Long-term photovoltaic succession promotes preferential ammonium accumulation in desert ecosystems Authors: Jianhua Xiao, Anlin Wang, Ya Tian, Qilin Tan, Tao Chen, Fei Liu, Zhaoshan Cai, Shijun Ma, Fenghua Sun, Hong Yang, Longfei Chen Published: 2026 ABSTRACT SummaryWhile nitrogen rigidly constrains productivity in desert ecosystems, the long-term impacts of rapidly expanding utility-scale photovoltaics (PVs) on soil nitrogen dynamics remain poorly understood. Here, we quantified the trajectories and environmental drivers of total nitrogen (TN), nitrate (NO3−-N), and ammonium nitrogen (NH4+-N) across three panel microhabitats over a 14-year desert chronosequence. Soil nitrogen accumulated progressively with increasing array age, with TN rising by 37%–54% after 12–14 years of operation. Crucially, NH4+-N accumulated preferentially within 2–3 years and remained persistently elevated under panels (UPs). Although NO3−-N also accumulated in this zone, it fluctuated in exposed areas (between-panel, BP and runoff, FL), indicating potential leaching losses. These findings highlight the decoupling of inorganic nitrogen dynamics, underscoring UP soils as persistent nitrogen sinks. Accordingly, we propose targeted interventions to optimize sustainable nitrogen management and mitigate environmental risks in desert PV ecosystems.Subject Areas: environmental science, environmental management, environmental monitoring, soil chemistry, energy resources, energy systems Introduction Nitrogen availability rigidly constrains primary productivity across Earth’s drylands, which occupy ∼41% of the terrestrial surface.1,2,3 Unlike forest or grassland ecosystems, where organic nitrogen dominates, desert soils are distinguished by inorganic nitrogen, particularly nitrate and ammonium, constituting the primary stable and biologically available pool.4,5,6 These inorganic nitrogen forms are highly responsive to hydrological pulses and microhabitat variability and therefore play a central role in nitrogen cycling in arid ecosystems.7,8,9,10,11 The global push toward carbon peaking and carbon neutrality has positioned solar photovoltaics (PV) energy as a core pathway for energy decarbonization and a central component of strategies to address the energy-climate crisis.12,13,14,15,16 Global PV capacity has expanded explosively, with 345.5 GW added in 2023 alone, accounting for 73% of newly installed renewable energy capacity.17 China contributed 216.87 GW of this addition, increasing its cumulative PV capacity to 610 GW, and its carbon-neutrality targets are expected to drive further expansion.12,18,19,20 Arid regions worldwide receive annual solar irradiance exceeding 2,200 kW h m−2, more than twice that of temperate regions,21 making them strategic priorities for utility-scale PV deployment.22 In September 2020, China pledged to peak CO2 emissions by 2030 and achieve carbon neutrality by 2060.23 Against this policy backdrop, and leveraging both abundant solar resources and strong policy support, the desert regions of northwest China have rapidly become primary hubs for large-scale PV development and currently host more than 60% of the nation’s centralized PV capacity.17,24,25 This large-scale land-use conversion has generated sharply contrasting microenvironments beneath and between panels (BPs);26 yet, the decadal-scale consequences for soil nitrogen pools and inorganic fractions remain highly uncertain. Inorganic nitrogen, primarily nitrate and ammonium, dominates the stable bioavailable pool in desert soils.7 Their spatial heterogeneity directly influences the intensity of plant resource competition and patterns of niche differentiation, thereby contributing to the maintenance of species diversity.1,7,27 PV have been tasked with the dual goals of “energy supply and ecological restoration.”17,18,28,29 However, achieving these goals is complicated by the fact that PV array structures substantially alter the surface energy balance and water cycling, and construction-phase disturbances (e.g., land leveling, vegetation clearance) markedly modify soil structure.16,26,30 These changes increase surface runoff, raise soil bulk density by 18%–25%, and reduce porosity,31 thereby impeding the physical redistribution of nutrients, intensifying erosion, and causing losses of soil carbon and nitrogen.32 Collectively, these processes constitute a novel form of anthropogenic disturbance that may increase the risk of land degradation.32,33,34,35 Previous studies have shown that, with increasing array age, contrasts among microhabitats beneath and around PV panels become more pronounced.36 Under-panel (UP) zones, persistent shading rapidly creates a cool-moist regime that suppresses nitrification, stimulates fungal-driven mineralization, and fuels sustained accumulation of soil organic matter and inorganic nitrogen.16,31,36,37 In stark contrast, exposed BP zones experience intensified wet-dry cycles that accelerate mineralization but also make nitrate highly vulnerable to leaching and volatilization, thereby making NO3−-N the main form by which nitrogen is lost.26,29,32 In addition, by intercepting and concentrating rainfall and dissolved nutrients from across the entire array slope, front-line runoff accumulation (FL) zones likely play a critical yet largely overlooked role in the long-term accumulation and redistribution of nitrogen and its inorganic fractions. Although short-term observations have revealed these contrasting early trajectories,26,31,38,39 it remains highly uncertain whether ammonium enrichment in UP zones and nitrate depletion in BP zones persist over decadal timescales, which factors ultimately govern their long-term dynamics, and how FL zones modulate nitrogen retention and loss within PV arrays. To elucidate the decadal-scale trajectory of total nitrogen (TN) and its inorganic fractions under large-scale PV disturbance in desert ecosystems, we selected eight utility-scale power stations in Northwest China’s deserts, spanning array ages of 1, 2, 3, 7, 8, 9, 12, and 14 years (Figure 1). We quantified the 14-year trajectories and environmental drivers of TN and its inorganic fractions (NH4+-N and NO3−-N) across three panel associated microhabitats: UP, BP, and FL zones. We hypothesized that (1) PV exert distinct long-term effects on different nitrogen pools (TN, NH4+-N, and NO3−-N) in desert soils, and (2) these differential effects are jointly controlled by array age, panel position, and PV-induced changes in soil physicochemical properties. This study aims to elucidate the long-term trajectory and potential mechanisms of soil nitrogen components in arid ecosystems under the influence of large-scale PV installations, providing important insights for optimizing soil nitrogen retention, guiding ecological conservation, and supporting the sustainable management and multi-functional land use of desert PV power stations. Location of the eight PV power stations in the arid desert region of northwest China and schematic diagram of the experimental sampling design Structural schematic of PV module and panel position sampling zones: under-panel (UP), between-panel (BP), and front-line runoff accumulation zone (FL). Results Response ratios of soil nitrogen fractions By contrast, in the initial stage (1 year of operation), soil nutrient levels in the mobile dune (MD) were overall poor, with a mean TN content of 0.127 g kg−1, NH4+-N of 3.05 mg kg−1, and NO3−-N of 3.07 mg kg−1. After 14 years of PV succession, however, TN increased to 0.184–0.207 g kg−1, NH4+-N to 5.46–5.91 mg kg−1, and NO3−-N to 3.68–4.15 mg kg−1 across the different panel positions, respectively. TN exhibited no significant response during the first three years of PV operation (RR ≈ 0), except for a localized increase in FL soils in year 3. Significant TN enrichment emerged UP in year 7 and extended to all positions (UP, BP, and FL) from year 8 onward, reaching 37.3%–54.5% by year 14 (Figure 2A). NO3−-N showed no significant response during years 1–3 (RR ≈ 0), but all panel-associated positions (UP, BP, and FL) were significantly enriched relative to MD in years 7 and 9; by year 14, however, significant enrichment persisted only in the UP zone, with intermittent depletion occurring in FL (year 8) and BP (year 12) (Figure 2B). In contrast, NH4+-N responded immediately and most strongly to panel installation. Significant enrichment occurred in UP soils in years 1–2 and in BP soils in year 2, and from year 3 onward RR was significantly higher than in MD across all panel-associated positions. The magnitude of this response far exceeded that of TN and NO3−-N, culminating in 50.4%–62.8% enrichment by year 14 (Figure 2C). Response ratios of soil nitrogen fractions in different microhabitats (BP, FL, UP) across a 14-year chronosequence of desert PV ecosystems (A, B, and C) represent the response ratios of total nitrogen (TN), nitrate nitrogen (NO3−-N), and ammonium nitrogen (NH4+-N), respectively. BP, between panel; FL, front-line runoff accumulation zone; UP, under panel. Statistical significance of RR values against the null hypothesis (RR = 0) was determined using two-tailed one-sample t tests. Values are means ± standard errors (SEs). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Temporal dynamics of nitrogen response ratios RRTN increased significantly with array age at all panel positions (UP, r = 0.57; FL, r = 0.54; BP, r = 0.46; p < 0.001). LOESS smoothing revealed a slow initial increase in RRTN during the first 3 years, accelerated enrichment between years 7–9 (11.1%–34.7%), and maximum accumulation between years 12–14 (37.3%–54.5%). UP soils consistently exhibited the earliest and strongest gains in TN (Figure 3A). Mixed-effects models confirmed that the age × panel position interaction explained 57.3% of the variance in TN (Figure 3E). UP soils consistently exhibited the earliest and strongest TN gains. NO3−-N showed significant positive trends with array age only in the UP (p = 0.0064) and FL (p = 0.0075) positions, with enrichment plateauing after year 12 (Figure 3B). Array age accounted for 40.7% of the variance in NO3−-N (p < 0.01), whereas panel position contributed only 4.1% (p < 0.05; Figure 3F). NH4+-N enrichment occurred rapidly and was structured far more strongly by panel position than by array age. RR of NH4+-N increased significantly with array age at all positions (p < 0.05); a large proportion of this enrichment occurred within the first 2–3 years after construction, with UP soils showing the largest early increases. Peak RR were reached between years 7 and 9 and remained high through year 14, with panel position alone explaining 41.0% of the variance in NH4+-N (compared with only 1.3% for array age; Figure 3C). Synergistic accumulation of TN and NH4+-N was strongest beneath panels, whereas NO3−-N and NH4+-N covaried significantly only in FL soils. Kernel density distributions of TN and NH4+-N RR were markedly right-skewed in UP soils, indicating a consistently higher probability of enrichment at this position (Figure 3D). Temporal trends, microhabitat heterogeneity, and relative importance of key drivers for soil nitrogen response ratios in desert PV ecosystems over a 14-year chronosequence (A–C) Linear regression models showing the temporal trends of response ratios for total nitrogen (TN, A), nitrate nitrogen (NO3−-N, B), and ammonium nitrogen (NH4+-N, C) in each microhabitat. Regression equations, coefficients of determination (R), and p values are provided for each habitat. Shaded areas represent 95% confidence intervals. Statistical significance was determined using two-tailed Pearson correlation tests. (D) Pairwise correlations (upper triangle), frequency distributions (diagonal), and boxplots (right) of TN, NO3−-N, and NH4+-N response ratios across microhabitats. (E–G) Variation partitioning analysis (VPA) quantifying the percent variation explained by location (microhabitat type), year (PV array age), and their interaction for TN (E), NO3−-N (F), and NH4+-N (G). The independent contribution percentages were calculated using hierarchical partitioning based on a linear regression model. Significance levels for each factor were derived from ANOVA (F tests).∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Soil property dynamics relative to undisturbed reference conditions Relative to the MD, SWC increased significantly in UP in years 2 and 14 and in FL in year 7, whereas other years and positions showed non-significant but consistently positive responses (Figure 4A). EC did not differ significantly among the three panel positions (BP, FL, and UP) during years 1–7. After year 7, EC in BP and FL tended to be higher than in UP and MD, although these differences were not statistically significant (Figure 4B). Soil pH in BP decreased significantly in years 1 and 2 compared with MD, but was significantly higher than MD in years 7 and 9. In contrast, FL exhibited significantly higher pH than MD from year 8 onward. UP showed slightly lower pH than MD in year 14, but this difference was not statistically significant overall (Figure 4C). SOM remained relatively stable in MD (0.37–0.45 g kg−(1) but accumulated rapidly in PV-affected zones. After year 7, all three panel positions showed highly significant SOM enrichment (p < 0.001), with response magnitudes following the order UP > FL > BP. In UP, SOM increased from 0.40 g kg−1 at the start of operation to 1.41 g kg−1 by year 14, corresponding to a 197.8%–213.3% increase relative to MD (Figure 4D). Total phosphorus (TP) exhibited a similar pattern to SOM, with a generally significant positive response from year 2 to year 14 (except year 3) and an overall increase of 50%–80% by year 14 (Figure 4E). In contrast, AP in BP, FL, and UP remained within a narrow range of 1.9–2.7 mg kg−1 throughout the chronosequence, with a significant positive response detected only in FL during years 12–14 (Figure 4F). Temporal dynamics and response ratios of key soil physicochemical properties in different microhabitats across a 14-year chronosequence of desert PV ecosystems (A–F) Temporal changes in soil physicochemical properties: (A) soil water content (SWC), (B) electrical conductivity (EC), (C) pH, (D) soil organic matter (SOM), (E) total phosphorus (TP), and (F) available phosphorus (AP). Values are means ± SE. (Bottom) response ratios of soil properties across different panel positions and array ages, relative to adjacent undisturbed plots. The dashed line at y = 0 represents the baseline reference of the original undisturbed soil properties for the corresponding year. Statistical significance for RR values (relative to 0) was determined using two-tailed one-sample t tests. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Drivers of soil TN and fractions Random forest variable importance (%IncMSE) identified year, SOM, TP, and Pre as the main predictors of variation in TN, NO3−-N, and NH4+-N (Figure 5). In addition, TN was strongly associated with location and DI, NO3−-N with DI, and NH4+-N with location, DI, and pH. SEM corroborated the dominant role of Year, which had significant positive direct effects on TN, NO3−-N, and NH4+-N (Figure 6ab). SOM also exerted a significant direct positive effect on TN, whereas the composite Climate variable had a significant negative direct effect on NO3−-N. Year further showed significant positive effects on both SOM and TP (p < 0.05), indicating that nutrient accumulation was largely age-driven. Location significantly influenced SOM and TP, but its indirect pathways to NH4+-N via SOM and TP were not significant (Figure 6C). Relative importance of environmental factors for soil nitrogen fractions, as determined by the percentage increase in mean squared error (%IncMSE) from a random forest model (A, B, and C) represent the driver importance for TN, NO3−-N, and NH4+-N, respectively. Year, array age; location, PV panel position; SWC, soil water content; EC, electrical conductivity; SOM, soil organic matter; TP, total phosphorus; AP, available phosphorus; Pre, precipitation; DI, drought index; ST, surface temperature. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Structural equation models (PLS-SEM) illustrating the direct and indirect effects of key drivers on soil nitrogen fractions in desert PV ecosystems (A, B, and C) represent the models for TN, NO3−-N, and NH4+-N, respectively. Single-headed arrows denote the hypothesized direction of causation. Red and blue arrows indicate significant positive and negative paths, respectively, while dashed arrows indicate non-significant paths (p > 0.05). Arrow width is proportional to the standardized path coefficient (numeric labels). The latent variable “Climate” represents the standardized weighted sum of precipitation and drought index (or surface temperature). The significant direct paths from “Year” and “Location” to nitrogen fractions represent the cumulative effects of unmeasured biological drivers (e.g., microbial community succession and enzyme activity) that are not mediated by the included physicochemical factors. Discussion PV development and operation alter environmental processes in desert soil systems.16,26,40 This study reveals the pivotal role of PV array age and panel microenvironment in influencing changes in nitrogen and inorganic component content in desert soils. Our first hypothesis, which proposed the existence of significant spatiotemporal patterns in soil nitrogen pools over time, was strongly supported by the data. Specifically, we found that during the early operational phase of the PV power station (years 1–3), TN concentrations exhibited minimal fluctuation and no significant accumulation, indicating that PV-induced disturbances had not yet exceeded the threshold of background variability. Overall TN levels remained relatively low. However, TN accumulation exhibited pronounced spatiotemporal heterogeneity with increasing array age. After seven years of operation, TN accumulation in UP markedly exceeded that in BP. The marked accumulation of total nitrogen beneath panels primarily resulted from the well-documented “cool-moist island” effect generated by PV arrays. This effect substantially reduced soil temperature and evaporative water loss while increasing soil moisture in arid and semi-arid regions.41 These ameliorated microclimatic conditions effectively relieved heat and drought induced constraints on microbial activity, thereby accelerating organic-N mineralization and enhancing inorganic-N retention, processes now firmly established as the predominant drivers of nitrogen enrichment beneath panels in both desertified grasslands and karst semi-arid ecosystems.42,43 The array age × panel position interaction accounted for 57.3% of TN variance (VPA), establishing microclimatic divergence driven by array age and panel position as the predominant driver of nitrogen spatial heterogeneity. NO3−-N exhibited pronounced differentiation among panel positions. Significant enrichment occurred at the UP position only in years 7, 9, and 14 and at the FL position only in years 7 and 9, whereas the BP position showed decreasing trends in some years (Figures 2B; 3B). NO3−-N depletion in BP soils primarily reflects its well-known susceptibility to loss in fully exposed, high-evaporation environments.8 Random forest analysis, VPA, and SEM together revealed that interannual hydroclimatic variation, rather than panel shading, exerted primary control on NO3−-N dynamics (panel position independent contribution only 4.07%; Figures 3F; 5B; 6B), rendering NO3−-N the dominant vector of nitrogen loss from BP soils. As a runoff accumulation area, the FL zone receives relatively more water input and could theoretically act as a “resource island” that favors plant growth in arid ecosystems. Although vegetation data were not measured in the present study, enhanced plant nitrogen uptake represents a plausible but unconfirmed complementary mechanism for the lack of NO3−-N accumulation in FL soils, in addition to leaching risk. Therefore, the observed NO3−-N dynamics in FL zones may be jointly driven by hydrological loss and potential biological consumption, which deserves further verification with vegetation observations in future research. NH4+-N, by contrast, responded immediately and intensely to panel shading. After only 3 years of operation, NH4+-N was significantly enriched at all panel positions (Figures 2C; 3C). From years 7–14, NH4+-N concentrations remained generally highest UP, markedly exceeding BP soils and equaling or slightly surpassing FL in most years. This immediate response is primarily driven by two established mechanisms: (1) cooler temperatures beneath panels strongly suppressed ammonia-oxidizing microorganisms, blocking the conversion of NH4+ to NO3–44 and (2) elevated soil moisture stimulated fungal-dominated organic-N mineralization, rapidly releasing available NH4+.45 Furthermore, a potential trade-off exists between soil alkalinity and nitrogen retention. Although the alkaline soil pH (> 8) theoretically promotes ammonia volatilization, our results showed a sustained accumulation of NH4+. We attribute this to two primary factors consistent with our observations. First, the significant increase in SOM under panels (Figure 4) enhances the soil’s capacity to retain cations like NH4+ through adsorption, counteracting potential losses. Second, the shading effect of panels significantly reduces near-surface temperature and wind velocity, which are critical drivers of ammonia volatilization rates.2,46 This “cool and sheltered” environment likely suppresses the kinetic loss of NH3−, thereby counteracting the chemical facilitation of volatilization by high pH. These results indicate that nitrate nitrogen is prone to loss or transformation, whereas in PV-disturbed desert soil systems an efficient strategy for conserving nitrogen resources is based on the more stable ammonium form, with dynamic transformations enabling adaptation to environmental fluctuations. This might be a trade-off strategy for soil nitrogen accumulation and morphological and functional transformation under the disturbance of PV power stations in arid areas. Desert ecosystems are constrained by chronic water deficiency, nutrient scarcity, and suppressed biological activity, and particularly under sustained disturbance from large-scale PV installations, the processes, and drivers of soil nitrogen accumulation, transformation, and transport differ markedly from those in forest and agricultural systems.2,7,47,48 Random forest analysis indicated that SOM, TP, Pre, and DI are the main environmental drivers of variation in TN and its fractions, which is consistent with the prevailing view that carbon, nitrogen, and phosphorus are co-limited and strongly regulated by hydrothermal conditions in desert ecosystems.49,50 Our 14-year chronosequence results showed that SOM and TP increased cumulatively by 197.8%–213.3% and 50%–80%, respectively (Figures 4D and 4E), and their high-value areas largely overlapped with zones of TN and NH4+-N enrichment and were significantly positively correlated with them. These results strongly validate the environmental-driver component of our second hypothesis. Previous studies have also demonstrated that the accumulation of SOM improves soil structure, enhances water-holding capacity, and supplies carbon substrates, thereby creating a favorable microenvironment for microbially driven nitrogen transformations such as mineralization and immobilization.51 From the perspective of nutrient co-limitation, a meta-analysis based on 1,734 observation sites worldwide further showed that phosphorus addition, by markedly increasing plant aboveground net primary production and microbial biomass nitrogen, enhances biological nitrogen retention and suppresses NO3− leaching, thereby driving a sustained accumulation of soil total nitrogen and producing a pronounced cumulative effect over time.52 Considering this body of evidence, it can be inferred that in the PV-disturbed desert soils examined in this study, the gradual accumulation of SOM and TP likely acts by simultaneously enhancing the supply of microbial substrates and P nutrients, thereby jointly promoting nitrogen retention and the sustained increase of TN and inorganic nitrogen fractions, thus providing mechanistic support for the nitrogen enrichment patterns we observed. Notably, the SEM revealed a different pathway pattern: although SOM and TP are important factors influencing TN and its fractions, neither showed a significant direct path coefficient in the SEM, whereas array age (year) exhibited a significant direct positive effect. This apparent discrepancy does not reflect conflict, but rather the complementary strengths of the two approaches: random forests quantify total predictive power, including variance shared via collinearity with other predictors,53,54 whereas structural equation modeling isolates direct, independent causal effects after accounting for all specified pathways.55 In this long-term engineered succession, SOM and TP rise in lockstep with operational years; consequently, the variance they share is credited by SEM to the single dominant driver that ultimately generates them both, operational years itself. In other words, the non-significant direct paths from SOM and TP do not undermine their critical ecological roles; rather, they reveal that their regulatory effects on nitrogen are fully subsumed within, and ultimately governed by the dominant temporal forcing of long-term PV succession. Crucially, it is important to clarify the ecological interpretation of this significant direct path from “Year” to “Nitrogen.” While time itself is not a biogeochemical reactant, our alternative model analysis (Table S4) demonstrated that physicochemical factors alone (SOM, TP, pH, and SWC) could not fully explain the observed nitrogen accumulation (RMSEA >0.12). This statistical residual suggests that “Year” acts as a proxy for unmeasured, time-dependent biological processes. We hypothesize that the progressive succession of N-fixing microbial communities (e.g., biological soil crusts) and the accumulation of soil enzymes, which were not directly quantified in this study, likely constitute the “missing link” driving the unexplained portion of nitrogen recovery. Previous studies in similar desert PV systems have confirmed that microbial abundance and enzymatic activity increase significantly with array age,32,56,57 supporting our inference that these biotic factors are the underlying drivers represented by the “Year” path. RF results further indicate that Pre and DI significantly influence TN and NO3−-N, while Pre and ST significantly influence NH4+-N. This aligns with the climate-driven nitrogen cycling characteristics observed in desert ecosystems.58 However, only climatic factors (Pre + DI or Pre + ST) in SEM exhibited a weakly significant direct effect on nitrogen. This disparity reflects the fundamental distinction between total predictive power and direct causal effect. Random forests ranked climatic variables highly because they covary strongly with interannual variation in water and thermal regimes.8,59 In contrast, SEM revealed that once the dominant effect of operational years was accounted for, the independent direct effects of these climatic factors on nitrogen fractions were greatly diminished or non-significant. This indicates that while interannual climatic conditions remain important contributing factors, their proximate effects on nitrogen and its inorganic components are primarily mediated through PV engineering disturbances that accumulate over time. This 14-year time-series study reveals a previously overlooked yet critical pattern in the evolution of soil nitrogen within desert PV power plants. PV array age is the primary driver of soil nitrogen accumulation and also exerts indirect effects by regulating soil nutrients and climatic factors. Differences in panel position lead to sharply contrasting trajectories of inorganic nitrogen fractions: NH4+-N is rapidly and persistently retained in UP zone, whereas NO3−-N becomes the main pathway of nitrogen loss in BP and FL zones. Our results highlight the necessity of position-specific nitrogen management in long-term PV projects. Based on these findings, we propose several practical, low-cost management interventions to optimize soil nitrogen retention and ecosystem stability: (1) preserve natural microtopography, retain the natural runoff-accumulating function of FL and BP zones by avoiding excessive artificial leveling. (2) Minimize mechanical disturbance, limit physical disruption to facilitate the natural recovery of biological soil crusts, which effectively enhance nitrogen fixation. And (3) introduce native vegetation, sow low-growing, drought-tolerant pioneer species in FL and BP zones to stabilize soil nitrogen through biomass uptake, thereby improving ecosystem multifunctionality. Ultimately, our study advances the mechanistic understanding of soil–nutrient interactions under sustained anthropogenic disturbance, providing a critical scientific foundation for harmonizing large-scale PV deployment with ecological restoration in arid regions. Limitations and future research To date, few studies have documented the decadal-scale cumulative effects of large-scale PV development on soil nitrogen fractions and their dynamics in desert ecosystems. However, several aspects warrant further investigation to fully generalize these findings. First, while our results from sandy desert ecosystems offer a robust baseline, the applicability to other arid soil types (e.g., gravel deserts and saline-alkali soils) remain to be tested. Future chronosequence studies across diverse edaphic gradients are needed to validate the universality of the “PV-induced nitrogen recovery” model. Second, our focus on bulk soil properties successfully identified significant nitrogen accumulation, yet the underlying microbial mechanisms remain inferred. Crucially, our structural equation modeling revealed that physicochemical factors alone were insufficient to fully explain the observed nitrogen recovery, suggesting that the direct effects of array age act as strong statistical proxies for unmeasured biological drivers. Therefore, future research should integrate high-throughput microbial sequencing and soil enzyme assays to explicitly quantify these “missing links,” thereby moving from statistical inference to mechanistic verification. Third, our sampling strategy targeted the surface soil layer (0–20 cm), which represents the most active zone for microclimate-driven changes and biological soil crust development. While this approach effectively captured the primary signal of ecosystem recovery, we acknowledge that deep-rooted desert vegetation and potential nitrate leaching processes extend beyond this depth. To fully resolve the vertical migration of nitrate and distinguish leaching losses from deep plant uptake, future studies should incorporate deep-profile sampling (e.g., up to 100 cm) and monitor water infiltration dynamics. Ultimately, this multi-layered approach will not only reveal the comprehensive patterns of nitrogen storage and vertical distribution but also elucidate the key biogeochemical pathways driving ecosystem recovery, thereby informing more precise, site-specific management strategies for sustainable desert solar development. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jianhua Xiao (xiaojh@lzb.ac.cn). Materials availability This study did not generate new unique reagents. Data and code availability The soil data reported in this study have been deposited at Mendeley Data and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. All original R scripts for data processing have been deposited at Zenodo and are publicly available as of the date of publication. DOIs are listed in the key resources table. Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request. Acknowledgments This research was funded by the Key Science and Technology Project of China Huadian Corporation Ltd. (HDKJ23-04-01-61), Gansu Provincial Science and Technology Planning Project (23ZDFA018 and 25YFFA007), National Natural Science Foundation of China (U23A2063), and the UK Royal Geographical Society with Institute of British Geographers Thesiger-Oman Fellowships grant (THES 03.26). Author contributions J.X.: data curation, formal analysis, investigation, software, visualization, writing – original draft, writing–review and editing, funding acquisition, and project administration; A.W.: formal analysis, investigation, software, visualization, and writing – original draft; F.S.: data curation, investigation, methodology, formal analysis, and writing – review and editing; Y.T.: conceptualization, investigation, methodology, and formal analysis; Q.T.: conceptualization, methodology, validation, and formal analysis; T.C.: investigation, methodology, and formal analysis; F.L.: investigation and methodology; Z.C.: investigation, methodology, validation, and formal analysis; S.M.: investigation, methodology, and formal analysis. H.Y.: conceptualization, methodology, validation, and writing – review and editing; L.C.: conceptualization, funding acquisition, project administration, resources, validation, supervision, and writing – review and editing. All authors edited the final manuscript. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDataSoil dataMendeley Datahttps://doi.org/10.17632/xpmnxdr986.2Software and algorithmsR v4.4.2https://www.r-project.orgRRID: SCR_001905RStudiohttps://www.rstudio.com/RRID: SCR_00432Data analysis and visualizationThis paper; Zenodohttps://doi.org/10.5281/zenodo.19447437 Method details Study area Northwest China’s deserts have become the country’s primary region for utility-scale photovoltaic deployment and remain the area with the greatest potential for further expansion.17,60,61We selected a continuous 1–14 years chronosequence using eight PVPS located in environmentally comparable sandy-desert regions of Northwest China (102°28′-109°42′E, 37°33′-40°26′N) (Figure 1). To ensure the validity and reliability of the space-for-time substitution approach, we applied strict and consistent site-selection criteria to minimize confounding effects from spatial heterogeneity, environmental differences, and installation variability. First, standardized PV configuration: To ensure consistent shading intensity, all stations utilized standardized fixed-tilt monocrystalline silicon panels with a uniform south-facing orientation (180°) and a consistent tilt angle of 36 ± 1°. Panel layout, inter-row spacing, and installation height were also maintained consistent across the chronosequence. Operation and maintenance were conducted under standardized protocols by major state-owned enterprises. Second, climatic comparability was guaranteed: All eight stations were situated within the same arid climatic zone (Aridity Index 0.06–0.11), characterized by comparable annual precipitation (113–265 mm, CV <24%), mean temperature (7.8°C–8.4°C, variation <0.3°C), and evaporation (1,913–2,300 mm).62 Annual sunshine duration exceeds 2,800–3,300 h with daily radiation ≥5.5 kW h m−2, and prevailing northwesterly winds frequently trigger dust storms in spring.63,64 Detailed site characteristics and climatic parameters are provided in Table S1. Meteorological data, including annual precipitation and annual mean temperature, were primarily obtained from two national-level authoritative databases: the China Meteorological Data Service Center (CMDSC, http://data.cma.cn) and the National Ecosystem Science Data Center (NESDC, http://www.nesdc.org.cn).65 Pre-installation vegetation cover was consistently <5%, dominated by mobile dunes and scattered pioneer psammophytes (e.g., Agriophyllum squarrosum, Artemisia spp.).40,62,66 Most importantly, we rigorously verified the homogeneity of initial soil background properties. All selected PV were originally constructed on flat mobile dunes with identical soil taxonomy (Aeolian sandy soil) and highly uniform initial texture (sand content >90%). A one-way ANOVA performed on adjacent reference plots (pristine mobile dunes located ∼300 m outside each station) confirmed no significant differences in key initial soil properties across the eight sites, including Total Nitrogen (p = 0.958), Ammonium Nitrogen (p = 0.857), Nitrate Nitrogen (p = 0.312), Soil Organic Matter (p = 0.867), and Soil Water Content (p = 0.(662) (Table S2). Although minor variations in Total Phosphorus and pH were observed (p < 0.05), likely reflecting local geological heterogeneity, our strict paired sampling design and the calculation of Response Ratios (RR) relative to the local control for each site effectively normalized these background variations. Soil sampling and analytical methods Soil sampling was conducted in August 2024. At each photovoltaic station, ten spatially independent replicate plots (>50 m apart) were established, positioned at least 100 m from the station boundary to eliminate edge effects. Within each plot, soil samples were collected from three distinct microhabitats: UP, BP, and FL. A five-point composite sampling method was employed: after removing surface litter and gravel, five soil cores (0–20 cm depth) were collected using a 5 cm diameter stainless steel auger and thoroughly homogenized into a single composite sample per microhabitat per plot (n = 10 per microhabitat per station). Concurrently, a strict paired control design was implemented by collecting ten composite samples from undisturbed mobile dunes (MD) located approximately 300 m upwind of each station, serving as the pristine edaphic baseline for that specific array age. The 0–20 cm sampling depth was strategically selected to capture the most responsive interface for nitrogen dynamics, which is dominated by rapid microclimatic modification, biological soil crust (BSC) development (concentrated in 0–5 cm), and shallow herbaceous root turnover.56,67 In total, 320 composite soil samples were collected for this study. Approximately 50 g of fresh soil was stored at 4°C for the determination of NH4+-N, NO3−-N, and soil water content (SWC), while the remaining soil was air-dried.10,26 Soil water content was determined gravimetrically by oven-drying at 105°C to constant weight.68 Soil pH was measured potentiometrically in a 1:2.5 soil:water suspension, and electrical conductivity (EC) was determined using a conductivity meter in a 1:5 soil:water suspension.67 Ammonium nitrogen (NH4+-N) and nitrate nitrogen (NO3−-N) were extracted with 2 mol L−1 KCl and quantified using a continuous-flow analyser (AA3, SEAL Analytical, Norderstedt, Germany).69 Soil organic carbon (SOC) was assayed by the Walkley-Black K2Cr2O7-H2SO4 oxidation method. Total nitrogen (TN) was determined by the Kjeldahl method after concentrated H2SO4 digestion, and total phosphorus (TP) via the alkali fusion-molybdenum antimony colorimetric method. Available phosphorus (AP) was measured by NaHCO3 extraction coupled with molybdenum-antimony colorimetry.26,38,70 Quantification and statistical analysis Data normality and basic statistics To assess the normality of the data, we conducted the Kolmogorov-Smirnov test. We then employed the response ratio (RR) to quantify differences in soil nitrogen dynamics and other physicochemical properties between the photovoltaic panel positions (BP/FL/UP) and the control zone (MD). where Xtreat represents the measured values in the treatment groups (BP/FL/UP), and XMD denotes the arithmetic mean of the MD control group in the same year. To eliminate interference from interannual climate fluctuations, a temporal baseline correction was applied. A t test was conducted on RR values (null hypothesis: RR = 0), where RR > 0 indicates an enrichment effect, and RR < 0 explains a depletion effect. Analyses were performed in tidyverse v.2.0.0. The specific output data structure of the RR values is presented in Table S3. Modeling and regression Non-linear trends were visualized using locally estimated scatterplot smoothing (LOESS) regression with 95% confidence intervals. Linear regression and correlation analyses were used to examine the relationship between PV array age and soil nitrogen RR values. Linear mixed-effects models (LMM) were implemented using the glmm.hp package in R to quantify the independent and interactive effects of array age and panel position.71 Variation partitioning analysis (VPA) was used to identify the variance contributions of individual factors. Machine learning and structural equation modeling Random Forest (RF) analysis was conducted using the rfPermute package in R to identify the key drivers of soil nitrogen variation. To ensure model stability and reproducibility, the number of trees (ntree) was set to 500, and the number of permutations for importance significance (num.rep) was set to 500. The number of variables sampled at each split (mtry) was set to the default value (p/3). Model accuracy was evaluated using the Out-of-Bag (OOB) error estimate, and predictor importance was quantified based on the percentage increase in mean squared error (%IncMSE).54 Guided by these variable-importance patterns, we then applied partial least squares structural equation modeling (PLS-SEM) to disentangle the direct and indirect effects of array age and the main climatic and soil properties drivers on TN and its inorganic fractions. The variable selection for SEM path settings followed three principles to ensure logical rationality: (1) The top five variables with the highest %IncMSE were chosen as core candidates to ensure significant explanatory power. (2) We constructed latent variables based on theoretical hypotheses (e.g., “Climate” from Pre and DI/ST) to simplify the model structure and reduce multicollinearity. (3) A sensitivity analysis on each candidate path allowed for the exclusion of variable combinations with poor fitting significance or unstable paths.72 To address multicollinearity (Variance Inflation Factor, VIF >(5) between precipitation (Pre) and the drought index (DI), a new composite variable, “Climate”, was created by standardizing and summing the relevant parameters. Specifically, Climate was calculated as Z(Pre) + Z(DI) for models of TN and NO3−-N, and as Z(Pre) + Z(ST) for the NH4+-N pathway. An interaction term (Year×Location) was also constructed to test the interaction between array years and panel positions. All latent variables were reflective indicators, and the model utilized a single-indicator measurement approach (directly utilizing observed variables). The statistical process included 500 bootstrap resamplings for stability assessment, and path significance was tested with the 95% confidence interval method (considered significant if CI did not include 0).55 The Goodness-of-Fit (GoF) index evaluated the overall PLS-SEM model fit. All graphs were produced in R 4.4.3, utilizing packages such as “plspm”, “randomForest”, “rfUtilities”, and “GGally”. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.116216. Contributor Information Jianhua Xiao, Email: xiaojh@lzb.ac.cn. Longfei Chen, Email: chenlongfei_mail@163.com. Supplemental information Document S1. Tables S1–S4 References Associated Data Supplementary Materials Document S1. Tables S1–S4 Data Availability Statement The soil data reported in this study have been deposited at Mendeley Data and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. All original R scripts for data processing have been deposited at Zenodo and are publicly available as of the date of publication. DOIs are listed in the key resources table. Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request.