Input real soil data from SAFE

[hrlai]

In the next step let's try to input real soil data from SAFE (?) This will be a hard test like what @lelavathy is doing with the hydrology.

For many soil variables we probably won't have spatially explicit data, but here's a first pass:

• When data are spatially implicit, model mean and variance and simulate them randomly across grids
• When data are spatially explicit, include a spatial autocorrelation term in the model and then in the simulation

[lelavathy]

@hrlai For the hydrology model, I’m currently working with coarser resolution climate data at 9 km for Maliau region. To match the 90 m resolution needed for the model, I’ve applied a simple interpolation using the nearest grid point approach. This allows us to spatially align the data with the finer model grid, although it doesn’t add new spatial detail—just ensures compatibility for now.

[hrlai]

@lelavathy when you upscale the 9-km climate data to 90-m, are you keeping the extend the same? If so, what is the bounding box for the VE input data?

[hrlai]

Building a spatially explicit model in glmmTMB is well until I try to make spatial prediction (5485250). At 90-m resolution, any spatial extent that does beyond 1 km will take up lots of RAM; expanding things into pairwise distance matrix is simply too slow.

I will try SPDE modelling using INLA instead.

[lelavathy]

@lelavathy when you upscale the 9-km climate data to 90-m, are you keeping the extend the same? If so, what is the bounding box for the VE input data?

I initially download ERA5 climate data at 9 km resolution for the broader Maliau region, using the bounding box (lat: 4.92 to 4.62, lon: 116.72 to 117.02). After interpolating this to 90 m resolution, I then crop it to a much smaller extent to minimize the runtime for the Virtual Ecosystem model.

The final bounding box for the VE input data at 90m resolution is:

Latitude: 4.7170137 to 4.7569565

Longitude: 116.9492683 to 116.9890846

[hrlai]

With @jacobcook1995 's soil campaign data I was able to reach working spatially-explicit model. Here's a snapshot of pH spatial predictions (this is wip):

Will think about this vs. simple interpolation.