You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be very useful to be able to be able to standardise time-series data of environmental variables between different formats,
as suggested by @jamesmaino
Some datasets like worldclim use tmax/tmin layers on a daily basis,
others use hourly/3 hourly temperature records, like SMAP.
Models written to work with hourly/3 hourly temperature can't use worldclim-style min/max data. But this could be resolved with a cyclic interpolator (sinosoidal or more complex) that presented data as if it was a regular GeoSeries containing regular GeoStacks with regular GeoArray layers.
There are a few components to this:
Interpolator wrapper: holds 2 arrays and fraction f to interpolate with
a function for calculating f for any value in the time series
an internal mutable wrapper inside the interpolators to make sure data shared across multiple timesteps are only loaded once -
and only copied to the GPU once.
There are also some performance gains to be had interpolating between files, especially when running models on the GPU. Loading file data and moving it to the GPU is often a lot more costly than doing anything to them - so loading files to GPU only once should be the goal.
We may also need a way of specifying the kind of array the data is loaded to initially, such as CuArray.
The text was updated successfully, but these errors were encountered:
It would be very useful to be able to be able to standardise time-series data of environmental variables between different formats,
as suggested by @jamesmaino
Some datasets like worldclim use tmax/tmin layers on a daily basis,
others use hourly/3 hourly temperature records, like SMAP.
Models written to work with hourly/3 hourly temperature can't use worldclim-style min/max data. But this could be resolved with a cyclic interpolator (sinosoidal or more complex) that presented data as if it was a regular
GeoSeries
containing regularGeoStack
s with regularGeoArray
layers.There are a few components to this:
and only copied to the GPU once.
There are also some performance gains to be had interpolating between files, especially when running models on the GPU. Loading file data and moving it to the GPU is often a lot more costly than doing anything to them - so loading files to GPU only once should be the goal.
We may also need a way of specifying the kind of array the data is loaded to initially, such as
CuArray
.The text was updated successfully, but these errors were encountered: