InVEST_surrogate

This repository contains code and data for developing a deep learning-based surrogate model of selected InVEST modules. The surrogate aims to replicate InVEST outputs efficiently for spatial optimization, especially in green infrastructure planning. The work is WIP.

1 Contents

• Rasters: Geospatial input files. These are clipped into uniform square samples for training.
• Sample: Shapefile defining sample boundaries across the region; used to clip raster inputs.
• InVEST_Model: Tabular inputs and workplace for selected InVEST modules—Habitat Quality, Urban Cooling, and Urban Nature Access. Tabular inputs are constant across all samples.
• Prosd: Processed tensors (torch inputs) clipped from rasters, saved for model training and InVEST calculation.
• Normal: Normalization parameters used for standardizing model inputs.
• Results: Ground truth outputs from InVEST simulations, used as training targets for surrogate models.
• StudySite: Data and configuration for running GI optimization on a selected case site.

2 Code Scripts

• CNN_InV_01: Clips raster data using the Sample shapefile and saves torch inputs in Prosd.
• CNN_InV_02: Runs InVEST models and saves outputs in Results.
• CNN_InV_03: Validates data, defines surrogate model architectures, and trains the models.
• CNN_InV_04: Evaluates trained model performance.
• CNN_InV_05: Applies surrogate and InVEST models to optimization on the case site and compares results.