My dissertation focused on methods for the spatial modeling and evaluation of tree canopy cover.

We compared models of tree canopy cover that used multi-date composite images to those that used harmonic regression coefficients from full time-series data. Both sets of data used Landsat imagery. We found that the harmonic regression coefficients contained more information and performed slightly better than other models.

We also studied methods for the sampling of a continuous variables derived from models using remote sensing data. We compared four sampling protocols (simple random and stratified random with equal allocation, proportional allocation, and optimal allocation). We found that optimal allocation sampling, which accounts for prediction uncertainty in the allocation of samples, provided error estimates with the highest confidence and showed high confidence with fewer samples.

Finally we collected photointerpreted tree canopy cover data using crowdsourcing on Amazon's mechanical turk. We found that the quality of participants' interpretations were affected by fatigue, time-money motivations (i.e. there was a motive to work quickly to make more money), and potential earnings (if the potential earnings were high, they worked faster).