1. Sam Dixon, Department of Geography
Forecasting reservoir inflows using remotely sensed precipitation estimates: a pilot study for the River Naryn, Kyrgyzstan
Sam Dixon, Department of Geography
Source: Dixon and Wilby (2015)
Contact:

2. Literature Global flood forecasting systems (e.g. GloFAS)
Kyrgyzstan wide TRMM evaluation (Karaseva et al., 2012)
Assessments of climate change impacts on future river flows (Wilby et al., 2011; Sorg et al., 2012)
Seasonal forecasting of runoff for large basins within Central Asia (Baumgartner et al., 2000)
Looking North across Toktogul reservoir (Source: Panoramio, 2007)

3. Aims
To assess the accuracy of a remotely sensed precipitation estimate (TRMM) for the strategically significant Naryn river basin, Kyrgyzstan.
To investigate the potential for river flow forecasting based on remotely sensed precipitation, surface temperature and gauged discharge, over monthly to seasonal horizons.

4. Methodology Comparison of TRMM estimates to Naryn gauge observations
Tests for normality of data
All TRMM cells, including permutations of concurrent and lagged cells, as well as moving average correlated with monthly flow at Toktogul
Combinations of predictor variable, averaging period and lag interval used to develop multiple linear regression models of monthly inflow
Models’ skill assessed via cross validation

5. Results
Monthly total precipitation recorded by the gauge at Naryn compared with the nearest (top) 0.5° TRMM and (bottom) 0.25° TRMM cell
Correlation between nearest 0.5° TRMM estimates and gauge daily (top) and monthly (bottom) precipitation at Naryn
Source: Dixon and Wilby (2015)

6. Results
Correlation (r) of gauged flows with lagged predictors averaged over one-six months: temperature (T); precipitation measured at Naryn (PN); precipitation estimates from TRMM for the basin area (PA), and optimum location (PO). For n = 130 and at p = 0.05 significance level, rcrit = 0.17.
Source: Dixon and Wilby (2015)

7. Results Cross-validation results for ZOF, Q1, Q2 and Q3 models
Metric
ZOF Q1 Q2 Q3
R2adj (%) 81 89 85 84
Nash- Sutcliffe Coeff.
0.797
0.878
0.829
0.818
Cross-validation results for ZOF, Q1, Q2 and Q3 models
Cross-validated model forecasts with lead-time one (Q1), two (Q2) and three (Q3) months compared with long-term monthly mean discharge (ZOF).
Source: Dixon and Wilby (2015)

8. Conclusions and further study
Correlations between observed precipitation at Naryn and 0.5° TRMM totals are weaker for daily than monthly totals
>80% of the variance in monthly inflows is explained with three month lead, and up to 65% for summer half-year average
Plans to develop non-linear statistical or deterministic algorithms
Forecast tools for other environmental hazards affecting reservoir operations
Dixon, S.G. & Wilby, R.L. (2015) Forecasting reservoir inflows using remotely sensed precipitation estimates: a pilot study for the River Naryn, Kyrgyzstan, Hydrological Sciences Journal, DOI: