Python/Numpy-based simulated annealing for optimizing timber land management.

Determine the best timber stand management strategies over time to optimize for multiple property-level objectives.

For example, the harvest scheduler can answer questions such as

What is the best combination of management prescriptions and temporal offsets in order to maximize both carbon sequestration AND timber revenue at the lowest cost?

The input data for the harvest scheduler is a 4D Array.

Variable of Interest X Stands X Rx X Time Period

It is easier to visualize this as multiple 3D arrays, one for each variable of interest.

The scheduler will

1. Select a random management for each stand
2. Assess the performance of that state by calculating an objective metric - a single number which we attempt to minimize.
3. Randomly change the management of a single stand and reassess the objective metric.
4. Decide whether to accept or reject the change based on the simulated annealing algorithm. Goto step 3 and repeat.

The calculation of the objective metric involves looking at each variable, for each stand over time (a 3D array) and distilling it down to a single number. In order to do this, each variable can be handled according to a preset strategy

• cumulative-maximize: Attempt to maximize the cumulative property-level sum of the variable over time.
• cumulative-minimize: Attempt to minimize the cumulative property-level sum over time.
• evenflow: Attempt to minimize the cumulative property-level standard deviation over time.
• evenflow-target: Attempt to minimize the variation over time around a set target. Target can be scalar or array of values over time.

See scheduler/prep_data.py for details.

This module contains a function prep_data.from_shp_csv which takes two arguments:

1. Path to a shapefile containing stand polygons. Required attributes, projection, etc. are all defined by the prep_data.py module. This will eventually be specified/documented but for now, just read the code carefully for the specs.
2. Path to the directory containing the csvs output from the FVS batch growth-yield process.

The data from the shapfile and csvs will be combined to form the data structures expected by the scheduler.

The data will be cached to disk to avoid recalculating expensive datasets. Remove the cache.* files in order to force a recalculation.

See test_scheduler.py for an example of usage of the scheduler interface.