The study aims to enhance the accuracy of the European Centre for Medium-Range Weather Forecasts (ECMWF) and Global Ensemble Forecast System (GEFS) reference evapotranspiration forecast at short to medium range (1-7 days) using the post-processing methods: Analog technique (AN) and Simple Linear Regression (LR) over the Indian subcontinent. The FAO, Penman-Monteith (PM) equation, is used for the estimation of reference evapotranspiration (ET0) reforecasts from meteorological reforecasts from ECMWF and GEFS models. The post-processing technique AN and LR was applied to the ET0 reforecasts and compared against the ET0 estimated using observed and reanalysis dataset. The deterministic evaluation metrics, such as Root Mean Square Error (RMSE) and Correlation Coefficient (R), were used for the performance assessment of raw ET0 forecast and post-processed ET0 forecasts. Results showed that short to medium range ET0 forecasts improved substantially using AN and LR post-processing methods over the Indian region. Assessment across the different climatic zones in India showed that raw and post-processed ET0 forecasts in the Tropical climate zone are more skillful than in the other climatic zones. A comparison of raw and post-processed ET0 forecasts across different seasons in India showed that model forecasts are more skillful during the winter season compared to the rest. Intercomparison of the models also show that overall the raw and post-processed ET0 forecasts from ECMWF are better than GEFS. Results emphasize the use of post-processing methods to enhance the skill of ET0 forecasts over the Indian subcontinent before their application in irrigation scheduling and water demand estimation purposes.
This study aims to enhance the accuracy and reliability of the Global Ensemble Forecast System’s (GEFS) precipitation forecasts over the Indian subcontinent using two post-processing techniques, namely the Analog method (AN) and Logistic Regression method (LR). The post-processing techniques and GEFS Numerical Weather Prediction Model (NWP) outputs were evaluated against the observed dataset using probabilistic and deterministic evaluation metrics. Results found that both the methods considerably improves short range to medium range (1-15 day) precipitation forecasts over India. Overall results showed that both the methods perform poorly during the monsoon seasons compared to other seasons. Basin analysis showed that both the methods underperform in the Western Ghats, while the performance is comparable and decent in other parts of India. Analysis of precipitation at different terciles showed that both the AN and LR methods underperforms at higher terciles compared to the lower ones. This is because the GEFS model itself was performing poorly in detecting the heavy precipitation events. The comparison of logistic regression and analog methods shows that the LR method outperforms the AN method in almost all the locations and lead times.