Skill of short- to medium-range monsoon rainfall forecasts from two global models over India for hydro-meteorological applications

Abstract

The South Asian region is dominated by summer monsoon rainfall which, in many years, is associated with severe floods leading to socio‐economic losses. In India, areas with high population are exposed to heavy rainfall events and their associated risks. Indian monsoon rainfall simulation and prediction at various spatial and temporal scales are challenging scientific tasks for the weather/climate modelling community. Skilful short‐ and medium‐range predictions of rainfall from global numerical models during monsoons are highly desirable for various hydro‐meteorological applications, including flood forecasting. For recent years (e.g. 2000–2016) higher resolution global advanced assimilation‐forecast systems have resulted in improved medium‐range rainfall forecasts. These model rainfall forecasts potentially could be used in hydro‐meteorological applications including water‐related disaster management/planning. In this study, deterministic rainfall forecasts from two state‐of‐the‐art global models (from the National Centers for Environmental Prediction and the Met Office) were evaluated against observed rainfall for the Indian summer monsoon season. Objective skill scores such as mean, bias, correlation co‐efficient, mean absolute error and root‐mean‐square error from Day 1 to Day 5 were examined for the monsoon season over India and at seven major river basins within India for 2013. It was seen that both of the models had useful skill for different regions and basins. For the Indus and Krishna basins, even the Day 5 forecasts were seen to be skilful. In general, the Met Office Day 5 forecasts were more skilful at all seven river basins. Finally, rainfall forecasts from the latest higher resolution Met Office Global Unified Model (implemented in July 2014) were compared with its immediate predecessor to document any further improvement during the monsoon period over India. It was seen that, at the medium range, the newer version of the Met Office model simulated Indian monsoon rainfall more realistically than the previous versions, with much reduced biases in the core monsoon trough region.

Assessment of Met Office Unified Model (UM) quantitative precipitation forecasts during the Indian summer monsoon: Contiguous Rain Area (CRA) approach

“…However, there is a considerable reduction in easterly bias (and rainfall wet bias) during the monsoon season of 2014 and 2015 (figure not shown). Some of these improvements in the forecast rainfall during recent years may be attributed to the increased horizontal resolution (∼17 km), a new dynamical core (ENDGame) and a revised physics package (GA6.1) (Prakash et al 2016;Rakhi et al 2016). The spatial distribution of average rainy day counts (>1 mm d −1 ) over Indian region (land points only) during the nine monsoon seasons of 2007-2015 is presented in figure 5.…”

Fast Biases in Monsoon Rainfall over Southern and Central India in the Met Office Unified Model

“…The aforementioned studies refer to longer climate simulations, but forecasting the Indian Summer Monsoon is also challenging at shorter timescales appropriate to numerical weather prediction (NWP) (Ranade et al 2014;Gadgil and Srinivasan 2012), and the MetUM also shows rainfall biases at NWP scales (Prakash et al 2016;Mitra et al 2013). Categorical yes/no forecasts of rainfall are generally good, but it is rather more difficult to produce good forecasts of rainfall amount (Joshi and Kar 2016;Kumar et al 2017).…”

Forecasting the monsoon on daily to seasonal time‐scales in support of a field campaign

“…Global Land (GL) version 6.1 defines the global land science configuration, which uses the Joint UK Land Environment Simulator (JULES) community land surface model (Best et al, 2011;Clark et al, 2011). Prakash et al (2016) showed good skill for five-day forecasts of the 2014 Indian monsoon season with this MetUM configuration. In order to provide more local detail in the forecasts for the campaign, a bespoke limited-area model (LAM) was developed covering a domain of 5…”