decision-support signals. The results can be applied in early warning systems and digital monitoring platforms using remote sensing and IoT

More by Nurzhamal Kashkimbayeva

• Evidence-Linked Measurement of Indirect Prompt Injection in Web Security Agents: Paired Trials and Application-Specific Evidence AuditsTechnologies · 2026
• Kazakhstan Agri Recommender — Data Collection PipelineZenodo · 2026
• Replication Data and Code for “Added Predictive Value and Temporal Transferability of MODIS and ERA5-Land Data for District-Level Wheat Yield Forecasting”Harvard Dataverse · 2026
• Multimodal Data Fusion for District-Scale Crop Yield Prediction in Northern KazakhstanIEEE International Conference on Smart Information Systems and Technologies · 2026