Hydroelectric plants are a key part of the future of renewable energy, accounting for approximately 40% of total renewable capacity according to the World Economic Forum. These plants, often complex, are subject to energy production variability due to rainfall.
Hydroelectric assets are required to be planned for production 9-12 months ahead, like a traditional source plant. Fundamental elements for correct program planning consist of the weather forecast and, in particular, rainfall and temperature forecasts. The last few years have seen, above all in Italy and Spain, the alternation of dry seasons followed by extreme acute events and floods.
Medium-long term forecasts, especially in European countries, are still very unreliable today, so much so that the most accurate value that can be used is given by the historical average. However, the meteorological anomalies of recent years have led to a strong deviation from this value as well, making it difficult to make reliable estimates.
The development of an advanced forecasting model, therefore, could finally make it possible to predict even acute events, in order to align the production estimates of hydroelectric plants.
Enel Green Power is looking for proposals for innovative models capable of providing accurate and reliable medium-to-long term (9-12 months ahead) forecasts of rainfall and temperature, especially in the Mediterranean area. In particular, this Challenge relates to the existing difficulty in predicting weather and climate events in Italy and Spain, except for the very short term.
Enel Green Power has looked at traditional climate models and meteorological processes internally, including the aptness of the North American Multi-Model Ensemble (NMME), European Centre for Medium-Range Weather Forecasts (ECMWF), etc. Forecast Centers the world over have extensive infrastructure designed to predict rainfall, temperatures, concentration of both, and acute weather events. Since the hereabove mentioned models didn’t provide reliable/representative results for our specific purposes, in particular for 9-12 months ahead mainly in Italy and Spain, Enel is looking for a model that uses "different/disruptive” approaches and variables, different from those normally used.
The model may use different kinds and blends of approaches (i.e. statistical, or physical, or machine learning modelling). The forecast must be passed according to macro-areas of homogenous climate conditions. We very much encourage innovation in this Challenge, for instance, your solution may take the form of a hybrid model using elements of traditional models combined with an innovative approach, or AI-backed methods for accurately predicting rainfall and temperature 9-12 months ahead. Solvers may leverage these existing forecasts or ensembles in their solution, but must be able to demonstrate the value added by their model, relative to any input datasets or foundational frameworks, for better chance of a full award.
This forecast model must be capable of predicting and quantifying seasonal rainfall and temperature at least 9-12 months ahead, on monthly and annual granularity, to allow the inclusion of these forecast values in the energy production models used by Enel Green Power. Furthermore, the model would ideally be able to predict acute events and prolonged periods of drought, in which the traditionally monitored quantities can deviate from the historical average values.
Across your solution, system, or proposal, Enel Green Power requires certain levels of data and consistent formatting for use in forecasting planning of hydroelectric assets in the mid-to-long term. Your proposal must:
Use historical data available for the last 20-30 years (if possible, but not exclusively using the ERA5 database),
Present the MAE (Mean Absolute Error) analysis related to these correlation anomalies and compare it with the MAE obtained using the simple historical mean,
Your correlations and MAEs must be compared with data obtained through the standard seasonal models such as the North American Multi-Model Ensemble (NMME), European Centre for Medium-Range Weather Forecasts (ECMWF), etc.
The minimum time step used can be the month.
The model must provide values in .csv or .xls format or in any case compatible with Enel Green Power standards.
Your submission, including your model and validated data, must provide Enel Green Power an estimation of the rainfall and temperature one year ahead in Italy and Spain:
- Give an estimation of rainfall/precipitation 9-12 months ahead, in millimetres/day on monthly and annual granularity,
- Present the correlation anomalies of precipitation values between the real anomaly (difference between the real and the mean value) of precipitation and the forecasted anomaly (hindcast). It must be done for a test period of at least 5 years, 2018-2022 included
- Give an estimation of temperature ranges 9-12 months ahead, in degrees Celsius, on monthly and annual granularity,
- Present the correlation anomalies for temperature values between the real anomaly of temperature and the forecasted anomaly, again across a test period of at least 5 years, 2018-2022 included
Provide a locally executable application to verify the data or model for Enel Green Power to use,
Forecast model should report on a monthly and annual basis,
Provide data for the Italy and Spain perimeter, labelled, according to macro-areas of homogenous climate conditions – wide area predictions, rather than day-to-day weather forecasts,
- The model for rainfall and temperature forecast should provide a percentage confidence level not above 5%.
Your proposal might also fulfil these nice-to-have requirements:
Acute events prediction: forecast of prolonged drought or floods during the 9-12 months ahead, with probabilities and validated data attached,
Snow Water Equivalent forecast, described as the equivalent amount of liquid water stored in the snow pack measured in millimetres, on monthly and annual granularity, which is particularly useful in the north of Italy,
- Provide predictions for the concentration of rainfall/precipitation in subregions of Italy and Spain, 9-12 months ahead.
The submitted proposal should consist of a detailed technical description including:
Forecasting model or system for temperature and rainfall that meets the Technical Criteria and Solution Requirements, with a particular focus on and accuracy around the regions of Italy and Spain;
Locally executable application to verify the solution;
Detailed description of results, performances and characteristic of the proposed model solution, particularly compared to the current models;
- Data, case studies, patents and journal references or any additional material that supports the proposed solution or model.
This Challenge contributes to the following UN Sustainable Development Goals:
- SDG 7: Ensure access to affordable, reliable, sustainable and modern energy for all
- SDG 9: Industry, Innovation and Infrastructure