The Integrated Multisatellite Retrievals for Global Precipitation Measurement (GPM) (IMERG) Level 3 estimates rainfall from passive microwave sensors onboard satellites that are associated with several uncertainty sources such as sensor calibration, retrieval errors, and orographic effects. This study aims to provide a comprehensive investigation of multiple machine learning (ML) techniques (Random Forest, and Neural Networks), to stochastically generate an error-corrected improved IMERG precipitation product at a daily time scale and 0.1°-degree spatial resolution over the Brahmaputra river basin. In this study, we used the operational IMERG-Late Run version 06 product along with several meteorological and land surface parameters (elevation, soil type, land type, soil moisture, and daily maximum and minimum temperature) to produce an improved precipitation product in the Brahmaputra basin. We trained, tested, and optimized ML algorithms using 4 years (from 2015 through 2019) of reference rainfall data derived from the rain gauge. The ML generated precipitation product exhibited improved systematic and random error statistics for the study area, which is a strong indication for using the proposed algorithms in retrieving precipitation across the globe. We conclude that the proposed ML-based ensemble framework has the potential to quantify and correct the error sources for improving and promoting the use of satellite-based precipitation estimates for water resources applications.

The accuracy of machine learning–based power outage prediction models (OPMs) is sensitive to how well event severity is represented in their training datasets. Unbalanced or overly dispersed event severity can result in random errors in outage predictions and underestimation in severe events or overestimation in weak ones. To improve accuracy in the prediction of storm-caused power outages, we introduce a novel method called “Conditioned OPM” that divides an OPM training dataset into subsets of events representative of the predicted event’s severity by calculating the quantile weight distance (QWD) between severe weather-related events in the dataset and the predicted event. Based on 102 storm events (including two hurricanes, Irene and Sandy), that have occurred since 2005 over Eversource Energy’s Connecticut service territory, we quantified the weather differences among predicted events, which we classified into three groups of severity: low, moderate, and high. The Conditioned OPM creates a subset of the historical events based on their classified severity group and uses that subset as the training dataset to predict the power outages. The study shows that the accuracy of event severity classification was 0.76, and the mean absolute percentage error (MAPE) decreased by about 30%; this method was also tested on forecast events and exhibited a low (20%) MAPE.

A growing number of electricity utilities use machine learning-based outage prediction models (OPMs) to predict the impact of storms on their networks for sustainable management. The accuracy of OPM predictions is sensitive to sample size and event severity representativeness in the training dataset, the extent of which has not yet been quantified. This study devised a randomized and out-of-sample validation experiment to quantify an OPM’s prediction uncertainty to different training sample sizes and event severity representativeness. The study showed random error decreasing by more than 100% for sample sizes ranging from 10 to 80 extratropical events, and by 32% for sample sizes from 10 to 40 thunderstorms. This study quantified the minimum number of sample size for the OPM attaining an acceptable prediction performance. The results demonstrated that conditioning the training of the OPM to a subset of events representative of the predicted event’s severity reduced the underestimation bias exhibited in high-impact events and the overestimation bias in low-impact ones. We used cross entropy (CE) to quantify the relatedness of weather variable distribution between the training dataset and the forecasted event.

Weather-related power outages affect millions of utility customers every year. Predicting storm outages with lead times of up to five days could help utilities to allocate crews and resources and devise cost-effective restoration plans that meet the strict time and efficiency requirements imposed by regulatory authorities. In this study, we construct a numerical experiment to evaluate how weather parameter uncertainty, based on weather forecasts with one to five days of lead time, propagates into outage prediction error. We apply a machine-learning-based outage prediction model on storm-caused outage events that occurred between 2016 and 2019 in the northeastern United States. The model predictions, fed by weather analysis and other environmental parameters including land cover, tree canopy, vegetation characteristics, and utility infrastructure variables exhibited a mean absolute percentage error of 38%, Nash–Sutcliffe efficiency of 0.54, and normalized centered root mean square error of 68%. Our numerical experiment demonstrated that uncertainties of precipitation and wind-gust variables play a significant role in the outage prediction uncertainty while sustained wind and temperature parameters play a less important role. We showed that, while the overall weather forecast uncertainty increases gradually with lead time, the corresponding outage prediction uncertainty exhibited a lower dependence on lead times up to 3 days and a stepwise increase in the four- and five-day lead times.

Understanding and quantifying the impact of severe weather events on the electric transmission and distribution system is crucial for ensuring its resiliency in the context of the increasing frequency and intensity of extreme weather events caused by climate change. While weather impact models for the distribution system have been widely developed during the past decade, transmission system impact models lagged behind because of the scarcity of data. This study demonstrates a weather impact model for predicting the probability of failure of transmission lines. It builds upon a recently developed model and focuses on reducing model bias, through multi-model integration, feature engineering, and the development of a storm index that leverages upon distribution system data to aid the prediction of transmission risk. We explored three methods for integrating machine learning with mechanistic models. They consist of: (a) creating a linear combination of the outputs of the two modeling approaches, (b) including fragility curves as additional inputs to machine learning models and (c) developing a new machine learning model that uses the outputs of the weather-based machine learning model, fragility curve estimates and wind data to make new predictions. Moreover, due to the limited number of historical outages in transmission networks, a storm index was developed leveraging a dataset of distribution outages to learn about storm behavior to improve model skills. In the current version of the model, we substantially reduced the overestimation in the sum of predicted values of transmission line probability of failure that was present in the previously published model by 10 times. This has led to a reduction of model bias from 3352% to 14.46-15.43%. The model with the integrated approach and storm index demonstrates substantial improvements in the estimation the probability of failure of transmission line and their ranking by risk level. The improved model is able to capture 60% of the outages within the top 22.5% of the ranked power lines, compared to a value of 34.9% for the previous model. With an estimate of the probability of failure of transmission lines ahead of storms, power system planning and maintenance engineers will have critical information to make informed decisions to create better mitigation plans and minimize power disruptions. Long term, this model can assist with resilience investments as it highlights areas of the system more susceptible to damage.