Page 23 - Preliminary Analysis Towards a Standardized Readiness Framework - Interim Report

Preliminary Analysis Towards a Standardized Readiness Framework

maximizing the area coverage for drone-based disaster response. Ad hoc network design
between drones helps in using multiple drones for surveillance and rescue. Models used
include Reinforcement Learning models, multi-agent systems and systems, and collaborative
intelligent solutions. Data collected include video and still images along with satellite images.
Simulations including sim2real approaches help curriculum learning (for achieving a smoother
learning curve from simple to complex scenarios. UAV network design is used to autonomously
perform essential tasks within disaster-stricken areas.

4�1�3 Management of Wildfires

This use case [51] uses predictive wildfire prevention and early detection of wildfire, in addition,
it adds value to local knowledge and aids in community empowerment. The output from the
use case is consumed by powerline utilities with KPIs on availability. A combination of cameras
(private) and satellite heat sensors (could be public) are used for prediction in this use case.
Datasets include streamed data from cameras and smoke and light detection. Post-fire analysis
using proprietary algorithms for fire prediction, detection, and fire propagation models. Pilot
deployments in India (2 reserves), Brazil, and Portugal are used to test the models.

4�1�4 Disaster Risk Management in Complex Geography

The use case [50] actors in this case are transporters, and other systemic actors such as
businesses, insurers, disaster management entities such as government agencies, and the
general public. Data analyzed includes open data from disaster management agencies (risk
data, real-time), and private data for creating context country-region specific advisories (satellite
images). This data is used for training models such as generation (advisory generation) and
prediction (forecasting), including continuous improvement models. Interoperability and
compatibility of cross-region data (e.g. early warning) and generated advisories are major
reasons for standards and metrics for AI readiness in this use case.

4�1�5 AI-based Chatbot for Farmers

This is an agricultural use case [49] that collates data from open data portals maintained and
updated by government actors. Time series and government data related to agriculture,
including crop production, land use, water use, market prices, weather patterns, and
government schemes are used for training models. GPT-like static models vs. Retrieval
augmented generation (RAG)-based dynamic updates to the policies database [49] are to be
studied to bring maximum benefits to farmers who use this solution. Satellite images to locate
the stakeholders and farmers along with time series market data on crop prices are other
factors to consider in this use case. The pilot study of agriculture-related AI technology on
7000 farmers in the Khammam district of Telangana (India) showed promising results, where
the net income of the farmers using the AI technology had been doubled ($800 per acre)
from the average income in 6 months [33]. The solution readiness may include cloud APIs for
subscribing/publishing of data from portals [49].

4�1�6 Networked ASEAN Peatland Forest for Net-Zero

This use case [48] proposes a tropical peatland fire weather index (FWI) system by combining
the GWL with the DC. Pilot deployments include a LoRa-based IoT system for peatland
management and detection in RMFR in Kuala Selangor, Malaysia. Verification of data with truth

15