The outbreak of seasonal crop diseases is a major risk to world food security and agricultural economies and the problem brings about massive loss in billions of dollars every year. Most disease management methods have been based on responding to the disease hence reducing their effectiveness in the control of the prevalence of the disease. In this review, the emerging paradigm of artificial intelligence-based predictive models in forecasting seasonal crop diseases is critically reviewed, especially on the architecture of deep learning models for temporal forecasts, and their economic consequences. Recent time-series modelling

techniques such as Long Short-Term Memory (LSTM) networks and Temporal Convolutional Networks (TCNs) are synthesized based on meteorological, phenological, and pathological data and used to predict the occurrence of a disease with lead times long enough to implement preventive measures. Additionally, we discussion economic modelling frameworks, which quantify the monetary advantages of early prediction  systems by increasing yield maintenance, diminished applications of pesticides and optimized resource directing by mathematical calculation. Case studies across large systems of crop-pathogens, such as rice blast, tomato late blight and wheat rust reveal that the prediction accuracy was higher than 85% and possible economic returns on investment were in the range of 300-500. We find such major flaws in extant literature such as inadequate incorporation of socioeconomic factors, lack of participation in various agroecological regions, and lack of strategies to deploy them to farmers. This review offers an all-inclusive basis to the researchers, policy-makers, and all interested agricultural stakeholders who might want to use AI technologies to promote sustainable disease management and economic health in agriculture.