Research & perspectives
← All research

Ruzizi I · Scientific Reports · 2026

When more water hides a hydropower problem

At Ruzizi I, rising efficiency tells only part of the story. A study of 24 years of operational records shows how increasing river discharge can mask declining hydraulic head and emerging operational constraints.

Read the published paper ↗
Study period
2000–2023
Installed capacity
29.8 MW
Setting
Downstream of Lake Kivu

Is higher output a sign of a healthier plant?

Hydropower depends on both the amount of water passing through a plant and the height difference that gives that water its energy. That height difference is called hydraulic head. More water can support generation even while the available head declines. We examined Ruzizi I, a 29.8 MW plant downstream of Lake Kivu, to understand what this combination means for its operating efficiency.

Looking beyond electricity output

The study combines monthly operational records from SNEL for January 2000 to December 2023 with ERA5-Land climate data. Trend analysis, comparisons of drought and wet years, linear regression, Random Forest, and Gradient Boosting were used to examine efficiency and its relationship with flow, head, and plant availability. The efficiency measure compares electrical output with the hydraulic energy input at each monthly timestep.

What we found

Higher flow can conceal a loss of head

The paper reports an efficiency increase of about 3.6 percentage points per decade alongside a gross-head decline of about 0.20 metres per decade. Discharge dominated the observed monthly efficiency relationship. The authors interpret this as a masking effect: better headline performance does not mean hydraulic constraints have disappeared.

Lake Kivu changes the climate response

Efficiency during drought years was not statistically distinguishable from normal years, while wet years showed higher efficiency. The paper attributes this pattern to upstream storage in Lake Kivu. This is a historical finding for this lake-buffered plant, not a claim that hydropower is immune to drought or future climate change.

Operations offer a possible improvement

An observed load-factor band of 78–82% was associated with higher average efficiency. The study estimates a potential gain of roughly four percentage points relative to historical operation, but recommends testing it with real-time optimisation simulations. Declining available capacity also points to constraints beyond water supply.

The observed efficiency record

Two plots of Ruzizi I efficiency from 2000 to 2023: monthly values on the left and a red 12-month moving average on the right, with substantial fluctuations and higher values toward the end of the record.

Each blue point represents a monthly efficiency value. The red curve in the right-hand panel is a 12-month moving average. The figure shows substantial variability within a longer-term increase; it does not establish the causes of that trend on its own.

Figure 3, Mugisho et al. (2026). © The authors. Reproduced unchanged; display size adjusted. CC BY 4.0 · Source paper

My contribution

As a co-author, I contributed to the study design, methodology, data curation, formal analysis, software, visualisation, and validation. I led the original manuscript draft and contributed to review and editing, as recorded in the paper’s author-contribution statement.

What this study can—and cannot—tell us

This is an analysis of one plant’s historical monthly records. The operational data cannot precisely separate sediment deposition, hydrological changes, and human regulation as causes of head loss. The modelling describes relationships in this dataset, rather than proving causal effects or forecasting future performance. The proposed operating range should be validated before being treated as an operational prescription.

Publication

Mugisho, M. J., Ahana, B. S., Posite, V. R., Ngayirwa, S., Mirindi, D., Mirindi, F., Abdelbaki, C. & Kumar, N. (2026). The efficiency paradox of discharge masking head loss in run-of-river hydropower generation. Scientific Reports, 16, 3048.