Satellite-data Based Geospatial Modeling Approach for Estimating Evapotranspiration and its parameters for Switchgrass Intercropped Pine Forests

Evapotranspiration (ET) comprises a large part of the forest water balance and serves as an indicator of forest growth and ecosystem productivity. Recently, there has been an increasing interest in energy from forest biomass as fossil fuel replacement, yet environmental and economic limitations have reduced feedstock available for bioenergy. Intercropping switch grass (Panicum Virgatum L.) in between pine (Pinus Taeda L.) trees has a potential for sustainable bioenergy feedstock production without land opportunity costs but can potentially alter ET/water use compared to traditional pine forests. However, measuring ET and its parameters across these forest systems is costly and time-consuming.

This study developed/tested geospatial models informed by satellite/aerial image based normalized difference vegetation index, soil adjusted vegetation index, vegetation vigor index, and other spectral information to estimate ET and its parameters. These parameters include stomatal conductance, leaf area index, canopy temperature, and soil moisture measured on treatment watersheds with young pines and natural understory (YP), switchgrass only (SG), and young pine intercropped with switchgrass (IC). These watersheds were replicated on three sites located across the Southeastern USA. Image analysis automation was created using Python scripting and geospatial models. Despite the growth inconsistency for the SG only treatment, estimation of ET parameters using this geospatial approach yielded acceptable accuracies (R² > 0.70). Models for the YP site performed rather poorly (R² = 0.28-0.76) compared to the IC (R² = 0.47-0.81), also with some inconsistencies in LandSat image pixel size, probably due to random growth of natural understory on YP. Advanced data mining supported Radial Basis Function Network models provided promising results for estimating ET and ET parameters in this intercropped system with their potential to apply on large pine forests. Some issues encountered in the study were dissimilar image spatial resolution for the intercropped vegetation spacing and difficulty in synchronizing the timing of high-resolution image data acquisition with field data collection. Addressing these issues will likely enhance the model predictive efficiency in future studies.