I am trying to predict what my rainfall field will look like at a future timestep using:

  1. Radar imagery of rainfall fields at previous timesteps: A set of 2D matrices where each element in each matrix corresponds to the rainfall intensity [mm/hr] and the dimensions of each matrix - the rows and columns, corresponds to the X,Y coordinate (i.e. the location). Therefore, each matrix denotes the rainfall intensities at every location at a specific timestep.

  2. Climate data: A spatial averaged (i.e. averaged over every location for each timestep) climate dataset. An example is illustrated below (the 't's refer to the timestep).

Spatial averaged (i.e. averaged over every location for each timestep) climate data

So my goal is to forecast what the rainfall field will look like (i.e. the output would be a 2D matrix with rainfall intensities at each location) at a future timestep (e.g. t + 1 hour) using inputs from BOTH (1) and (2). I know I can use the radar image alone and pass it through a CNN and it should work fine but I want to incorporate both climate data and radar image data.

Would I use a multi data/multi-type input Neural Network and concatenate at the end? If so, then how would I do so and are there any resources I can be directed to? If not, then what should I do?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.