Validating quantitative precipitation forecast for the Flood Meteorological Office, Patna region during 2011–2014

Abstract

In order to issue an accurate warning for flood, a better or appropriate quantitative forecasting of precipitation is required. In view of this, the present study intends to validate the quantitative precipitation forecast (QPF) issued during southwest monsoon season for six river catchments (basin) under the flood meteorological office, Patna region. The forecast is analysed statistically by computing various skill scores of six different precipitation ranges during the years 2011–2014. The analysis of QPF validation indicates that the multi-model ensemble (MME) based forecasting is more reliable in the precipitation ranges of 1–10 and 11–25 mm. However, the reliability decreases for higher ranges of rainfall and also for the lowest range, i.e., below 1 mm. In order to testify synoptic analogue method based MME forecasting for QPF during an extreme weather event, a case study of tropical cyclone Phailin is performed. It is realized that in case of extreme events like cyclonic storms, the MME forecasting is qualitatively useful for issue of warning for the occurrence of floods, though it may not be reliable for the QPF. However, QPF may be improved using satellite and radar products.

Appendix

The simplest forecast can be categorically either ‘yes’ or ‘no’ forecast of occurrence of a phenomenon like a rain event. In order to understand the verification measures used for forecast evaluation let us consider a contingency table as follows:

ᅟ ᅟ

In this table, X1 indicates the case when the event (the flood in this case) is actually observed (observed occurrence) and also predicted to occur (forecast occurrence). Y1 indicates an event which is not observed in reality (observed non-occurrence) whereas it is predicted to occur. When an event is predicted not to occur (forecast non-occurrence) whereas it occurs in reality then it is classified as X2 otherwise (observed non-occurrence) it is denoted by Y2. In view of this, the total number of forecasts generated is given by

$$\text{TF} = \mathrm{X}1 + \mathrm{X}2 + \mathrm{Y}1 + \mathrm{Y}2.$$

Now, the probability of detection (POD) can be written as:

$$\text{POD}=\frac{\mathrm{X}1}{\mathrm{X}1+\mathrm{X}2}. $$

False alarm rate (FAR) is given by the following equation:

$$\text{FAR}=\frac{\mathrm{Y}1}{\mathrm{X}1+\mathrm{Y}1}. $$

Missing rate (MR) is calculated as:

$$\text{MR}=\frac{\mathrm{X}2}{\mathrm{X}1+\mathrm{X}2}. $$

Similarly, the values of correct non-occurrence of C-NON can be computed as:

$$\mathrm{C}\textnormal{-}\text{NON} =\frac{\mathrm{Y}2}{\mathrm{Y}1+\mathrm{Y}2}. $$

The critical success index (CSI) is given by:

$$\text{CSI}=\frac{\mathrm{X}1}{\mathrm{X}1+\mathrm{X}2+\mathrm{Y}1}. $$

Bias for occurrence (abbreviated as BIAS) of the event can be computed as:

$$\text{BIAS}=\frac{\mathrm{X}1+\mathrm{Y}1}{\mathrm{X}1+\mathrm{X}2}. $$

Percentage correct (PC) values can be computed as:

$$\text{PC}=\frac{\mathrm{X}1+\mathrm{Y}2}{\text{TF}}\times 100\% . $$

True skill score (TSS) can be determined by the following formulation:

$$\text{TSS}=\frac{\mathrm{X}1}{\text{X1}+\text{X2}}+\frac{\mathrm{Y}2}{\text{Y1}+\text{Y2}}-1. $$

Similarly, another skill score known as Heidke skill score (HSS) can be computed in a relatively complex manner as follows:

$$\begin{array}{@{}rcl@{}} \text{HSS}=\frac{2\left\{ {\left( {\mathrm{X}1} \right)\left( {\mathrm{Y}2} \right)-\left( {\mathrm{X}2} \right)\left( {\mathrm{Y}1} \right)} \right\}} {\begin{array}{l} \left( {\mathrm{X}2} \right)^{2}+\left( {\mathrm{Y}1} \right)^{2}+2\left( {\mathrm{X}1} \right)\left( {\mathrm{Y}2} \right)\\ ~~~+\left( {\mathrm{X}2+\mathrm{Y}1} \right)\left( {\mathrm{X}1+\mathrm{Y}2} \right)\end{array}}. \end{array} $$

(1)