Gurara Reservoir · Kaduna State, Nigeria

Twenty-five years of water change, made visible.

A satellite-led evidence platform for understanding Gurara Reservoir’s surface-water history, seasonal behaviour and climate-conditioned outlook.

Explore the evidence Read the manuscript

Study record: 2001–2026
Valid observations: 196 months
Forecast horizon: To 2030

Map showing the location and footprint of Gurara Reservoir

Study area55.84 km² recurring-water footprint

Latest observationApril 2026

36.81 km²visible surface-water area

Above normal34.9% above the April mean

Area is not storage volume.

The evidence

A reservoir shaped by formation, season and climate.

The record shows a significant long-term increase, but the dominant structural change is the reservoir’s formation after 2001—not climate change alone.

Long-term trend+1.82km² per year · Sen slope

Seasonal lowApril27.29 km² historical mean

Seasonal highAugust43.91 km² historical mean

Preferred indexMNDWI3.46 km² cross-sensor MAE

Historical record

Monthly surface-water area

Monthly Gurara Reservoir surface-water area from 2001 to 2026

Plain-language finding

“The reservoir has grown, but the record is also the story of its creation.”

Any interpretation of the positive trend must account for the transition from a natural drainage basin to an impounded water body.

Annual summary

Observed change by year

Annual mean and observed range. The long-term trend includes the reservoir-formation period.

Seasonal pattern

When water extent is highest

April has the lowest historical mean; August has the highest. Monthly sample counts differ because of cloud filtering.

Observe the change

One frame for every year.

The annual Landsat sequence reveals the reservoir forming and then settling into a cycle of seasonal expansion and contraction.

20012026

Annual Landsat animation of Gurara Reservoir from 2001 to 2026 — Annual composites · Landsat 5, 7, 8 and 9 ·

Climate relationships

Association is evidence—not causation.

Same-month climate relationships are exploratory. Reservoir operations, abstraction, formation history and delayed runoff responses can also influence water extent.

Correlation analysis

Climate and surface-water area

Model interpretation

Random Forest feature importance

Permutation importance varies by forecast horizon and should be interpreted cautiously because tree-model test skill was limited.

Model evaluation

Complex AI did not automatically win.

Persistence, seasonal persistence, SARIMA, Random Forest and XGBoost were evaluated chronologically. Simple models remained competitive at longer horizons.

Test-period performance

Best model by horizon

Horizon	Selected model	RMSE	MAE	NSE	KGE
1 month	SARIMA	8.014	5.393	0.075	0.516
3 months	SARIMA	9.648	7.011	−0.341	0.317
6 months	Seasonal persistence	9.836	5.029	−0.292	0.088
12 months	Persistence	9.836	5.029	−0.292	0.088

Values are in km² where applicable. Negative NSE indicates performance below a test-period mean benchmark.

Forecast and scenarios

Near-term guidance, long-term uncertainty.

SARIMA provides the strongest short-horizon result. Beyond April 2027, projections should be read as planning scenarios rather than precise forecasts.

Forecast peak55.64 km²December 2026

Forecast endpoint40.94 km²April 2027

Best short modelSARIMA8.01 km² test RMSE

Selected horizons

Operational forecast values

Horizon	Month	Area	Lower 95%	Upper 95%
1	May 2026	41.890	29.449	54.331
3	Jul 2026	42.992	26.009	55.837
6	Oct 2026	53.645	34.536	55.837
12	Apr 2027	40.936	19.735	55.837

Mapped uncertainty

A plausible spatial extent—not an exact shoreline

Dark blue shows the lower 95% extent, cyan the central projection and light blue the upper 95% extent.

Animated monthly Gurara forecast extent scenarios

Climate-conditioned outlook

Explore scenarios through 2030

The central case holds climate stress multipliers unchanged and provides the core planning trajectory.

2030 annual mean: 50.12 km²
Rainfall adjustment: No change
Temperature adjustment: No change

Central climate scenario animation for Gurara Reservoir

Projection atlas

Uncertainty, seasonality and climate-model spread

Five bias-corrected NEX-GDDP-CMIP6 models under SSP2-4.5 and SSP5-8.5 inform these scenarios. Results after April 2027 are low-confidence planning scenarios.

Monthly scenarios through 2030

Central 80% and 95% intervals widen substantially with horizon.

Annual scenario means

Projected climate inputs

Uncertainty growth

Projected seasonal cycle

Scenario differences

Scenario heatmap

CMIP6 ensemble spread

Projected water balance

History and projection

The observed record and central projection share one view to make the forecast origin explicit.

Scenario definitions

Documented stress tests

Scenario	Rainfall	PET	Temperature
Central	1.000	1.000	0.0 °C
Dry	0.850	1.100	+1.0 °C
Wet	1.150	0.950	−0.5 °C

December 2030

Scenario outcomes

Scenario	Median	Lower 80%	Upper 80%	Lower 95%	Upper 95%
Central	55.837	31.481	55.837	17.695	55.837
Dry	54.797	27.983	55.837	14.763	55.837
Wet	55.837	35.005	55.837	20.639	55.837

Research design

From satellite pixels to decision evidence.

A reproducible workflow connects Earth observation, climate records, validation, forecasting and uncertainty communication.

01
Define
Derive a recurring-water reservoir footprint from JRC Global Surface Water.
02
Observe
Extract monthly Landsat and Sentinel-2 surface-water area.
03
Validate
Compare water indices and retain MNDWI using cross-sensor agreement.
04
Model
Benchmark statistical and machine-learning forecasts chronologically.
05
Communicate
Publish forecasts, uncertainty and explicit limits for decision support.

Cross-sensor validation

Why MNDWI was retained

Across 73 valid Landsat–Sentinel-2 overlap months, MNDWI produced MAE 3.463 km², RMSE 6.838 km² and bias −0.967 km².

Research boundaries

Clear limits make the evidence more useful.

The platform reports what the implemented evidence supports and avoids claims the available observations cannot validate.

Area is not volume

Depth, bathymetry, sedimentation, releases, abstraction and measured storage were unavailable.

Validation is satellite-based

Cross-sensor agreement improves confidence but does not replace field observations or labelled ground truth.

Optical data have gaps

Cloud and coverage filters rejected 35.5% of monthly Landsat records.

Long forecasts are uncertain

Longer-horizon skill was weak and uncertainty appropriately widens through 2030.

No deep-learning claim

Only 196 months passed quality control and the longest continuous valid run was 23 months.

Scenarios are not weather

CMIP6 projections support stress testing; they do not predict exact monthly conditions or shorelines.

Research paper

Multi-Sensor Earth Observation and Machine Learning for Monitoring and Forecasting Reservoir Surface Water Dynamics: A Case Study of Gurara Reservoir, Nigeria

Meet the researchers behind the Gurara Reservoir study and access the complete manuscript for the full methodology, analysis, results and discussion.

Read the manuscript Research files

Alamin Musa Magaga
Daberechi Okorie
Yazid S Mikail
Daniel Akanbi
Meron Abate
Abubakar Isa-Abubakar
Nimatallahi Masuud

Open research

Inspect the evidence yourself.

Download the reviewed manuscript, operational datasets and decision-support reports used in this application.

PDF· ManuscriptResearch manuscript↗ CSV · Historical dataQuality-controlled analysis dataset↗ CSV · Forecast12-month operational outlook↓ CSV · ScenariosMonthly scenarios through 2030↓ CSV · ScenariosAnnual scenario summaries↓ JPG · Historical imageryAnnual reservoir contact sheet↓ JPG · Forecast imageryForecast extent contact sheet↓ GIF · Historical imageryAnnual Landsat timelapse↓ GIF · Forecast imagery12-month forecast extent animation↓ GIF · Central scenarioCentral spatial scenario animation↓ GIF · Dry scenarioDry spatial scenario animation↓ GIF · Wet scenarioWet spatial scenario animation↓ MD · Technical reportFinal research report↓ MD · BriefingDecision-support report↓ MD · Technical report2030 projection report↓