https://www.mdu.se/

Agrivoltaic system combines solar energy and agriculture which is an effective way to utilize the lands full potential. Crops can be grown between vertical panels or under tilted panels among other designs. Combining solar panels and agriculture leads to optimization of space. This degree project evaluates a comparison of several parameters measured both in a vertical agrivoltaic system and a reference control plot located in Kärrbo Prästgård, Västerås, Sweden. Specifically, the correlation between the ground albedo and normalized difference vegetation index (NDVI) under the two treatments are investigated. Correlations between the albedo and the NDVI against different weather parameters are also explored. Linear regression models are developed separately for the albedo and NDVI with the most correlated parameters. In addition, because the albedo in the reference is not the same as the albedo in the agrivoltaic system, a linear regression model linking the albedo of the agrivoltaic system, and the albedo of the reference system is further developed. With this latter model, power production from the vertical agrivoltaic system is simulated under different albedo considerations: using measured albedo from the agrivoltaic system, using predicted albedo from the linear regression model, and using measured albedo from the reference system. These power estimations are then compared to the real power production from the agrivoltaic system. The study employs MATLAB for data analysis, models development and power production simulations.

The study compared the correlation between ground albedo and NDVI in an agrivoltaic system and a control plot. The albedo model revealed that the reference system could explain 87% of the albedo variance in the agrivoltaic system, but the NDVI model showed that the reference system could only account for 39.6% of the variation in the agrivoltaic system. Furthermore, for the power production comparison, using the actual measured albedo in the agrivoltaic showed the most accurate power, employing predicted albedo through the linear regression model showed the second highest and using the albedo measured in the reference showed the least accurate.