As a fundamental hydrological variable, rainfall contributes significantly to the land surface and atmospheric processes. There have been many applications for rainfall forecasting, such as drought monitoring and optimizing irrigation water management, flood forecasting and early warning systems, reservoir water level management for hydropower generation and more. Forecasting rainfall is challenging for meteorologists due to the variability in rainfall timing and quantity. As a result of its persistence and complexity, rainfall forecasting has piqued the interest of academics. Moreover, heavy rainfall can result in severe floods, causing significant fatalities and economic damage. Therefore, a more accurate rainfall forecast can aid in the formulation of appropriate measures that can reduce the risk of flooding and/or provide advanced early warning to people whose property may be within the flood damage zone. It is recognized that existing models use complex statistical models, which are often neither computationally nor technically feasible for many jurisdictions, hence downstream applications are unaffected by them. It is therefore being explored as a possible solution to these shortcomings to employ machine learning algorithms in combination with real-time monitoring technology. Accordingly, the current study presents a comparative analysis using different extreme learning machines suitable for specific downstream applications. Various models are defined for multi-step ahead forecasting of the rainfall in Quebec City, Canada. A simple mathematical formula derived from the current study could be applied to practical engineering problems.