I'm a hydrologist and pretty new to machine learning, but would like to use sensor data that I have (rainfall time series for example) and combine it with GIS data (grids/rasters of topography for example) as the input to a neural network to then produce a variable of interest (streamflow for example). I can easily take a 1d-array of daily rainfall and match it up with a 1d-array of stream flow and set up and train a regression perceptron (vanilla) network in Keras, but I am having a hard time wrapping my head around how it would be best to combine the single timeseries values with static but spatially distributed topography data to create a single input array of input/training data to then ultimately try to predict the 1d-array of streamflow data.
Any assistance regarding how to format these data sets would be appreciated. Thanks!
Article A Spatial Data Infrastructure Integrating Multisource Hetero...
One way is to use convectional neural network. The other is Laplace approximation for summation or integration of high dimensional integrals. Also you can you the method of saddle point approximation.
I agree with Ali
Article A study of optimal model lag and spatial inputs to artificia...
https://arxiv.org/pdf/1604.08880
Article Application of Deep Convolutional Neural Networks for Detect...
Article Modeling urban evolution using neural networks, fuzzy logic ...
https://ascelibrary.org/doi/abs/10.1061/(ASCE)0733-9496(1995)121:6(499)
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training