Open-Source Development of The Python Agrivoltaic Simulation Environment and Case Studies with PASE 1.0

Driven by the urge to expand renewable energy generation, the development of agrivoltaics is currently accelerating. However, harmonious deployment requires evaluating both photovoltaic and crop yields, to ensure simultaneous compliance with energetic and agricultural objectives of stakeholders within evolving local legal contexts.
Based on the community’s priority needs in terms of modelling, this thesis has developed a first version of the Python Agrivoltaic Simulation Environment (PASE), a
unique open-source tool for predicting the agricultural and photovoltaic productivity
of a variety of agrivoltaic systems. In fact, a review of previous work on agrivoltaic
modelling carried out as part of this thesis highlighted the cruel lack of a generic tool,
modular, well balanced between the different components to be modelled, as well as
being open and auditable by peers. This work therefore presents PASE 1.0, a MIT-licensed modelling framework developed in partnership with companies and research
groups to assess the land productivity of agrivoltaic systems. The long-term development vision of PASE is explained and the various benefits expected from it are
described, as well as the open-source business model established with the partners and
its subsequent developments. PASE 1.0 version architecture is also presented and the
various calculation modules are described. Case studies then illustrate how PASE 1.0
effectively fulfills three primary requirements encountered by agrivoltaics stakeholders: predict irradiation on relevant surfaces, estimate agricultural and energy yields
as well as water use efficiency, and ease understanding of processes underlying field
observations.
Two pilot agrivoltaic sites equipped with pyranometers were reproduced in PASE
to evaluate the ray casting model and its assumptions. These showed errors of 7.7
and -5.7 % on the total irradiation received at ground level during the measurement
periods. For each demonstrator, the model was evaluated on days with contrasting sky
conditions. For one day in particular, the model showed an accuracy equivalent to that
of bifacial_radiance.
PASE 1.0 was used in conjunction with the PVLib and PVFactors tools to assess
the suitability of a vertical agrivoltaic installation in Chile, in a region facing severe
and recurring droughts. This work highlighted the potential for bimodal production
of vertical installations to avoid curtailment problems when the photovoltaic potential
was high around midday. For irrigated agriculture of the region, PASE 1.0 evaluated
the amount of potential evapotranspiration that could be saved thanks to the shading
induced by photovoltaic modules and their estimated wind-break effect at 1410 m3
/ha.
PASE 1.0 ability to predict photovoltaic and agricultural yields as well as the land
equivalent ratio over several years was demonstrated for a wheat crop as part of the
BIODIV-SOLAR pilot project. The potential of vertical bifacial agrivoltaic installations to offset energy production was once again highlighted, and a more detailed study
showed the periods and sky conditions that were favourable for this type of installation compared with a south-facing photovoltaic plant. Evaluation of crop yields over
several years showed that the presence of the photovoltaic modules led to an average
yield reduction of 18.9 % with SIMPLE and 4.21 % with STICS. These differences
highlighted the importance of the crop model choice and the associated formalisms. A
sensitivity analysis of inter-row spacing also demonstrated the usefulness of PASE to
design systems according to the criteria set out in the legal frameworks.
Finally, the results of the agronomic trial for the year 2023 on the same pilot site
with spring wheat were presented, showing a low overall yield and a better yield in
the agrivoltaic zone. PASE 1.0 was used in co-simulation with STICS to complete
the interpretation of these agronomic observations. The simulation made it possible to
hypothesise that part of the difference in yield could be attributed to the reduction in
heat stress resulting from the shading of the photovoltaic modules. It was also possible
to detect a nitrogen stress that would explain the low overall yields.