Thamalakane Flow Forecast 29 April 2010

Disclaimer

This forecast is provided as is, without any guarantees as to skill or error. The forecast may be disseminated, displayed, or publicised without the explicit permission of the author, however, with proper acknowledgement of the source (Prepared by Piotr Wolski, available at: www.orc.ub.bw/datacat/thamalakane2010.php)

Important

The flow forecast for Thamalakane River (unlike the one for the Okavango River presented elsewhere at this site) is based on observed data: Okavango River discharges at Mohembo and rainfall measurements at stations belonging to the Department of Meteorological Services, located in Maun and in Shakawe.

These data form input to hydrological models, the output of which is presented in the graphs below.

Two models are used here: a regression model; and a semi-empirical reservoir model of the Okavango Delta, a.k.a. HOORC model.

Models are just models - they are not perfect. In the Okavango they have to simulate a range of conditions present in a rather complex system, and this has to be done using very limited data. The performance of these models has to be judged by looking at the entire period of observations/simulations, not just at a particular year. Ultimately, for practical purposes the output of simulations has to be considered as indicating an order of magnitude of forecast variables.

95 % confidence interval expresses errors associated with the model. It essentially says that, accounting for the error, there is a 95% chance that the actual value of the forecasted variable will fall within that interval. There is, however, a 5 % chance that the value will be outside of that interval.

Even more important

The forecasting excercise indicates that a very large flood is expected in Maun this year. Discharges in excess of 100 Mm³/month (or approximately 40 m³/s) are forecasted.

Discharges in Thamalakane River do not translate to water levels in a straightforward manner. We have not yet calculated what water levels are to be expected, but by analogy they will be similar to these present in the mid 1970s. For reference, we present a figure showing Thamalakane water levels in the 1970-2009 period. Within the accuracy of the forecast, water levels in the order of 50-100 cm higher than the maximum water level recorded in 2009 are expected.

The confidence in the forecast is fairly high. Two independent, and in principle different, models converge indicating similar total annual flow in 2010. Although the margins of error of the models are large, the simulated values are well in excess of these margins.

Observed and simulated total annual discharge of Thamalakane River at Maun Bridge

Observed and simulated (HOORC model) monthly discharges of Thamalakane River at Maun Bridge for the period 1970-2010

Observed and simulated (HOORC model) monthly discharges of Thamalakane River at Maun Bridge for the period 1970-2010

Observed water levels of Thamalakane River at Maun Bridge for the period 1970-2009

And not so important...

How does the regression model work?

The Regression model analyses the relationship between flows of the Thamalakane River at Maun (or so called dependent variable) and factors that potentially influence these flows (or independent variables). These variables have been identified as: inflow at Mohembo, rainfall at Maun and the discharge of the Thamalakane in the previous year. In principle, a relationship between dependent and independent variable(s) is developed based on the observed time series of these variables. Knowing independent variables and the coefficients describing these relationships allows prognostication of the value of the dependent variable. Several versions of a regression model for Thamalakane discharges have been done in the last 25 years. The version used here is a bit more complicated than the earlier ones, but its detailed description is beyond the scope of this page. The model is based on annual data and considers total annual inflow at Mohembo, average of total annual rainfall at Maun and Shakawe, and the previous year's total annual outflow at Maun. For 2010 these values are taken as 12400 Mm³, 890 mm and 184 Mm³ respectively. The parameters of the model have been derived (or the model has been calibrated) using values of dependent and independent variables observed during the 1969-2009 period.

The prognosis presented here is done only in April 2010, and thus neither total annual inflow (which is that between November 2009 and October 2010) nor total annual rainfall (between July 2009 and June 2010) are known at this stage. However, we have made some educated guesses for inflow and rainfall during the rest of the year based on the analysis of past records. Importantly, whether or not these educated guesses will be correct does not matter much for the final output of the model. This is because the bulk of water available from inflow and rainfall has been already 'delivered' and eventual errors in our estimates contribute little to the error of the final result.

How does the HOORC model work?

HOORC model is a rather complex one, and it is in details described in Wolski et al. 2006. The model treats the Delta as an array of hydrological units. It simulates the flow of water and the processes affecting water balances in these units (evaporation, transpiration, infiltration to groundwater), fully accounting for all water present in the system. It uses inflows at Mohembo, rainfall and temperatures measured at Maun and Shakawe as inputs, and produces a range of outputs, including Thamalakane River flows at Maun.

The model has been calibrated for the period 1969-2002. So the flows after 2002, including the forecast discharges in 2010, are effectively 'predicted' by it. The performance (or agreement of observed and simulated discharges) of the model during 2003-2009 can thus be considered a measure of how good the forecast for 2010 is.

Acknowledgements

The forecast is based on data provided by Department of Water Affairs and Department of Meteorological services.