Hackathon CI 2022
Drought prediction over West Africa: using standardized precipitation index (SPI)

Instructions from CI2022 Hackathon:
The goal is to train a model that receives 3 time steps of SPI as input (SPI_t-2, SPI_t-1, SPI_t) and then predicts 1 time step of SPI, but 3 months ahead, SPI_t+3. You do not have to predict SPI_t+1, SPI_t+2 or anything else then SPI_t+3 \\
The input dimension is time_stepsxlatxlon (3x13x20) and the output dimension is 1xlatxlon 

To train your model you can use all of the provided CESM data. But to make it work for the task you should learn to predict 3 months ahead given 3 past time steps

The three months input can be any months of the year. It can be (jan, febr, march) , it can be (jun,jul,aug), it can be (dec, jan, febr)

For the final evaluation you will use your already trained model and not do any retraining

We will provide new data, but only the input data and keep the output/target. You will use this new input data to create a prediction and then submit this prediction to us. We will then compare the prediction to the target

The new input data will be multiple samples/instances of 3 months of SPI and for each of these samples you will create a prediction (1 time step of SPI 3 months ahead)

Based on the starter_kit python scripts provided by Paula Harder, we use multiple linear regression as a baseline and try to build other ML models that can perform better over the MLR.