Skip to main content
. 2021 Aug 5;75:103231. doi: 10.1016/j.scs.2021.103231

Algorithm 1.

2-Step Optimization Workflow for Data Fusion Process.

1. Input: 2D input maps (Xclimate) for each of the 5 climate features, 2D input maps (XSEG) compacting all 11 socioeconomic-governmental factors, size of lead-time (Tlead) for Sub-Scenarios (a-f), values of model's hyperparameters (H1,H2,), threshold values for precision metric (PT,1,PT,2), validation dataset, testing dataset
2. Output: 2D output maps (Y) representing spatial classes distribution for all 251 countries with defined lead-time using the proposed approach
3. for eachXclimate ← Step I of Optimization Process
do
a. initialize training of deep learning prediction model with user-defined H1,H2, values & Xclimatewith selected Sub-Scenario (a-f) for defined Tlead in Step I optimization process
b. derive Y predictions on validation and testing datasets via trained model with defined Tlead & H1,H2, values
c. model evaluation of predictions on testing dataset via P score
ifPPT,1
Proceed
d. store P score with corresponding H1,H2, values & type of Sub-Scenario for defined Tlead
4. select optimal Xclimate with highest P(>PT,1) score with corresponding H1,H2, values & type of Sub-Scenario for defined Tlead
5. do
a. fuse optimal Xclimate with XSEG
b. initialize training of deep learning prediction model with selected H1,H2, values & type of Sub-Scenario for defined Tlead from (4) ← Step II of Optimization Process
c. derive Y predictions on validation and testing datasets with trained model
d. model evaluation of predictions on testing dataset via P score
ifPPT,2
Proceed
store best prediction model for near real-time predictions
Else
re-train prediction model with additional hyperparameters (H1,H2,) tuning
6. end