ssembatya-etal2025TBD

Analyzing the Relative Influences of Hydrologic Information and Dams Hydropower Scheduling Decisions on Electricity Price Forecasts

Henry Ssembatya¹, Jordan D. Kern^1*, Nathalie Voisin^2,3*, Scott Steinschneider ⁴, and Daniel Broman²

¹ North Carolina State University, Raleigh, NC, USA
² Pacific Northwest National Laboratory, Richland, WA, USA
³ University of Washington, Seattle, WA, USA
⁴ Cornell University, Ithaca, NY, USA

* Corresponding authors: jkern@ncsu.edu, Nathalie.voisin@pnnl.gov

Abstract

Hydropower operators use hydrologic forecasts to better manage water releases and reservoir storage, including hydropower generation schedules. In this paper, we examine whether short- to medium-range hydrologic forecast errors are a significant driver of errors in electricity price forecasts. Week-to-week changes in reservoir inflows influence the supply of low marginal cost hydropower, which can, in turn, impact market prices. Using softly coupled hydrologic, hydropower scheduling, and power systems models representing the operations of 267 dams spanning the U.S. Western Interconnection, we quantify the importance of short- to medium-range hydrologic forecast accuracy in correctly forecasting wholesale electricity prices. As a point of comparison, we also quantify the influence of dam operators own hourly production decisions on realized market prices. We find that dams revenue-maximizing behavior (i.e., scheduling generation to align with the periods of high forecasted prices) causes larger magnitude deviations from price forecasts than hydrologic forecast errors. Our findings suggest that optimization of hydropower scheduling should anticipate price effects due to the production decisions of hydropower facilities.

Journal reference

Ssembatya, H., Kern, J. D., Voisin, N., Steinschneider, S., & Broman, D. (2025). Analyzing the Relative Influences of Hydrologic Information and Dams Production Decisions on Electricity Price Forecasts. Submitted to Applied Energy

Code reference

Ssembatya, H., Kern, J. D., Voisin, N., Steinschneider, S., & Broman, D. (2025). Supporting code for Ssembatya et al. 2025

Data references

Input data

| Dataset | Repository Link | DOI | |:-----------------------------------------------:|:---------------------------------------------------------------:|:----------------------------------------:| | STARFIT Output files | xx | xx | | FIScH Output | https://github.com/HydroWIRES-PNNL/broman-etal2025wrr | xx |

Output data

| Dataset | Repository Link | DOI | |:----------------------------------------------------------:|-----------------------------------------------:|:-------------------------------------------------:| | DCOPF1 (Perfect, Persistence) LMPs and generation | https://zenodo.org/uploads/14041719 | https://doi.org/10.5281/zenodo.14041719 | | DCOPF2 (Perfect, Persistence) LMPs and generation | https://zenodo.org/uploads/14041719 | https://doi.org/10.5281/zenodo.14041719 |

Contributing modeling software

| Model | Version | Repository Link | DOI | |:--------:|:-------:|:--------------------------------------------------:|:---:| | FIScH | v0.4.0 | https://github.com/HydroWIRES-PNNL/fisch | NA | | GO-WEST | NA | https://github.com/romulus97/IM3-GO-WEST | NA | | starfit | v0.1.0 | https://github.com/IMMM-SFA/starfit | NA |

Reproduce my experiment

Clone this repository to get access to the scripts used in the experiment. Run the 4 instances of the DCOPF model ("DCOPF1 Perfect", "DCOPF1 Persistence", "DCOPF2 Perfect", "DCOPF2 Persistence"). Use the results of LMPs and generation from the 4 DCOPF runs, as well as outputs from FIScH to analyze the trends in changes in price forecasts errors corresponding to changes in streamflow forecast or the DP model's (FIScH) optimal scheduling.

*Note that the nomenclature "GO1" is used interchangeably with "DCOPF1". *Note that the nomenclature "GO2" is used interchangeably with "DCOPF2".
*Note that the nomenclature "Delta1" is used interchangeably with "flow".
*Note that the nomenclature "Delta2perfect" is used interchangeably with "schedulingperfect".
*Note that the nomenclature "Delta2persistence" is used interchangeably with "scheduling_persistence".

| Script Number | Script Name | Purpose | | --- | --- | --- | | 1 | deltas_v2_github.py | Computing "eflow" and "escheduling" based on LMPs ("duals") | | 2 | standard_dev_github.py | Computing the standard deviation of "eflow" and "escheduling" |

Reproduce my figures

Use the following scripts to reproduce figures used in this publication.

| Figure Numbers | Script Name | Description | |:--------------:|:-------------------------------------------------------:|:------------------------------------------------------------------------------------------:| | 2 | distribution_plot_onlyperfect_violin_manuscript.py | Comparing the differences in LMPs across all hours, nodes, and years. | | 3 | jointplot_manuscript_github.py | Comparing the standard deviation for "eflow" and "escheduling" across nodes. Also plotted against annual hydro generation. | | 4 | seasonal_std_avg_deltas_github.py | Comparing the standard deviation for "eflow" and "escheduling" by year and season. | | 5 | std_avg_plot_onlyperfect_github.py | Visualizing the standard deviation for "eflow" and "escheduling" by node. | | 6 | revenue_impactsv2_github.py | How changes in price errors affect revenues by year. | | 7 | 2000_percentage_plot_wecc_delta2pf_revdistr_github.py | Plotting the percentage changes in revenue by node |