Playing Catch up in Food Forecasting Technology

Context: India needs a technically capable workforce that can master ensemble weather and ?ood forecast models.

Overview of Flood Forecasting in India:

• Use of Deterministic Forecasting Model: In this model, local agencies makes a decision if a flood forecast uses the words “Rising” or “Falling” above a water level at a river point. 
• Agencies Involved:
• The Indian Meteorological Department (IMD): issues meteorological or weather forecasts.
• Central Water Commission (CWC): issues flood forecasts at various river points.
• Disaster Management Authorities and local administration: act as the end-user agencies. 

• Use of Technology: IMD has 35 advanced Doppler weather radars to help it with weather forecasting. 

• Doppler weather radars can measure the likely rainfall directly (known as Quantitative Precipitation Estimation or QPE) from the cloud re?ectivity over a large area.
• It has extended the “lead time” (time gap between issuance of forecast and occurrence of a flood) by up to three days.
• Introduction of  Ensemble Model in India: The IMD has begun testing and using ensemble models for weather forecast through its 6.8 Peta ?ops supercomputers (“Pratyush” and “Mihir”).

Shortcomings with Indian flood forecasting

• Limitations of Deterministic Forecasts: 

• Any small change in initial conditions of a weather model results in output that is unexpected. 

• Therefore, beyond a lead time of three days, a deterministic forecast becomes less accurate. 

• Presence of multiple agencies leading to delays (Poor Water Governance): Following factors impact the flood forecasting, leading to a reduction in the “lead time”.

• Time taken by the IMD to estimate and forecast rainfall.
• Time taken by the CWC to integrate the rainfall forecast (also known as Quantitative Precipitation Forecast or QPF) with ?ood forecast. 
• Time taken by the CWC disseminated this data to the end-user agencies.
• Outdated Statistical Methods (CWC’s technological gap): have neutralized the advantage of advanced technology
• The current method is of the type gauge-to-gauge correlation and multiple coaxial correlations and enables a lead time of fewer than 24 hours.
• These statistical methods fail to capture the hydrological response of river basins between a base station and a forecast station and cannot be coupled with the QPF.

• Non-uniform model across India: 

• India has recently moved to use hydrological (or simply rainfall runo? models) capable of being coupled with QPF. 
• Therefore, a lead time of 3 days is sporadic in India and at a select river points.
• IMD’s technological gap: 
• India will need at least an 80 100S band dense radar network to cover its entire territory for an accurate QPF. 
• Else, the limitations of altitude, range, band, the density of radars and its extensive maintenance enlarge the forecast error in QPF which would ultimately re?ect in the CWC’s ?ood forecast. 

An Overview of Ensemble Technology

• Probabilistic Forecasting Model: It assigns different probabilities to different scenarios of water levels and regions of inundation.
• An example of a probabilistic forecast can be like: chances of the water level exceeding the danger level is 80%, with likely inundation of a village at 20%.
• Worldwide Use: The United States, the European Union, and Japan have already shifted towards “Ensemble ?ood forecasting” along with “Inundation modelling.” 

Way Forward: 

• Ensure Technological Parity: The forecasting agency (i.e. CWC) should try to achieve technological parity with the IMD to a couple of ensemble forecasts to its hydrological models.

• Modernize Telemetry infrastructure: and also raise technological compatibility with river basin specific hydrological, hydrodynamic, and inundation modelling. 

• Employ a technically capable workforce: that is well versed with ensemble models and capable of coupling the same with ?ood forecast models. 

• Employ Artificial Intelligence: Google AI has adopted the hydrological data and forecast models derived for diverse river basins across the world for training AI to issue ?ood alerts in India. 
• This by-passes the data de?ciencies and shortcomings of forecasts based on statistical methods. 

Conclusion: Probabilistic forecasts will provide end-user agencies ample time to decide, react, prepare and undertake risk- based analysis and cost- e?ective rescue missions, reducing ?ood hazard across the length and breadth of India.