This research study presents a framework for applying physically based global illumination techniques to the creation of software models of light fields that are then validated against actual light fields measured in physical experiments. A prior experiment was performed by horticulture scientists in which the light field of an empty plant growth chamber was measured using quantum sensors at fixed spatial intervals. The result was a light map consisting of a 9 x 45, fixed-width, two-dimensional graph of sensor readings that described the intensity of radiant energy present in the chamber at the chosen locations. A single observation of the growth chamber was made resulting in a single data set consisting of 45 different, location-sensitive irradiance observations.

To test this framework a series of simulations were performed in which the physical attributes of the growth chamber were duplicated as closely as possible in a virtual growth chamber software model. Modeled attributes included physical dimensions, wall and light reflectivity, and full-spectrum light characterization. Light transport was modeled using a physically based, global illumination rendering technique called photon mapping. Virtual sensors that recorded the intensity of the light that transmitted through their surface were placed in the virtual chamber at the same position and interval as the ones that were used in the physical experiment. The output of the virtual chamber experiments were represented as a graph in the same configuration as the one in the physical experiment. The experiment was conducted using a modified version of pbrt, a physically based, extensible renderer developed by Matt Pharr and Greg Humphreys [1].

As photon mapping uses a stochastic algorithm, many repetitions of the virtual chamber experiment were performed and the mean and standard deviation were recorded as a global measure for each chamber as well as for each individual sensor location. The global means of the physical and virtual chambers were compared using a simple linear regression analysis and were shown to have a p < 0.0001 at α < 0.05, thus indicating a significant association between them. The correlation coefficient between the chambers was –0.66 indicating a fairly strong negative relationship between the two groups.

A robust statistical comparison of sensor-to-sensor similarity between the physical and virtual was not possible due to the limited number of observations in the physical data. However visual comparison of the patterns of response between a given physical and virtual sensor does show a significant inverse relationship between the two. It is unclear why the relationship between the virtual and physical chambers is inverse; potential causes and solutions are discussed.

An important conclusion reached in this study is that photon mapping as implemented in pbrt demonstrates the ability to function as a tool for irradiance modeling as well as image synthesis.