The UC Davis Experimental Ecovoltaic Park (DEEP), located south of Interstate 80 and north of Putah Creek (38.520268, −121.739191), is a single-axis GPV spanning 62 acres (0.25 km²) constructed in 2015 on agricultural land owned by the University of California, Davis (Davis, California, USA). Historically, the land was a natural wetland adjacent to wildflower-dominated grasslands and riparian woodlands, typical of the California prairie biome (Holstein, 2011). Situated in a Mediterranean climate, Davis has an annual global horizontal irradiance of 1,854 kWh/m², average temperatures ranging from 8.9°C to 23.9°C, and annual precipitation of 498 mm, primarily between October and April (U.S. Climate Data, 2023). The soil is predominantly characterized as the Yolo series, featuring well-drained alluvium (SoilWeb, 2023).

Adjacent to the UC DEEP are diverse agricultural activities, including crop cultivation (e.g., alfalfa, tomatoes, and sunflowers), horse ranching, and apiary operations. With a nameplate capacity of 13 MWac (16.3 MWdc), the GPV generates 33 GWh annually, meeting approximately 14% of the electricity demand of the campus. Originally developed by SunPower, the facility ownership was transferred to Arevon Energy in 2020. The system employs multicrystalline PV modules mounted on 1.37-m high piles with single-axis tracking (east–west direction). When fully tilted (~53°), the lowest frame edge is 0.54 m above the ground, with a pole-to-pole spacing of 4.57 m.

Vegetation management includes thrice-yearly mowing (February, June, and September) to maintain plant heights of 100–150 mm and remove biomass. This practice minimizes shading that may reduce power output, prevents physical contact with cabling and other GPV elements, and mitigates fire risks by controlling fuel loads (Randle-Boggis et al., 2020; Uldrijan et al., 2021). Additional measures such as herbicide application, soil fertility depletion, or revegetation were not implemented. Consequently, noxious, non-native weeds are prevalent across the site. The research area, covering approximately 7.6 acres (0.031 km²) in the central region of the facility, was permitted for our ecological studies (Figure 1A).

The UC DEEP is characterized by a mixed-use footprint, with 14% of the area supporting little to no vegetation due to gravel roads, bare ground used as lay-down areas for on-site materials, and infrastructure required for operations and maintenance activities. The remaining 86% was classified into five distinct micro-patch types based on the diurnal shading patterns created by the PV panels:

NS represents conditions closest to a natural, unshaded grassland ecosystem, serving as a baseline for evaluation. In contrast, FS is subject to persistent shading, offering insight into environments with highly modified light availability. The other three micro-patches (MS, AS, and BS) have intermediate shading conditions that vary in timing, potentially influencing microclimate and vegetation.

These five micro-patches can also be grouped into three functional zones based on their relative position to the solar arrays: (i) within-array, comprising BS and FS located within solar arrays, (ii) near-array, comprising MS and AS located along the periphery of the solar arrays, and (iii). beyond-array, comprising NS in fully open areas farther away from the solar arrays.

The classification of these micro-patches reflects the ecological heterogeneity introduced by single-axis GPVs. This framework enables the investigation of how shade-induced microclimates drive flora composition and productivity, with broader implications for managing solar installations as dual-use or multifunctional landscapes that optimize ecological and operational outcomes.

To ensure robust data collection, four permanent 48-m² plots (22 m length × 2.7 m width) were randomly selected and delineated within each micro-patch type, resulting in 20 plots across the permitted research area (Figure 1B). Monthly data collection was conducted from December 2022 to November 2023, spanning 12 sampling periods to capture seasonal variations.

We chose one plot per micro-patch type to monitor meteorology, including ambient temperature (AT; °C), dew point temperature (DP; °C), relative humidity (RH; %), and wind speed (WS; m s⁻¹) at 15-min intervals by digital 5,400-AG weather meters (Kestrel Meters, Boothwyn, PA, USA) mounted on posts 0.6 m height above the soil surface of center of each plot for 6–7 days in the second or the third week of each month. Vapor pressure deficit (kPa) was derived from AT and RH following Ward and Trimble (2003): V P D = 0.6108 ∗ e 17.27 ∗ A T A T + 273.4 ∗ ( 1 − R H 100 )

Arithmetic averages of daily, daytime, nighttime, morning, and afternoon, as well as daily maximum, minimum, and range, were calculated for AT, DP, RH, WS, and VPD. Photosynthetic active radiation (PAR; μmol m⁻²s⁻¹), daily light integral (DLI; mol m⁻²d⁻¹), and photoperiod (hours) were concurrently monitored with DLI-400 meters (Apogee Instruments, Logan, UT, USA) at 3-min intervals on the same posts of weather meters. Daytime, morning, and afternoon mean PARs were calculated on a daily basis. Growing degree days (GDD) was derived from the average of daily maximum and minimum ATs minus a base temperature threshold of 5°C: G D D = ( A T max − A T min ) / 2 − 5

Volumetric water content (VWC; %) and soil temperature (ST; °C) were measured at 10 cm depth on six randomly selected spots per plot with TDR-315H (Acclima, Meridian, ID, USA)—using a portable time domain reflectometer—from 10:00 a.m. to 2:00 p.m., yielding 100 records every month/sampling event (5 micro-patches x 4 plots/micro-patch x 5 records/plot).

Vegetation communities were evaluated by five randomly positioned 1 m² x 1 m² quadrats per plot (i.e., 100 quadrats, or 100 m², per month/sampling event). Within each quadrat, we (i). surveyed presence–absence, canopy coverage, blooming frequency, height profile, and aboveground live biomass to the lowest taxonomy (further details could be referred to Note 1 in Supplementary material 1), (ii). estimated bare ground and dead tissue coverages, (iii). calculated maximum height and total aboveground live biomass as well as the coverage ratios of narrowleaf: broadleaf, heliophilous: sciophilous, and anemophilous: entomophilous, and (iv). derived species richness, Simpson diversity, Shannon diversity, and Pielou’s evenness based on the presence–absence using the “vegan” package (Oksanen et al., 2019) of R 4.2.1 (R Core Team, Vienna, Austria).

All statistical analyses were conducted using R 4.2.1 software.

The primary goal was to assess how microclimate variables (e.g., VPD, WS, PAR, GDD, VWC) and vegetation variables (e.g., presence–absence, mean height, litter coverage, live biomass) were influenced by the interactions between micro-patch type and seasonal or monthly variations. To achieve this, generalized linear mixed-effect models (GLMMs) were applied, with micro-patch type (five levels: NS, MS, AS, BS, and FS), month (12 levels: December to November), or season (four levels: winter, spring, summer, and autumn), and their two-way interactions as fixed factors. Random factors, including date (for meteorology, excluding ST and VWC) and plot (for vegetation, ST, and VWC), were incorporated to account for hierarchical data structure or repeated measures.

The distribution of the response variables was determined using the “fitdistrplus” package (Delignette-Muller and Dutang, 2015), ensuring that appropriate model families were chosen. For instance, Gaussian was employed for continuous real-number data (e.g., AT, VPD, diversity indices). Assumptions of normality and homoscedasticity were checked using the Shapiro–Wilk test, and when necessary, data were transformed using the “bestNormalize” package (Peterson and Peterson, 2020) before model fitting. For other types of data: Binomial, Beta, Gamma, and Poisson were more appropriate for binary numbers (e.g., presence–absence), continuous probability numbers (e.g., RH), positive real numbers (e.g., height), and natural numbers (e.g., species richness), respectively,

GLMMs were conducted using the “lme4” package (Bates et al., 2015). To address issues such as zero inflation, overdispersion, or autocorrelation, the “DHARMa” package (Hartig and Hartig, 2017) was used for diagnostics, and the “glmmTMB” package (Brooks et al., 2023) was applied for corrected modeling when needed. Model selection was guided by the lowest Akaike information criterion (AIC) to ensure optimal fit. Significant effects of sampling periods or micro-patches were examined further using Tukey’s honestly significant difference (HSD) post-hoc tests with the “emmeans” package (Lenth and Lenth, 2018), providing pairwise comparisons between treatment levels.

To analyze community composition, non-metric multidimensional scaling (NMDS) was performed using the “vegan” package, leveraging Jaccard dissimilarity and 999 permutations to visualize differences in weed community assemblages based on presence–absence data. NMDS, as a non-parametric approach, is well suited for ecological datasets with uncertain or non-linear relationships. The dimensionality of NMDS plots was optimized by selecting stress values below the recommended threshold for accurate ordination (Martin, 2022). Differences among levels of season, micro-patch type, and their interactions were tested using permutational multivariate analysis of variance (PERMANOVA).

To complement the unconstrained NMDS and enhance the interpretability of species co-occurrence patterns, linear discriminant analysis (LDA), a supervised clustering approach, was performed using the “MASS” package (Ripley et al., 2013; Priyadarsini et al., 2015). To identify species specifically associated with each season and micro-patch combination, indicator species analysis was also conducted using the “indicspecies” package (De Caceres et al., 2016).

Our comprehensive statistical framework enabled detailed exploration of both microclimate and vegetation dynamics, providing robust insights into how these ecological variables respond to spatiotemporal variations of the environment.

Wind speeds were the highest in NS, with an annual mean of 3.89 m/s (± 2.58 m/s), followed by MS (3.00 m/s ± 2.28 m/s), AS (2.75 m/s ± 1.87 m/s), BS (2.70 m/s ± 2.05 m/s), and FS (2.10 m/s ± 1.95 m/s) (Figure 2A; Table 1). However, these differences were not statistically significant across all months and seasons (Supplementary Tables S1, S2 in Supplementary material 2). A similar trend was observed for daytime PAR, with MS, AS, BS, and FS receiving 91, 88, 83, and 11% of the PAR observed in NS, respectively (Figure 2B). PAR in MS and AS did not significantly differ from BS during the morning and afternoon, respectively, due to shadows cast by the PV panels (Figures 2C,D; Supplementary Figure S2 in Supplementary material 1). The daily photoperiod was longest in winter and shortest in autumn in NS, although the annual mean photoperiod was comparable across all micro-patches, ranging from 13.4 to 13.8 h per day (Table 1; Supplementary Figures S3A–D in Supplementary material 1).

AT, DP, VPD, and GDD peaked in July or August (Supplementary Figure S1 in Supplementary material 1). Among the micro-patches, AS consistently exhibited the lowest GDD as well as the mean, maximum, daytime, afternoon, nighttime, and range of AT and VPD, while RH and DP exhibited the opposite patterns (Table 1; Supplementary Figures S3E,F in Supplementary material 1). The highest maximum AT and VPD were observed in BS (AT: 27.4°C ± 10.8°C; VPD: 2.86 kPa ± 2.32 kPa), and the greatest ranges of these measures occurred in MS (AT: 20.7°C ± 7.21°C; VPD: 2.78 kPa ± 2.31 kPa), although statistical analysis revealed no significant differences across micro-patches (Supplementary Tables S1, S2 in Supplementary material 2). Interestingly, mean, daytime, and nighttime AT in FS were comparable to those in NS (Table 1). In the morning, NS and AS were the warmest, whereas in the afternoon, MS displayed higher AT and VPD than all other micro-patches (Figures 2E–H).

Temperature of soil averaged 24.8°C (± 5.38°C) in NS, which was 2.4°C, 3.0°C, 3.6°C, and 5.0°C warmer than in FS, BS, AS, and MS, respectively (Figure 3A; Table 1). Annual VWC was highest in BS (13.9% ± 13.0%), followed closely by FS (13.2% ± 13.6%), AS (13.0% ± 12.8%), and MS (12.6% ± 13.3%), with the lowest levels observed in NS (11.9% ± 13.7%) (Figure 3B; Table 1). The statistical differences between BS and FS, as well as among AS, MS, and NS, were marginal (Supplementary Tables S1, S2 in Supplementary material 2). Over time, FS experienced the driest conditions in December and February (up to −7.1%) but transitioned to the moistest from April through June (up to +6.8%) (Figure 3C). No significant differences in VWC were observed among the micro-patches during other months (Supplementary Tables S1, S2 in Supplementary material 2).

A total of 37 species were identified at the UC Davis Ecovoltaic Solar (Supplementary Table S1 in Supplementary material 1), including 16 dominant broadleaves (observed across multiple quadrats during more than one sampling event; Figure 4), 12 rare broadleaves (Supplementary Figures S4A–L in Supplementary material 1), and 9 narrowleaves (Supplementary Figures S4M–T in Supplementary material 1). Of the five species native to California, four—Croton setiger, Epilobium ciliatum, Gilia tricolor, and Matricaria discoidea—were rarely encountered, while the invasive Erigeron canadensis was common. Approximately 60% of species were shade-tolerant, 53% were insect-pollinated, and 30% were perennial or semi-perennial.

The presence and coverage of certain broadleaves, such as Malva neglecta, Silybum marianum, and Erodium cicutarium, declined dramatically in March, July, and October following mowing events (p < 0.05) (Supplementary Figure S5 in Supplementary material 1). Mowing also caused notable reductions in the height and biomass of M. neglecta and Dittrichia graveolens (Supplementary Figure S7 in Supplementary material 1; Supplementary Table S3 in Supplementary material 2). While the blooming frequency of M. neglecta decreased from summer to autumn, the floral resources of Polygonum aviculare and Erigeron bonariensis displayed the opposite trend (p < 0.01) (Supplementary Table S4 in Supplementary material 2).

Aboveground biomass of certain species varied significantly across micro-patches. For example, both D. graveolens and P. aviculare exhibited greater biomass in near-array zones than the other two zones, while Lactuca serriola was most productive in FS (p < 0.05) (Supplementary Figure S9B in Supplementary material 1; Supplementary Tables S3, S4 in Supplementary material 2). Blooming frequencies also showed variation: M. neglecta bloomed more abundantly in NS, Erigeron bonariensis in AS and BS, and P. aviculare in MS (p < 0.05) (Supplementary Figure S9A in Supplementary material 1). However, no consistent patterns were observed for plant height (Supplementary Tables S3, S4 in Supplementary material 3).

Narrowleaves emerged in early winter and were absent between August and October, with peak presence and coverage in April and May when species identification was enabled due to spikelet development (Supplementary Figures S4M–T in Supplementary material 1; Supplementary Table S1 in Supplementary material 1). Avena fatua, Bromus hordeaceus, Bromus tectorum, Elymus repens, and Phleum pratense were observed only in NS; Hordeum murinum and Polypogon monspeliensis were exclusive to the near-array zone, whereas Festuca perennis and Pennisetum clandestinum were widespread across all micro-patches. Narrowleaf presence and coverage were significantly lower near-array (p < 0.05) (Supplementary Figure S7 in Supplementary material 1; Supplementary Tables S3, S4 in Supplementary material 2).

Two-dimensional NMDS achieved excellent fit (R² = 0.982, stress = 0.135), with NMDS1 and NMDS2 explaining 60.9 and 30.8% of variability, respectively (Figures 5A–B). Species clustering was apparent near-array (between MS and AS) and within-array (between BS and FS), as confirmed by LDA (Figures 5C,D). PERMANOVA indicated significant seasonal and micro-patch effects on species composition (p < 0.0001), though centroids overlapped between (i) summer and autumn and (ii) MS and AS (Supplementary Figure S12 in Supplementary material 1; Supplementary Table S6 in Supplementary material 2).

Species richness, diversity, and evenness reached the maximum in June, with the highest and lowest values in BS and near-array, respectively (Figures 6A,B; Supplementary Tables S5, S6 in Supplementary material 2). The coverage ratios of heliophilous-to-sciophilous and anemophilous-to-entomophilous species peaked in summer and autumn, with the highest values in AS and MS, respectively (Supplementary Figure S10 in Supplementary material 1). Indicator species analysis revealed early-season (e.g., S. marianum), late-season (e.g., H. echioides), and all-season (e.g., M. neglecta) groupings, as well as species characteristic of each micro-patch (Supplementary Table S2 in Supplementary material 1).

The maximum height of the plant community peaked in summer and was the tallest in BS (110 cm ± 81.0 cm) (Supplementary Figure S1 in Supplementary material 1). However, live aboveground biomass of BS was 53–88% of the other micro-patches (Table 3). Bare ground exposure increased significantly in March and April after the first mowing event (p < 0.0001) and was consistently the highest in FS except during spring (Figures 6C,D; Supplementary Tables S5, S6 in Supplementary material 2). Dead litter began accumulating in May, and the coverage was 35% higher in NS than in the other micro-patches during summer (p < 0.05) (Figures 6E,F; Table 2).

The shading and wind-sheltering effects of single-axis GPVs were evident across all micro-patches within the within-array and near-array zones (Armstrong et al., 2016; Fagnano et al., 2024). PAR was intercepted by 9, 12, 17, and 89% in MS, AS, BS, and FS, respectively, compared to NS (Figure 2B; Table 1). These reductions in sunlight exposure could be adjusted by deploying meteorological sensors at varying heights, enabling comparisons with values reported in other studies (Liu et al., 2019; Graham et al., 2021; Sturchio et al., 2024). Unlike fixed-tilt PV systems, the diurnal rotation of panels resulted in a 3% lower PAR in AS than MS (p < 0.05), attributed to the higher intensity of afternoon irradiance than morning conditions (Figures 2C,D). Additionally, this rotational design prevents permanent shadows; even FS received an annual mean PAR of 89 μmol m⁻²s⁻¹ (Table 1; Moscatelli et al., 2022; Uldrijan et al., 2022; Vaverková et al., 2022). A similar trend was observed for wind speed, with NS recording 1.30, 1.41, 1.44, and 1.85 times the values of MS, AS, BS, and FS, respectively (Figure 2A; Table 1). These values exceed the range (1.3–1.6 times) observed beneath fixed-tilt panels at 0.5–1.2 m heights (Hassanpour Adeh et al., 2018), corroborating the findings by Li et al. (2024) that single-axis tracking systems may be more effective in reducing airflow. However, the differences in wind speed across micro-patches were only marginally significant, likely due to the dominance of north–south winds that align with the orientation of module strings, reducing cross-panel disruptions (Oomen, 2024).

Air temperature, as well as derived variables VPD and GDD, were primarily driven by seasonal trends instead of micro-patch types (Supplementary Figures S1A–D in Supplementary material 2; Yue et al., 2021; Sturchio et al., 2022; Yavari et al., 2022). Contrary to our expectations, AS, rather than FS, exhibited the lowest air temperature, VPD, and GDD, along with the most dampened variations among all five micro-patches (Table 1). For instance, while AS was fully exposed to sunlight in the morning and experienced similar air temperatures to NS (AS: 15.0 ± 7.15°C; NS: 15.1 ± 7.37°C), afternoon shading resulted in AS being on average 0.5°C cooler than NS (Figures 2E–H). By contrast, although afternoon shading also occurred in BS and FS, they were on average 0.7°C and 0.5°C warmer than NS, respectively, attributable to a combination of heat dissipation from adjacent operating solar panels and reduced wind speeds, counteracting the cooling and humidifying effects of cast shadows (Hassanpour Adeh et al., 2018; Wu et al., 2022). These findings suggest that shading may not always mitigate evaporative demand for vegetation during periods of high solar irradiance—particularly for stems situated very close to the panels (Makaronidou, 2020; Fagnano et al., 2024). The nighttime “PV heat insulation” hypothesis proposed by Armstrong et al. (2016)—whereby fixed-tilt panels obstruct skyview and constrain the escape of long-wave outgoing radiation—was not observed in this study. Instead, AS remained the coolest micro-patch, albeit without statistical significance, likely related to the minimal energy absorbed by the ground surface during the daytime (Table 1).

Unlike air temperature, the topsoil temperature (0.1 m depth) of NS was consistently the warmest (up to 4.9°C) throughout the entire year (Table 1). However, the temperatures of near-array micro-patches were still lower than those of within-array (Figure 3A). These patterns suggest that impacts of emitted long-wave radiation from PV panels still exist but diminish with distance (Marrou et al., 2013b; Wu et al., 2022). In contrast, VWC exhibited notable seasonal variations (Figure 3C). In December and February, the diversion and redistribution of precipitation by PV panels reduced water inputs to the FS; this “umbrella” effect was mitigated during periods of ample rainfall in January and March through three potential pathways: (i) belowground hydrological processes, (ii) lateral transfer of concentrated fluxes via micro-gullies and capillaries, and (iii) water flow along driplines and supporting frames (Armstrong et al., 2016; Elamri et al., 2018; Choi et al., 2020; Makaronidou, 2020; Sturchio et al., 2024). As the Mediterranean climate entered its dry season in April, with precipitation dropping to nearly zero, a combination of relatively low soil temperatures and reduced evapotranspiration allowed FS to retain 3–7% more soil moisture than other micro-patches through August (Figure 3C; Yang et al., 2017; Hassanpour Adeh et al., 2018; Barron-Gafford et al., 2019; Tanner et al., 2020; Lambert et al., 2021). On an annual average, VWCs of FS (13.9%) and BS (13.0%) were 0.6–2% higher than in other micro-patches (Figure 3B; Table 1). However, these values were much lower than the range of 11–34% reported in other studies conducted on single-axis or fixed-tilt PV systems over 2 m in height in less arid regions (Liu et al., 2019; Yue et al., 2021; Fagnano et al., 2024).

Consistent with Vaverková et al. (2022), the tracking GPV infrastructure created favorable conditions for annual broadleaves (Supplementary Table S1 in Supplementary material 1). However, the formation of repeating mosaics of unique environmental zones—that is, micro-patches—played a significant role in shaping vegetation assemblages, driven largely by adaptations to the PAR gradient (Liu et al., 2019; Graham et al., 2021; Kannenberg et al., 2023; McCall et al., 2024). Among dominant broadleaves, heliophilous species such as C. arvensis, C. solstitialis, and H. incana were confined to the NS micro-patch, while E. cicutarium was notably absent within-array (Figure 5B). In contrast, sciophilous species such as C. pycnocephalus and H. echioides were restricted to within-array zones, whereas E. bonariensis and L. serriola exhibited higher abundances near-array (Supplementary Figure S6 in Supplementary material 1). As highlighted by LDA, the distinct composition across the three broader zones (within-array, near-array, and beyond-array) was pronounced (Figure 5D). Nevertheless, the coverage ratio of heliophilous-to-sciophilous species did not show an increasing trend along gradients in micro-patches with more extensive skyviews (Supplementary Figure S10A in Supplementary material 1), suggesting that factors like interspecific, asymmetric competition may also play a key role (Uldrijan et al., 2022). For instance, P. aviculare and D. graveolens appeared to benefit from the morning and afternoon shade of MS and AS, respectively, during summer and autumn, as evidenced by their significantly higher presence, coverage, and biomass (Supplementary Figures S6, S9 in Supplementary material 1). In the early season, E. cicutarium formed dense, basal “carpet” in MS and AS, preempting available space and suppressing less competitive cohorts such as annual grasses and S. marianum (Supplementary Figures S6, S7 in Supplementary material 1). Consequently, species diversity and evenness near-array were both approximately 19% lower than in the other two zones (Table 3). The higher species richness observed in BS (3.13) than in FS (2.88) and NS (2.66) aligned with Liu et al. (2019), suggesting that the moderate shading (> 15%) imposed by PVs may promote niche partitioning of flora.

Although no consistent patterns of height at the species level were observed, the maximum height of the vegetation stand was lowest in NS (80 cm) and highest in BS (110 cm)—the only micro-patch with vegetation exceeding the lower leading edge of PV panels (106.7 cm) (Supplementary Figure S11B in Supplementary material 1). Beyond phenotypic plasticity, such as stem elongation to enhance light capture (Weigelt et al., 2021; Kannenberg et al., 2023), this variation is more likely tied to the relative abundance of vegetation assemblages. In particular, six of the eight tall broadleaf species (mean height > 30 cm) were tolerant of partial shade, with C. album and E. bonariensis being particularly prevalent in BS (Table 2; Supplementary Figure S6 in Supplementary material 1; Uldrijan et al., 2021; Uldrijan et al., 2022). Unlike findings from studies at conventional fixed-tilt GPVs, the annual mean aboveground biomass showed no significant difference between within-array patches (FS: 180 g/m²; BS: 157 g/m²) and beyond-array patches (NS: 177 g/m²) (Supplementary Figure S11A in Supplementary material 1; Hassanpour Adeh et al., 2018; Elamri et al., 2018). However, AS was 46 and 73% more productive than MS and the other micro-patches (Table 3), aligning with Sturchio et al. (2024), who reported that reduced heat stress from afternoon shading benefits community yield. The availability of overall floral resources provided by dominant broadleaves did not vary significantly across micro-patches, except the most abundant E. bonariensis, M. neglecta, and P. aviculare, which were characterized by multiple inflorescences per stem and exhibited blooming frequencies congruent with their presences (Supplementary Figure S8 in Supplementary material 1). In contrast to Uldrijan et al. (2021) on fixed-tilt GPVs, the highest coverage ratio of anemophilous-to-entomophilous species was observed in MS and AS—rather than FS—where P. aviculare and D. graveolens predominated, respectively, during the mid- to late-growing season (Supplementary Figure S10B in Supplementary material 1; Brownsey et al., 2013). Their wind-dispersed pollen poses potential challenges: (i) deposition on PV module surfaces, impairing power generation efficiency and increasing maintenance costs and (ii) triggering human allergies (Sanz Saiz et al., 2020; Nowak et al., 2023).

Statistical analysis using GLMMs revealed significant spatiotemporal variations in vegetation coverage. Bare ground exposure was most pronounced in (i) March, coinciding with the first mowing event, and (ii) FS, where low germination and high mortality rates affected species unable to tolerate extreme sunlight deficiency, except in spring (Figures 6C,D; Schindler et al., 2018; Lambert et al., 2021; Uldrijan et al., 2022). Vegetation coverage remained relatively stable throughout the entire year with annual means exceeding 80% on most micro-patches (NS: 87.4%; AS: 84.4%; MS: 87.9%; BS: 85.9%) except FS (72.2%) (Table 3). These are comparable to the 80.5% documented for within-array patches by Beatty et al. (2017) but lower than the 90.5% measured in July by Liu et al. (2019). The much lower coverage reported for NS in Liu et al. (2019) (13.4%) likely reflects their lower annual precipitation (200 mm) of the study site compared to Davis (429 mm), underscoring the role of aridity in amplifying the benefits of solar arrays on vegetation productivity (Barron-Gafford et al., 2019). Dead litter accumulation began in May as early-season graminoids (e.g., H. murinum and P. clandestinum) and forbs (e.g., C. pycnocephalus, E. cicutarium, and S. marianum) entered senescence (Figure 6E; Supplementary Figures S5, S7 in Supplementary material 1). NS was the most drought-prone micro-patch, characterized by the warmest soil temperature and the fastest moisture depletion, resulting in the highest dead litter coverage during summer (Figure 3; Armstrong et al., 2016). Broadleaf species conducive to burning were scarce, except near-array where relatively flammable species such as D. graveolens and S. tragus dominated during summer and autumn (Bernau and Eldredge, 2018). Although annual grasses (e.g., B. hordeaceus, B. tectorum, and A. fatua) are significant contributors to layered fuel beds that can easily ignite and propagate fire (Vaverková et al., 2022), their occurrences were either rare—particularly within-array—or did not overlap with peak drought periods. As a result, the fire hazard associated with these grasses may remain minimal under the current conditions.

At the UC DEEP where weed pressure is intensive, adhering to a mowing-only strategy poses risks in the following six aspects:

Moreover, mowing can also inadvertently exacerbate weed problems. For species that propagate via underground rhizomes or tubers (e.g., thistles, lettuces, and fleabanes), mowing may encourage resprouting from residual stems. If conducted during seed production phases of the life cycle, mowing can instead facilitate the spread of seeds (Bossard and Lichti, 2000). These factors have contributed to the low diversity of native species at the UC DEEP after 7 years of operation.

While eradicating all noxious weeds on-site may be unrealistic, adopting alternative approaches may help enhance maintenance efficiency and socio-ecological benefits (Randle-Boggis et al., 2020; Uldrijan et al., 2023). No single method can effectively manage all noxious weeds across the diverse micro-patches without tradeoffs. Instead, multiple control practices tailored to specific site conditions should be implemented in parallel or sequence for optimal outcomes (see Note 2 in Supplementary material 1 for alternative options and Supplementary Table S3 in Supplementary material 1 for their effectiveness against species observed at the UC DEEP) (DiTomaso et al., 2013; US Fish and Wildlife Service, 2014).

Given the scale of the site (>1 acre) and specific challenges, such as the toxicity of certain broadleaf species to grazing livestock, we recommend that O&M personnel at the UC DEEP, or similar GPVs, implement a targeted approach. Specifically, repeated shallow cultivation combined with herbicide spot-spraying could be conducted over at least 2 years to deplete the weed seed bank, followed by the sowing of sacrificial cover crops to proactively occupy the ground surface. Extra resources should be allocated to BS due to its tallest vegetation profiles and highest species richness, and to the near-array zone to address potential fuel loads and pollen spillover. Careful planning and execution of control measures are essential to ensure they align with site performance goals and avoid unintended impacts on PV modules or surrounding ecosystems (McCall et al., 2023).

For solar energy developers aiming to balance low long-term maintenance costs, biodiversity net gain, esthetic enhancement, and heritage preservation, the establishment of a permanent native vegetation community offers an ideal solution. Such vegetation should at least meet several criteria: (i) competitive but non-invasive, (ii) short-statured to remain below the leading edge of PV panels, and (iii) fire-retardant, featuring traits that decrease flammability, such as high live fine fuel moisture and low levels of volatiles, waxes, and resins (Blackhall and Raffaele, 2019). By integrating these strategies, GPV can realize a confluence of ecosystem services beyond clean energy generation, fostering resilient and multifunctional landscapes.

This study has several limitations that warrant consideration. First, the findings are based on a single year of data collection, which may not capture interannual variability driven by climatic fluctuations or successional changes in vegetation. Additionally, the study was conducted at a single site, limiting the generalizability of the results to other GPV installations with different climate, soil, and ecological conditions. The exclusive focus on a single management practice—mowing—further constrains the scope, as alternative strategies such as grazing, herbicide application, or integrated approaches may yield different outcomes. Moreover, the fixed height and depth of microclimate measurements, while sufficient for characterizing surface conditions, may overlook vertical gradients and subsurface processes critical to ecosystem functioning.

Future research should prioritize multi-year or long-term monitoring to better understand temporal dynamics and their implications for vegetation management and ecosystem services (Lambert et al., 2021). Expanding studies to include GPVs in diverse climate regions, soil types, and array configurations—such as elevated, bifacial, or dual-axis panels—would enhance the applicability of findings across various solar installations. Additionally, incorporating a wider range of management practices and optimizing microclimate measurement protocols to capture more comprehensive profiles could provide deeper insights into the interplay between tracking PV systems, microclimate, and vegetation dynamics.