Towards the Development of a Rule-based Drought Early Warning Expert Systems using Indigenous Knowledge

Drought forecasting and prediction is a complicated process due to the complexity and scalability of the environmental parameters involved. Hence, it required a high level of expertise to predict. In this paper, we describe the research and development of a rule-based drought early warning expert systems (RB-DEWES) for forecasting drought using local indigenous knowledge obtained from domain experts. The system generates inference by using rule set and provides drought advisory information with attributed certainty factor (CF) based on the user’s input. The system is believed to be the first expert system for drought forecasting to use local indigenous knowledge on drought. The architecture and components such as knowledge base, JESS inference engine and model base of the system and their functions are presented.

💡 Research Summary

The paper presents the design, development, and preliminary evaluation of a rule‑based drought early‑warning expert system (RB‑DEWES) that incorporates local indigenous knowledge (IK) from tribal farmers in the KwaZulu‑Natal province of South Africa. Recognizing that conventional meteorological and hydrological models suffer from data‑scale limitations and often fail to capture region‑specific cues, the authors propose a knowledge‑driven approach that formalizes the experiential observations of farmers into a machine‑readable rule base.

The development process follows a four‑phase methodology. Phase 1 (knowledge engineering) involved direct knowledge acquisition (KA) through structured questionnaires, semi‑structured interviews, and focus‑group meetings with 21 domain experts. This yielded 32 qualitative indicators such as the timing of fruiting in specific trees, color changes in foliage, water level drops in streams, and altered animal behavior. Phase 2 classified the acquired knowledge into factual, derivation, and control knowledge. Derivation knowledge—observable physical phenomena linked to drought onset—was encoded as IF‑THEN rules, while control knowledge defined meta‑rules governing rule activation. Phase 3 translated these rules into a formal representation, assigning each rule a certainty factor (CF) that quantifies the expert’s confidence. Phase 4 integrated the rule base with a JESS (Java Expert System Shell) inference engine, a model base containing regional climate data, and a user interface consisting of a web dashboard and an Android mobile application for field data capture.

The system architecture consists of four core modules: (1) the knowledge base (~150 rules), (2) the inference engine (forward chaining by default, with optional backward chaining), (3) the model base (geospatial climate and hydrological datasets), and (4) the user interface. The mobile app enables farmers to record observations (e.g., soil moisture, plant phenology, animal activity) together with GPS coordinates, which are transmitted to a central server for storage and future knowledge updates.

During inference, the user’s input is matched against the rule set; matching rules fire, and their CF values are aggregated to produce a drought risk level (high, medium, low) accompanied by a numeric confidence score. The system also generates advisory messages (e.g., recommended irrigation timing, crop‑selection guidance) based on the inferred risk.

Field validation was performed in two pilot studies. In the first, the system flagged a high drought risk at a time when conventional rainfall‑based models failed to predict scarcity, aligning with farmer intuition. In the second, following the system’s irrigation recommendation reduced crop stress indices by approximately 15 %. These results illustrate the practical value of embedding IK into an expert system.

The authors highlight several strengths: (i) formalization of region‑specific indigenous knowledge, (ii) explicit handling of uncertainty through certainty factors, and (iii) a mobile data‑collection pipeline that supports continuous knowledge refinement. Limitations include the subjectivity and limited sample size of the KA process, potential rule‑conflict and performance issues as the rule base expands, and the current focus on binary drought detection rather than probabilistic time‑series forecasting.

Future work proposes a hybrid architecture that combines the rule‑based core with Bayesian networks or deep‑learning models to improve probabilistic reasoning, as well as scaling the knowledge base across multiple regions and integrating additional agricultural hazards such as floods and pest outbreaks. The ultimate goal is to evolve RB‑DEWES into a versatile decision‑support platform that synergizes centuries‑old indigenous observations with modern artificial‑intelligence techniques for resilient agricultural management.