# Gurara Reservoir Surface-Area Scenarios Through 2030

## Scope

This analysis projects monthly visible surface-water area from May 2026 through
December 2030. Months 1-12 are near-term forecasts with moderate confidence.
Months 13-56 are low-confidence climate-conditioned scenarios, not deterministic
forecasts.

The projected variable is satellite-visible water area, not storage volume,
water level, inflow, or release.

## Methods

- A SARIMA(1,1,1)(1,1,1,12) model provides the 56-month temporal baseline.
- Five NASA NEX-GDDP-CMIP6 models under SSP2-4.5 and SSP5-8.5 provide future
  rainfall and temperature.
- CMIP6 precipitation and temperature were monthly bias-corrected against
  CHIRPS and TerraClimate observations over 2001-2014.
- Potential evapotranspiration was estimated with Hargreaves-Samani and
  ratio-corrected against observed TerraClimate PET climatology.
- A Ridge model fitted to the post-impoundment period converts climate
  differences into surface-area adjustments.
- Monte Carlo simulation combines SARIMA predictive uncertainty with spread
  across ten CMIP6 model-pathway members.
- Every result is constrained to 0-55.84 km2.

The climate-conditioned Ridge model used 208
monthly training rows and obtained a time-series cross-validation RMSE of
6.80 km2.

## Documented Scenario Assumptions

| scenario   |   rainfall_multiplier |   pet_multiplier |   temperature_delta_c |
|:-----------|----------------------:|-----------------:|----------------------:|
| central    |                  1.00 |             1.00 |                  0.00 |
| dry        |                  0.85 |             1.10 |                  1.00 |
| wet        |                  1.15 |             0.95 |                 -0.50 |

Dry and wet adjustments are stress-test assumptions applied to each
bias-corrected CMIP6 member. They are not probabilities and should not be
interpreted as separate emissions pathways.

## December 2030 Outcomes

| scenario   |   projected_area_km2 |   lower_80_km2 |   upper_80_km2 |   lower_95_km2 |   upper_95_km2 | confidence_label   |
|:-----------|---------------------:|---------------:|---------------:|---------------:|---------------:|:-------------------|
| central    |                55.84 |          31.48 |          55.84 |          17.69 |          55.84 | low                |
| dry        |                54.80 |          27.98 |          55.84 |          14.76 |          55.84 | low                |
| wet        |                55.84 |          35.01 |          55.84 |          20.64 |          55.84 | low                |

![Monthly scenarios and uncertainty](../figures/09_scenarios_through_2030.png)

## Annual Summaries

| scenario   |   year |   month_count |   projected_area_mean_km2 |   projected_area_min_km2 |   projected_area_max_km2 |   annual_mean_lower_95_km2 |   annual_mean_upper_95_km2 |
|:-----------|-------:|--------------:|--------------------------:|-------------------------:|-------------------------:|---------------------------:|---------------------------:|
| central    |   2026 |             8 |                     47.75 |                    41.23 |                    55.41 |                      30.32 |                      55.58 |
| central    |   2027 |            12 |                     48.55 |                    40.01 |                    55.23 |                      25.98 |                      55.84 |
| central    |   2028 |            12 |                     48.96 |                    40.31 |                    55.75 |                      21.60 |                      55.84 |
| central    |   2029 |            12 |                     48.72 |                    39.06 |                    55.84 |                      16.96 |                      55.84 |
| central    |   2030 |            12 |                     50.12 |                    41.54 |                    55.84 |                      13.22 |                      55.84 |
| dry        |   2026 |             8 |                     46.39 |                    40.72 |                    53.18 |                      28.92 |                      55.52 |
| dry        |   2027 |            12 |                     46.60 |                    38.52 |                    52.91 |                      24.03 |                      55.84 |
| dry        |   2028 |            12 |                     46.41 |                    37.71 |                    53.14 |                      18.89 |                      55.84 |
| dry        |   2029 |            12 |                     46.41 |                    36.95 |                    53.78 |                      14.41 |                      55.84 |
| dry        |   2030 |            12 |                     47.94 |                    39.44 |                    54.80 |                      10.70 |                      55.84 |
| wet        |   2026 |             8 |                     48.24 |                    41.16 |                    55.84 |                      30.79 |                      55.54 |
| wet        |   2027 |            12 |                     49.61 |                    41.00 |                    55.84 |                      27.04 |                      55.84 |
| wet        |   2028 |            12 |                     49.78 |                    41.19 |                    55.84 |                      22.71 |                      55.84 |
| wet        |   2029 |            12 |                     49.71 |                    40.36 |                    55.84 |                      18.01 |                      55.84 |
| wet        |   2030 |            12 |                     50.90 |                    42.47 |                    55.84 |                      14.90 |                      55.84 |

The 2026 annual row covers May-December only. Later annual rows contain all
twelve months.

![Annual scenario summaries](../figures/10_annual_scenarios_through_2030.png)

## Uncertainty Analysis

|   horizon_months | date     |   projected_area_km2 |   width_80_km2 |   width_95_km2 | confidence_label   |
|-----------------:|:---------|---------------------:|---------------:|---------------:|:-------------------|
|                1 | May 2026 |                41.23 |          16.39 |          24.65 | moderate           |
|               12 | Apr 2027 |                40.01 |          28.18 |          37.52 | moderate           |
|               24 | Apr 2028 |                40.31 |          32.86 |          41.94 | low                |
|               36 | Apr 2029 |                39.06 |          36.52 |          47.63 | low                |
|               48 | Apr 2030 |                41.54 |          37.78 |          51.09 | low                |
|               56 | Dec 2030 |                55.84 |          24.36 |          38.14 | low                |

The 95% interval expands substantially as the projection horizon increases.
The model therefore provides more information about seasonal direction and
relative scenario differences than about an exact water area several years
ahead.

![Uncertainty growth](../figures/12_uncertainty_growth.png)

## Seasonal Analysis

| scenario   | lowest_month   |   lowest_mean_km2 | highest_month   |   highest_mean_km2 |   seasonal_range_km2 |
|:-----------|:---------------|------------------:|:----------------|-------------------:|---------------------:|
| central    | April          |             40.23 | December        |              55.58 |                15.35 |
| dry        | April          |             38.16 | November        |              53.54 |                15.38 |
| wet        | April          |             41.25 | November        |              55.84 |                14.58 |

Across all scenarios, the projected seasonal minimum occurs in April. The
projected area then increases through the wet season and reaches its largest
values around October-January. The dry scenario lowers the seasonal curve
throughout the year rather than changing the timing of the seasonal cycle.

![Projected seasonal cycle](../figures/13_projected_seasonal_cycle.png)

## Scenario Sensitivity

| comparison        |   mean_difference_km2 |   largest_absolute_difference_km2 |
|:------------------|----------------------:|----------------------------------:|
| Dry minus central |                 -2.12 |                              3.12 |
| Wet minus central |                  0.85 |                              2.13 |

Scenario separation remains materially smaller than the long-horizon 95%
uncertainty interval. This means the stress tests are useful for comparing
directional climate effects, but they should not be presented as sharply
separated deterministic futures.

![Scenario differences](../figures/14_scenario_differences.png)

## Climate Ensemble Analysis

|    year |   rainfall_median_mm |   rainfall_min_mm |   rainfall_max_mm |   pet_median_mm |   water_balance_median_mm |   temperature_median_c |
|--------:|---------------------:|------------------:|------------------:|----------------:|--------------------------:|-----------------------:|
| 2026.00 |              1510.25 |            982.13 |           1969.44 |          974.45 |                    535.80 |                  24.91 |
| 2027.00 |              1681.20 |           1411.80 |           2087.61 |         1631.26 |                     32.02 |                  25.56 |
| 2028.00 |              1574.84 |           1417.51 |           2072.84 |         1637.13 |                    -98.17 |                  25.86 |
| 2029.00 |              1724.76 |           1388.72 |           2591.69 |         1622.88 |                     98.05 |                  26.01 |
| 2030.00 |              1684.26 |           1469.75 |           2293.73 |         1597.40 |                     76.91 |                  25.33 |

The ranges in annual rainfall and water balance show disagreement among the
five CMIP6 models and two SSP pathways. This ensemble disagreement is retained
in the projected area intervals rather than hidden behind a single climate
trajectory.

![Projected climate scenarios](../figures/11_projected_climate_scenarios.png)

![CMIP6 ensemble spread](../figures/15_cmip6_ensemble_spread.png)

![Projected climatic water balance](../figures/16_projected_water_balance.png)

## Planning Indicators

The following thresholds are descriptive planning indicators, not official dam
operating levels:

- Lower-area indicator: 70% of the validated footprint
  (39.09 km2).
- High-area indicator: 90% of the validated footprint
  (50.25 km2).

| scenario   |   months_below_39.1_km2 |   months_above_50.3_km2 |   months_low_confidence |
|:-----------|------------------------:|------------------------:|------------------------:|
| central    |                       1 |                      23 |                      44 |
| dry        |                       3 |                      23 |                      44 |
| wet        |                       0 |                      23 |                      44 |

These counts use scenario medians. They do not represent probabilities of
crossing a storage, flood, or water-supply threshold.

## Historical Context

![Observed record and projection](../figures/17_history_and_projection.png)

The projection begins from the April 2026 observation and retains the strong
seasonal cycle seen in the post-impoundment record. The widening band makes the
transition from observation to uncertain scenario explicit.

![Monthly scenario heatmap](../figures/18_scenario_heatmap.png)

## Integrated Interpretation

- Differences between dry, central, and wet median projections are smaller than
  the long-horizon uncertainty bands.
- Upper intervals frequently reach the validated 55.84 km2 footprint, while
  lower intervals widen substantially toward 2030.
- This means the model cannot support a precise shoreline or exact area claim
  several years ahead.
- The scenario medians remain useful for comparing climate stress assumptions,
  while the intervals communicate the much wider range of plausible outcomes.
- Seasonal timing is more stable than exact magnitude: all scenarios identify
  April as the low-area month and late-year months as the high-area period.
- CMIP6 model spread is a major source of uncertainty and should remain visible
  in decision-support products.

## Confidence and Limitations

1. May 2026-April 2027: moderate-confidence near-term forecast period.
2. May 2027-December 2030: low-confidence climate-conditioned scenario period.
3. Dam releases, abstractions, operating rules, inflow gauges, bathymetry, and
   measured storage were unavailable.
4. The configured Gurara study rectangle is used as a local catchment-climate
   proxy; it is not the complete published 2,150 km2 catchment boundary.
5. CMIP6 provides climate projections, not month-specific weather forecasts.
6. Spatial maps distribute projected area using historical water persistence;
   they are not future satellite images or exact shoreline forecasts.

## Outputs

- `outputs/projections_2030/gurara_monthly_scenarios_2026_2030.csv`
- `outputs/projections_2030/gurara_annual_scenarios_2026_2030.csv`
- `outputs/projections_2030/sarima_56_month_baseline.csv`
- `outputs/projections_2030/bias_corrected_cmip6_scenario_climate.csv`
- `outputs/figures/09_scenarios_through_2030.png`
- `outputs/figures/10_annual_scenarios_through_2030.png`
- `outputs/figures/11_projected_climate_scenarios.png`
- `outputs/figures/12_uncertainty_growth.png`
- `outputs/figures/13_projected_seasonal_cycle.png`
- `outputs/figures/14_scenario_differences.png`
- `outputs/figures/15_cmip6_ensemble_spread.png`
- `outputs/figures/16_projected_water_balance.png`
- `outputs/figures/17_history_and_projection.png`
- `outputs/figures/18_scenario_heatmap.png`