Design of Optimal Expansion of Rain Gauge Network in the Himalayan Region for Monitoring Extreme Events
Abstract
The Himalayan region has a complex topography and atmospheric circulation, resulting in high temporal and spatial variability in precipitation. Extreme weather events, like cloudbursts which lead to secondary events like flash floods, have played a devastating role in the life of 52 million people who live in the Himalayas. Precise localized rainfall data is needed for weather prediction and yet, reliable on-ground data collection infrastructure is vastly absent in the Himalayas. The 229,000 sq miles of the Indian Himalayan region has about 77 operating automatic weather stations (AWS) maintained by Indian Meteorological Department. To monitor extreme events, we propose a mechanism to identify optimal locations for expanding the existing network.
We trained a Neural Network (NN) model with a large number of parameters (e.g. slope, elevation, temperature, etc.) to improve the accuracy of the gridded precipitation data. We divided the Himalayan terrain into a set of representative regions (a.k.a strip). The strips with the maximum number of AWS were selected. Kriging was done on the selected strips using the AWS data to generate gridded precipitation data. We also used satellite precipitation data to augment the data set while generating the gridded precipitation data. We then trained NN on these strips with 50 years of weather data to capture the rainfall characteristics. The trained NN model was used on other strips to create gridded data. Previous studies on the design of rain gauge networks have found that strategically designed rain gauge networks can provide high accuracy precipitation forecasts with fewer numbers of AWS. However, existing techniques to identify the new AWS locations do not consider the spatial and temporal variability of the mountainous terrain. NN is used to propose an expansion of the AWS network from the list of accessible locations to improve the overall network accuracy. Our results show that the proposed method can be applied to detect the extreme rainfall events in the Himalayas. The blended gridded daily rainfall estimates have the highest probability of detection of cloudburst events selected for the study region. The geo-spatial Kriging used for the generation of gridded data includes topographical factors which leads to a reduction in RMSE values during cross-validation.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMA183.0008S
- Keywords:
-
- 3314 Convective processes;
- ATMOSPHERIC PROCESSES;
- 3355 Regional modeling;
- ATMOSPHERIC PROCESSES;
- 1817 Extreme events;
- HYDROLOGY;
- 4313 Extreme events;
- NATURAL HAZARDS