Statistical Seasonal Streamflow Forecasting

Purpose: Forecast ensemble seasonal streamflows
Developer: Academic, research and operational fraternities
Key Features: Uses past climate information; Compatible with different types of data; Generates forecasts at a few months lead time ahead
Latest Release: N/A
OS Platform: Code can be developed for different platforms
Cost: Free
Related Software:
Website: A statistical multi-model ensemble seasonal streamflow forecasting framework

Introduction

Seasonal precipitation is chiefly responsible for seasonal streamflows, such as monsoon rainfall- driven streamflows. However, rain being the form of precipitation and whatever rains it immediately becomes surface flow without being stored do not allow using precipitation in forecasting seasonal streamflows in advance. It is a known fact that complex nonlinear land-ocean-atmospheric feedbacks set conditions for seasonal precipitation and temperatures. These feedbacks, either regional or large-scale (e.g., ENSO, IOD) in nature do provide significant information on monsoon rainfall and consequently on seasonal streamflows in advance. Knowing the causes, i.e., either winter precipitation or regional or large-scale climate mechanisms, and its effect, i.e., seasonal streamflow, allow for statistical forecasting based techniques.

Statistical seasonal streamflow forecasting techniques involve forming and quantifying the relationship between seasonal streamflows (effect) and its influencing variables (causes), and then estimating streamflows from the relationship. Traditionally, statistical techniques issue a forecast as a single-valued or deterministic forecast, which lacks uncertainty and consequently misguides a decision maker. From the decision-making perspective, it is very important to know the uncertainty surrounding forecasts. Ensemble forecasts, also known as probabilistic forecasts, provide forecast uncertainty..

There are many ensemble-forecasting techniques available, e.g., non-parametric multi-model ensemble seasonal streamflow forecasting technique (See for technical details, Regonda et al., 2006). The technique (i) considers various hydroclimate and land-ocean-atmosphere variables (e.g., sea surface temperatures, snow, winds, pressure anomalies, humidity) at regional- and large-spatial scales (e.g., ENSO, IOD) for up to 12 months ago based on correlations with seasonal streamflows; (ii) forms multiple models based on different combinations of selected variables; (iii) fits a non-parametric model for all models; (iv) selects a suite of models that have similar performance; (v) generates ensembles of streamflow forecasts from each of the suite of models and then combines ensembles..

Being data driven, the technique has limitations similar to many other statistical techniques. Typically, statistical technique under (or often times poorly) perform for new extreme conditions that are not seen in the data used to calibrate the technique. Therefore, data sets of longer periods of the record, which comprise a variety of scenarios are preferred.

Advantages and Limitations

Advantages Limitations
  • Uses past climate information, without necessarily requiring future climate data
  • Has flexibility to use different types of data including public domain data, e.g., satellite estimated streamflows, reanalysis data
  • Easy to develop and transferable to other regions
  • Often times poorly performs for new extreme conditions that are not part of the calibration
  • Requires data sets of longer periods

Illustrative Screens

Modified boxplots of January 1st ensemble seasonal streamflow forecasts generated in cross-validation mode for Blue Nile River, Bile Nile River Basin. Source: World Bank

Sample Applications

Africa East Asia and the Pacific Europe & Central Asia Latin America & the Caribbean Middle East and North Africa South Asia

World Bank - All rights reserved.