A method for refining construction and parameter optimization of a photovoltaic land surface process model suitable for desert areas
By embedding a photovoltaic-specific parameterized module into the CLM framework and optimizing key parameters, the accuracy problem of existing models in simulating desert photovoltaic power plants was solved, achieving high-precision climate-carbon cycle simulation and virtual simulation of photovoltaic carbon sequestration technology.
Patent Information
- Application Number
- CN202611565040.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-10
- Publication Date
- 2026-11-10
AI Technical Summary
Existing land surface process models fail to fully consider the key physical processes brought about by photovoltaic arrays when simulating desert photovoltaic power plants, resulting in low simulation accuracy and failing to meet the evaluation requirements of desert photovoltaic ecological carbon sequestration technology.
A photovoltaic land surface process model adapted to desert areas was constructed. Through the construction of multi-source observation datasets, photovoltaic-specific parameterization schemes, parameter optimization and model verification, and embedding the CLM framework, the model fully simulates the radiation capture, shading attenuation, panel heat transfer and precipitation redistribution processes of photovoltaic arrays, and optimizes key parameters.
It significantly improves simulation accuracy, reducing carbon flux simulation error from ±45% to ±12%, supports virtual simulation of photovoltaic carbon sequestration technology, and provides a reliable climate-carbon cycle assessment tool.
Smart Images

Figure CN123069718A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of land-atmosphere interaction numerical simulation, desert ecometeorology, and carbon cycle simulation technology. Specifically, it relates to a high-resolution land surface process model development and parameter optimization method driven by multi-source observation data, tailored to the unique underlying surface of desert photovoltaic power plants. This invention can be applied to the assessment of local climate effects of desert photovoltaic power plants, simulation of water and heat fluxes, quantitative analysis of carbon cycles, and virtual simulation and optimized deployment of photovoltaic ecological carbon sequestration technology solutions. Background Technology
[0002] Land surface process models (such as CLM developed by the National Center for Atmospheric Research and Noah-MP of the National Center for Environmental Prediction) are core numerical tools for simulating energy balance, water and heat exchange, and carbon flux exchange between the Earth's surface and the atmosphere. They are widely used in climate change assessment, ecological process simulation, and carbon cycle research.
[0003] In recent years, with the large-scale construction of photovoltaic power stations in the desert areas of Northwest my country, the impact of photovoltaic arrays on local microclimates, water and heat balance, and carbon cycle processes has received increasing attention. Existing research shows that large-scale photovoltaic systems produce significant regional biogeophysical impacts by altering underlying surface properties such as albedo, roughness, and vegetation cover. Photovoltaic facilities significantly influence local microclimates by changing surface radiation budget and energy distribution. In desert regions, the drastic changes in underlying surface properties before and after the construction of photovoltaic power stations—from bare soil / sparse vegetation to a composite underlying surface of photovoltaic panels and artificial vegetation—completely alter the original water and heat budget balance, thereby regulating the regional ecosystem's carbon cycle processes.
[0004] Existing land surface process models are primarily designed for natural underlying surfaces such as forests, grasslands, farmland, and bare soil, neglecting the unique physical processes brought about by photovoltaic (PV) arrays. Current climate models simulating PV-induced climate effects mostly modify surface albedo to characterize PV radiative forcing, ignoring land-atmosphere interactions such as turbulent transport and soil water and heat transfer brought about by PV arrays, resulting in simulation biases. Key missing physical processes include: solar radiation capture, absorption, and reflection by PV panels; surface radiation attenuation and spatial redistribution due to PV panel shading; convective heat exchange between PV panel surface and air; precipitation interception, diversion, and redistribution by PV panels; and aerodynamic roughness and turbulent exchange characteristics altered by PV arrays.
[0005] Due to the lack of physical processes, the general land surface model used for simulating photovoltaic power plants has large errors. Field observations show that the net radiation simulation deviation of the general model for photovoltaic power plants can reach more than 30%, and the net ecosystem carbon exchange (NEE) simulation deviation exceeds 50%, which cannot meet the needs of quantitative assessment of the carbon sequestration effect of desert photovoltaics.
[0006] Meanwhile, there is currently a lack of systematic long-term multi-site observation datasets of the underlying surface of photovoltaic power plants, and a lack of parameterization schemes specific to photovoltaic scenarios. Desert photovoltaics are distributed in extremely arid and vulnerable areas, with significant differences in light, precipitation, and temperature in different regions. Most of the key parameters related to vegetation, stomata, and soil in existing models come from the natural ecosystems of humid areas, and there is a lack of localized parameter systems for desert photovoltaic scenarios.
[0007] In summary, the core contradiction of existing technologies is that the research and evaluation of desert photovoltaic ecological carbon sink enhancement technology urgently requires high-precision numerical simulation tools, but existing land surface models are poorly adapted to desert photovoltaic scenarios, have low simulation accuracy, and lack specific parameters, making it difficult to quantify the local climate-carbon cycle impact mechanism of photovoltaic power plants and unable to provide a reliable platform for virtual simulation of carbon sink enhancement technology. Summary of the Invention
[0008] To address the shortcomings of existing technologies, this invention provides a method for the refined construction and parameter optimization of photovoltaic land surface process models adapted to desert regions. It solves four major technical problems: constructing multi-source observation datasets for desert photovoltaics, lack of parameterization for photovoltaic array-specific physical processes, localized calibration of model parameters, model verification, and analysis of photovoltaic-desert interaction mechanisms. This enables high-precision simulation of local climate, water and heat transfer, and carbon flux in desert photovoltaic power plants.
[0009] The technical solution of this invention: A method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas, comprising the following steps: S1. Screening of typical sample areas and construction of a multi-source observation network; S2. Construction of photovoltaic-specific parameterized solutions and embedding of CLM framework; S3, hierarchical iterative optimization of key parameters; S4. Model Validation and Mechanism Scenario Simulation.
[0010] The typical sample area of S1 includes: ① the oasis edge transition zone, with an average annual precipitation of 200-250 mm; ② the Gobi desert zone, with an average annual precipitation of 150-180 mm; ③ the extremely arid desert zone, with an average annual precipitation of less than 50 mm. Multi-source observation network deployment: A comparative observation system is deployed inside and in the surrounding natural desert area of each photovoltaic power station; a main observation tower is set up inside the power station, equipped with an open-circuit eddy covariance system, a four-component radiometer, a spectroradiometer, and a multi-layer wind speed, wind direction, temperature, humidity, and pressure gradient observation system; auxiliary observation points are set up under and between the panels, equipped with small automatic weather stations, soil water and heat observation systems, and soil respiration observation devices; background observation towers are deployed in the natural desert area 1-2 km away from the boundary of the power station, and the main observation tower and the background observation tower are equipped with the same observation equipment; Data Acquisition and Preprocessing: Daily scale observation data were continuously acquired for no less than two complete calendar years; outlier removal, ultrasonic virtual temperature correction, quadratic coordinate rotation, time delay correction, WPL air density effect correction, and frequency response correction were performed on the 10Hz high-frequency raw data to output the 30-minute average flux; flux data quality control was completed to remove precipitation periods, instrument malfunctions, and low-quality data, and a desert photovoltaic multi-source observation dataset containing four major categories of variables: radiation, meteorology, soil, and flux was constructed.
[0011] The S2 specifically includes: (1) selecting Community Land Model of CLM5.0 and above as the underlying basic model; (2) adding four photovoltaic-specific parameterized modules to the CLM source code: photovoltaic panel radiation capture module, shading attenuation module, panel-atmosphere heat exchange module, and precipitation redistribution module; the photovoltaic panel radiation capture module takes into account the photovoltaic panel tilt angle, azimuth angle, photovoltaic panel geometric dimensions and spacing, and photovoltaic panel transmittance / reflectance spectral characteristics, calculates the solar altitude angle and azimuth angle in combination with the solar position model, calculates the interception, absorption, and reflection of direct radiation by the photovoltaic panel, and outputs the net radiation flux of the photovoltaic panel surface and the panel surface temperature; the shading attenuation module divides the land surface into the projection area under the panel and the non-shading area between the panels, calculates the direct radiation transmission ratio and the scattered radiation ratio respectively, and outputs the total radiation of the land surface under the panel and the photosynthetically effective radiation attenuation. Coefficient; The panel-atmosphere heat exchange module, based on the temperature difference between the photovoltaic panel and the ambient air, combined with wind speed and the geometric characteristics of the photovoltaic panel, uses the forced convection empirical formula to solve the convective heat transfer coefficient, and outputs the photovoltaic panel-atmosphere sensible heat flux and the daily variation of the panel temperature; The precipitation redistribution module sets the precipitation interception evaporation coefficient, the panel flow ratio, and the edge dripping area humidification coefficient, and outputs the soil moisture input and spatial non-uniformity coefficient at different locations under the panel; (3) The four photovoltaic exclusive parameterization modules are embedded into the CLM source code, the surface type identifier of the desert photovoltaic power station is added, the mapping relationship between the photovoltaic array geometric parameters and the mode grid is established, and the energy balance and water balance equations of the surface flux main program are modified to obtain the CLM_PV photovoltaic land surface process mode.
[0012] The hierarchical iterative optimization of the S3 key parameters includes: (1) identifying the key parameter group to be optimized, including the vegetation parameter group, stomatal conductance parameter group, soil parameter group, and photovoltaic parameter group; the vegetation parameter group includes vegetation functional type, vegetation coverage, leaf area index, and canopy height; the stomatal conductance parameter group includes maximum carboxylation rate Vcmax, maximum electron transport rate Jmax, and stomatal conductance slope parameter; the soil parameter group includes soil thermal conductivity, soil hydraulic diffusion coefficient, and saturated hydraulic conductivity; the photovoltaic parameter group includes photovoltaic panel albedo, convective heat transfer coefficient, and radiative transmittance; (2) using Bayesian parameter inversion combined with particle The hierarchical iterative optimization strategy of sequential data assimilation by sub-filter; using the observation dataset obtained by S1 to construct a likelihood function with the goal of minimizing the root mean square error between simulated and observed values, combined with the prior distribution of parameters, the posterior probability distribution and optimal estimate of parameters are obtained by Monte Carlo sampling of MCMC Markov chain; using particle filtering to sequentially assimilate time-series observation data and dynamically update parameter status; prioritizing the calibration of high-sensitivity parameters and then calibrating low-sensitivity parameters; (3) based on different latitudes of different regions, photovoltaic array density, drought index, and annual temperature difference, a parameter localization lookup table with the mapping relationship between station-parameter group-optimal value is generated.
[0013] The S4 model validation and mechanism scenario simulation include: (1) Independent model validation: using independent observation data that has not participated in parameter calibration, the simulation effect of the CLM_PV model on 2m air temperature, specific humidity, wind speed, net radiation, sensible heat flux, latent heat flux, and net ecosystem carbon exchange (NEE) is validated. The evaluation index includes the coefficient of determination R. 2 Root mean square error (RMSE); the optimized model metrics satisfy: R 2 ≥0.85, RMSE≤10%, carbon flux simulation error controlled within ±12%; (2) Set up two sets of comparative scenario simulations: Scenario A is a photovoltaic scenario, and the CLM_PV all photovoltaic exclusive modules are enabled to simulate the underlying surface of the photovoltaic power station; Scenario B is a non-photovoltaic scenario, and the photovoltaic exclusive modules are turned off to simulate the underlying surface of the natural desert before construction; fix the driving data, initial conditions, and time steps of the two scenarios, compare the output results, quantify the impact of the photovoltaic power station on the local microclimate, water and heat flux, and carbon flux, and analyze the radiation-temperature-moisture-carbon coupling contribution chain; (3) Carry out virtual simulation of carbon sink enhancement technology, simulate carbon flux response under different plate spacing, vegetation configuration, and soil microbial regulation measures, and output desert photovoltaic carbon sink enhancement optimization scheme.
[0014] Preferably, the sensor installation height of the main observation tower is at least 5m higher than the highest point of the photovoltaic panel; no less than 3 sets of repeated observations are set at the auxiliary observation points under the panel and between the panels, and the average value is used to represent the microhabitat observation results.
[0015] Furthermore, the four photovoltaic-specific parameterized modules are written as independent subroutines, and data transfer between modules is completed through the built-in clm_varpar and clm_varcon data structures of CLM.
[0016] Preferably, Bayesian parameter inversion uses the DREAM or MCMC toolbox, and particle filtering uses the PF-Lab open-source library; single-site parameter calibration uses a high-performance computing node with 32 cores or more, running continuously for 7-14 days, and is accelerated by parallel computing.
[0017] Furthermore, 20%-30% of the time period is reserved from the original observation data as independent verification data, covering multiple seasons and multiple weather types.
[0018] The beneficial effects of this invention are as follows: 1. For the first time, a dedicated land surface process model CLM_PV is developed for desert photovoltaics. The four major physical processes of photovoltaic radiation capture, shading attenuation, panel heat exchange, and precipitation redistribution are fully embedded into the CLM open source framework, filling the gap in the existing models that lack key physical processes of photovoltaic arrays and solving the pain point that traditional models cannot characterize the interaction between photovoltaics and desert land and atmosphere.
[0019] 2. Significantly improved simulation accuracy: Based on multi-source observation data from multiple stations in the eastern, central, and western parts of the Hexi Corridor for more than two years, Bayesian inversion + particle filtering hierarchical iterative parameter calibration was adopted; the carbon flux simulation error was reduced from ±45% in the general model to within ±12%, and the key flux variable R... 2 With a value ≥0.85, the RMSE is reduced by more than 40% compared to the general CLM, providing a reliable numerical tool for the quantitative assessment of desert photovoltaic carbon sinks.
[0020] 3. Quantitative analysis of the photovoltaic-desert interaction mechanism: By controlling the comparison of scenarios with and without photovoltaics, the independent ecological and climatic effects brought about by photovoltaic construction are separated, and the causal chain of radiation attenuation-temperature change-water redistribution-carbon flux response is fully analyzed, providing a simulation platform for mechanism research.
[0021] 4. Supports virtual simulation of photovoltaic carbon dioxide enhancement technology; it can conduct virtual experiments on schemes such as panel spacing, vegetation configuration, and soil regulation, reducing the time and economic cost of field experiments, and supporting the standardized process output and regionally differentiated deployment of desert photovoltaic ecological carbon dioxide enhancement.
[0022] 5. Strong scalability: Developed based on the CLM open-source modular architecture, it has good portability; with a supporting localized parameter lookup table, it can be extended to the desert photovoltaic areas of Xinjiang and Inner Mongolia, providing a general technical tool for the assessment of the ecological effects of desert photovoltaics and carbon sink management nationwide. Attached Figure Description
[0023] Figure 1 This is the logical relationship diagram of the four photovoltaic-specific parameterized modules in the CLM_PV mode; Figure 2 This is a schematic diagram of the hierarchical iterative optimization process for key parameters; Figure 3 This is a schematic diagram illustrating the simulation results of a scenario comparing "with and without photovoltaics". Detailed Implementation
[0024] A refined construction and parameter optimization method for photovoltaic land surface process models adapted to desert areas, comprising four core steps S1-S4: S1 Typical Sample Area Selection and Multi-Source Observation Network Construction (1) Typical sample area selection: Photovoltaic power stations in the three climate gradient deserts of the Hexi Corridor in Gansu Province were selected as sample areas to ensure the representativeness of the observation area: Eastern sample area: Photovoltaic power station in Gulang County (oasis edge transition zone, annual precipitation 200-250mm); Central sample area: Photovoltaic power station in Shandan County (typical Gobi desert zone, annual precipitation 150-180mm); Western sample area: Photovoltaic power station in Dunhuang City (extremely arid desert zone, annual precipitation <50mm).
[0025] (2) Multi-source observation equipment is deployed inside each photovoltaic power station and in the surrounding natural desert to set up a pairwise comparison observation system. The main observation tower inside the power station is located in the middle of the power station. The tower is equipped with: a 10Hz sampling open-circuit eddy covariance system (three-dimensional ultrasonic anemometer, CO2 / H2O infrared gas analyzer) to observe sensible heat, latent heat, CO2, and momentum flux; a four-component radiometer to observe long and short wave up and down radiation; a spectroradiometer to obtain photosynthetically active radiation (PAR); and a multi-layer gradient observation system to realize wind speed, wind direction, temperature, humidity, and air pressure observation at heights of 2m, 5m, 10m, 20m, and 30m. Small automatic weather stations and soil water and heat observation systems are set up at auxiliary observation points under and between the panels; soil temperature observation gradients are 5cm, 10cm, 20cm, 40cm, and 80cm; soil moisture content observation gradients are 10cm, 20cm, 40cm, and 80cm; soil heat flux plates are buried at 5cm and 20cm; and a static box soil respiration observation device is provided to verify the eddy carbon flux results. The peripheral background observation tower is located 1-2 km outside the power station boundary, with the underlying surface consistent with the original desert and Gobi before the station's construction. The configuration of the observation equipment is exactly the same as that of the main observation tower of the power station. The sensor installation height of the observation tower is at least 5 m higher than the highest point of the photovoltaic panel; no less than 3 sets of repeated observations are set at the points under the panel and between the panels, and the average value is used to characterize the microhabitat.
[0026] (3) Data acquisition and preprocessing: Data was continuously acquired for no less than two complete natural years, covering daily-scale observation data during the transition between cold and warm seasons; 10Hz high-frequency raw flux data was processed using EddyPro software, performing outlier removal, ultrasonic virtual temperature correction, quadratic coordinate rotation, time delay correction, WPL air density correction, and frequency response correction, and outputting 30-minute average flux; flux quality control was carried out to remove precipitation, instrument failure, and low-quality data; finally, a desert photovoltaic multi-source observation dataset was constructed, which includes four major categories of variables: radiation, meteorology, soil, and flux, providing a foundation for subsequent parameter development and model calibration.
[0027] Construction of photovoltaic-specific parameterized solutions and embedding of CLM framework (1) Basic model selection: CommunityLandModel of CLM5.0 or higher is selected as the underlying framework. The modular code structure of this model is easy to develop and has mature land surface physical process simulation capabilities.
[0028] (2) Four photovoltaic-specific parameterized modules were developed: ① Photovoltaic panel radiation capture module: Input photovoltaic panel tilt angle θ, azimuth angle φ, photovoltaic panel geometry, panel spacing, and spectral transmittance / reflectivity; call the solar position model to calculate the solar altitude angle and azimuth angle at each moment; simulate the process of photovoltaic panel interception, absorption, and reflection of direct radiation; output the net radiative flux of photovoltaic panel surface and panel temperature. ② Shading attenuation module: Divide the surface in the grid into the projected shading area under the panel and the non-shading area between the panels; calculate the direct radiation transmission ratio and the scattered radiation distribution ratio of the two areas respectively; output the total surface radiation under the panel and the photosynthetically effective radiation attenuation coefficient, realizing the spatial redistribution of surface radiation under photovoltaic shading. ③ Panel-atmosphere heat exchange module: Based on the difference between photovoltaic panel surface temperature and ambient air temperature, combined with wind speed and photovoltaic panel geometry, the convective heat transfer coefficient is solved using the forced convection empirical formula; calculate the sensible heat exchange between the photovoltaic panel and the atmosphere; output the photovoltaic panel-atmosphere sensible heat flux and the daily variation of panel surface temperature. ④ Rainfall redistribution module: Set the precipitation interception and evaporation coefficient on the panel surface, the flow diversion ratio on the panel surface, and the humidification coefficient of dripping at the panel edge; simulate the redistribution of precipitation intercepted by the photovoltaic panel and dripping along the panel surface; output the soil moisture input at different locations under the panel, as well as the spatial non-uniformity coefficient of soil moisture.
[0029] (3) CLM_PV Model Integration: The four modules are developed as independent subroutines, following the CLM code specifications. Data interaction between modules is completed through the built-in clm_varpar and clm_varcon data structures of CLM. The NetCDF format underlying surface dataset is modified using the CLM's built-in surface dataset compilation tool to expand the surface attribute parameter library. The CLM underlying surface database is modified to add an underlying surface type identifier for "desert photovoltaic power station". Geometric parameters such as photovoltaic array tilt angle, panel height, and coverage extracted by remote sensing are aggregated to the model grid through area weighting. A grid-photovoltaic parameter lookup table is constructed to establish the mapping relationship between photovoltaic array geometric parameters and model grid. The subgrid type judgment branch of the surface flux main program is modified to add desert photovoltaic-specific calculation logic. The surface flux main program is modified to introduce photovoltaic radiation interception, photoelectric conversion, and under-panel water and heat redistribution source and sink terms into the original energy and water balance equations. The photovoltaic physical process is coupled into the model energy balance and water balance equations to obtain the CLM_PV photovoltaic land surface process model.
[0030] Hierarchical Iterative Optimization of Key Parameters (1) Identify the key parameter group to be calibrated, see Table 1 Table 1. Specific parameters and dominant physical processes of the key parameter group (2) Hierarchical Iterative Parameter Optimization Algorithm Combined with Bayesian Parameter Inversion and Particle Filter Sequential Assimilation: Bayesian Parameter Inversion: A likelihood function is constructed using the S1 measured dataset, with the objective of minimizing the root mean square error between simulation and observation; combined with the prior distribution of parameters, MCMC Markov chain Monte Carlo sampling is used to output the posterior probability distribution of each parameter and the optimal estimated value; Particle Filter Sequential Assimilation: Two years of time-series observation flux data are sequentially assimilated into the model, and the parameter state is dynamically updated to adapt to the time-varying characteristics of parameters with the environment; Hierarchical Strategy: High-sensitivity parameters (Vcmax, soil thermal conductivity, etc.) are calibrated first, followed by low-sensitivity parameters, balancing computational efficiency and optimization accuracy. Computational Tools: Bayesian inversion uses the DREAM / MCMC toolbox; particle filtering uses the PF-Lab open-source library; single-site parameter calibration uses high-performance computing nodes with more than 32 cores, running continuously for 7-14 days, prioritizing parallel computation.
[0031] (3) The localized parameter lookup table is constructed based on the latitude of the three sample areas, the photovoltaic panel coverage / panel spacing, the drought index, and the annual temperature range. A localized lookup table is generated to form a mapping relationship between "site-parameter group-optimal parameter value". In areas without dense observation sites, the parameters in the lookup table are directly retrieved to carry out simulation, thereby expanding the model area.
[0032] Model Validation and Mechanism Scenario Simulation (1) Independent Model Validation: 20%-30% of the observation period from the original observation data is reserved as an independent validation set. The validation set covers different seasons and weather types and is not used for parameter calibration. Evaluation variables: 2m air temperature, specific humidity, wind speed, net radiation Rn, sensible heat flux H, latent heat flux LE, net ecosystem carbon exchange NEE. Evaluation index: coefficient of determination R. 2 Root mean square error (RMSE), mean deviation (MBE); Expected performance indicators: R 2 ≥0.85; RMSE is reduced by more than 40% compared to the general CLM model; carbon flux simulation error is controlled within ±12%.
[0033] (2) Two simulation scenarios were set up to compare the presence and absence of photovoltaic (PV) power, ensuring that the driving data, initial conditions, and time steps were all consistent, with only the PV module switching on and off: Scenario A (with PV): CLM_PV enabled all PV-specific modules to simulate the real composite underlying surface of the PV power station; Scenario B (without PV): CLM_PV disabled the PV-specific modules to simulate the original desert and Gobi underlying surface before the station was built. The results of the two simulations were compared to quantitatively separate the contributions of the PV power station to local temperature and humidity, wind speed, water and heat flux distribution, and carbon flux; the complete coupled contribution chain of "radiation-temperature-moisture-carbon flux" was analyzed to reveal the PV-desert ecological interaction mechanism.
[0034] (3) Virtual simulation of carbon flux enhancement technology is carried out based on the verified CLM_PV model to conduct virtual scenario experiments, simulate carbon flux response under different photovoltaic panel spacing, different vegetation configuration schemes under the panels, and soil microbial regulation measures; complete the scheme screening without conducting large-scale field experiments, and output the desert photovoltaic ecological carbon flux enhancement optimization deployment scheme.
[0035] Example 1: Construction and Parameter Optimization of Desert Photovoltaic Model in Hexi Corridor, Gansu 1. The observation system was deployed and implemented using the Wuwei Jiuduntan Photovoltaic Desertification Control Base, the Shandan Gobi Photovoltaic Power Station, and the Dunhuang Gobi Photovoltaic Power Station as pilot areas. The main observation tower was installed away from the edge of the power station, and the sensor height was 5m higher than the highest point of the photovoltaic panel. Three sets of soil observation repeat points were set up under the panel and between the panels. A 1-2km perimeter native desert control observation tower was also provided. Continuous annual observations were carried out for more than 24 months. EddyPro was used to process the 10Hz eddy intensity raw data, and WPL correction, coordinate rotation and other quality control were completed to construct a desert photovoltaic multi-source dataset.
[0036] 2. The secondary development of the CLM-PV mode is based on the official CLM5.0 source code, and four independent photovoltaic subroutines are written; a new underlying surface type identifier is added; the geometric parameters of the photovoltaic panel (tilt angle, panel spacing, coverage) are mapped to the grid; the energy and water balance equations are modified to couple the photovoltaic panel radiation, heat exchange, and precipitation redistribution processes; and the modules call the native CLM structure to complete data interaction.
[0037] 3. Layered iterative parameter optimization: Bayesian MCMC inversion was carried out using the DREAM toolkit, and particle filtering time-series assimilation was implemented using PF-Lab; high-sensitivity parameters such as Vcmax and soil thermal conductivity were calibrated first, followed by low-sensitivity parameters; parallel computing was performed using a 32-core high-performance node, with each site running for approximately 10 days; a localized desert photovoltaic parameter lookup table was output based on the three sites.
[0038] 4. Model validation and scenario simulation: 25% of the observation period is reserved as an independent validation dataset, covering typical winter and summer seasons; R is calculated. 2 Accuracy assessments were completed using RMSE and MBE; two control simulations were set up with and without photovoltaics, with fixed driving forces and unchanged initial conditions; the impact of photovoltaics on local water and heat and carbon flux was quantified; virtual simulations were carried out with different panel spacing and under-panel vegetation configurations to output a photovoltaic ecological carbon sequestration technology solution suitable for the Hexi Corridor.
[0039] Although the present invention has been described in conjunction with embodiments, the present invention is not limited to the above embodiments. Any equivalent modifications to the technical solutions of the present invention that are within the scope of the inventive concept shall fall within the protection scope of the present invention.
Claims
1. A method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert regions, characterized in that, Includes the following steps: S1. Screening of typical sample areas and construction of a multi-source observation network; S2. Construction of photovoltaic-specific parameterized solutions and embedding of CLM framework; S3, hierarchical iterative optimization of key parameters; S4. Model Validation and Mechanism Scenario Simulation.
2. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 1, characterized in that, The typical sample area of S1 includes: ① the oasis edge transition zone, with an average annual precipitation of 200-250 mm; ② the Gobi desert zone, with an average annual precipitation of 150-180 mm; ③ the extremely arid desert zone, with an average annual precipitation of less than 50 mm. Multi-source observation network deployment: A comparative observation system is deployed inside and in the surrounding natural desert area of each photovoltaic power station; a main observation tower is set up inside the power station, equipped with an open-circuit eddy covariance system, a four-component radiometer, a spectroradiometer, and a multi-layer wind speed, wind direction, temperature, humidity, and pressure gradient observation system; auxiliary observation points are set up under and between the panels, equipped with small automatic weather stations, soil water and heat observation systems, and soil respiration observation devices; background observation towers are deployed in the natural desert area 1-2 km away from the boundary of the power station, and the main observation tower and the background observation tower are equipped with the same observation equipment; Data Acquisition and Preprocessing: Daily scale observation data were continuously acquired for no less than two complete calendar years; outlier removal, ultrasonic virtual temperature correction, quadratic coordinate rotation, time delay correction, WPL air density effect correction, and frequency response correction were performed on the 10Hz high-frequency raw data to output the 30-minute average flux; flux data quality control was completed to remove precipitation periods, instrument malfunctions, and low-quality data, and a desert photovoltaic multi-source observation dataset containing four major categories of variables: radiation, meteorology, soil, and flux was constructed.
3. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 1, characterized in that, The S2 specifically includes: (1) selecting Community Land Model of CLM5.0 and above as the underlying basic model; (2) adding four photovoltaic-specific parameterized modules to the CLM source code: photovoltaic panel radiation capture module, shading attenuation module, panel-atmosphere heat exchange module, and precipitation redistribution module; the photovoltaic panel radiation capture module takes into account the photovoltaic panel tilt angle, azimuth angle, photovoltaic panel geometric dimensions and spacing, and photovoltaic panel transmittance / reflectance spectral characteristics, calculates the solar altitude angle and azimuth angle in combination with the solar position model, calculates the interception, absorption, and reflection of direct radiation by the photovoltaic panel, and outputs the net radiation flux of the photovoltaic panel surface and the panel surface temperature; the shading attenuation module divides the land surface into the projection area under the panel and the non-shading area between the panels, calculates the direct radiation transmission ratio and the scattered radiation ratio respectively, and outputs the total radiation of the land surface under the panel and the photosynthetically effective radiation attenuation. Coefficient; The panel-atmosphere heat exchange module, based on the temperature difference between the photovoltaic panel and the ambient air, combined with wind speed and the geometric characteristics of the photovoltaic panel, uses the forced convection empirical formula to solve the convective heat transfer coefficient, and outputs the photovoltaic panel-atmosphere sensible heat flux and the daily variation of the panel temperature; The precipitation redistribution module sets the precipitation interception evaporation coefficient, the panel flow ratio, and the edge dripping area humidification coefficient, and outputs the soil moisture input and spatial non-uniformity coefficient at different locations under the panel; (3) The four photovoltaic exclusive parameterization modules are embedded into the CLM source code, the surface type identifier of the desert photovoltaic power station is added, the mapping relationship between the photovoltaic array geometric parameters and the mode grid is established, and the energy balance and water balance equations of the surface flux main program are modified to obtain the CLM_PV photovoltaic land surface process mode.
4. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 1, characterized in that, The hierarchical iterative optimization of the S3 key parameters includes: (1) identifying the key parameter group to be optimized, including the vegetation parameter group, stomatal conductance parameter group, soil parameter group, and photovoltaic parameter group; the vegetation parameter group includes vegetation functional type, vegetation coverage, leaf area index, and canopy height; the stomatal conductance parameter group includes maximum carboxylation rate Vcmax, maximum electron transport rate Jmax, and stomatal conductance slope parameter; the soil parameter group includes soil thermal conductivity, soil hydraulic diffusion coefficient, and saturated hydraulic conductivity; the photovoltaic parameter group includes photovoltaic panel albedo, convective heat transfer coefficient, and radiative transmittance; (2) using Bayesian parameter inversion combined with particle The hierarchical iterative optimization strategy of sequential data assimilation by sub-filter; using the observation dataset obtained by S1 to construct a likelihood function with the goal of minimizing the root mean square error between simulated and observed values, combined with the prior distribution of parameters, the posterior probability distribution and optimal estimate of parameters are obtained by Monte Carlo sampling of MCMC Markov chain; using particle filtering to sequentially assimilate time-series observation data and dynamically update parameter status; prioritizing the calibration of high-sensitivity parameters and then calibrating low-sensitivity parameters; (3) based on different latitudes of different regions, photovoltaic array density, drought index, and annual temperature difference, a parameter localization lookup table with the mapping relationship between station-parameter group-optimal value is generated.
5. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 1, characterized in that, The S4 model validation and mechanism scenario simulation include: (1) Independent model validation: using independent observation data that has not participated in parameter calibration, the simulation effect of the CLM_PV model on 2m air temperature, specific humidity, wind speed, net radiation, sensible heat flux, latent heat flux, and net ecosystem carbon exchange (NEE) is validated. The evaluation index includes the coefficient of determination R. 2 Root mean square error (RMSE); the optimized model metrics satisfy: R 2 ≥0.85, RMSE≤10%, carbon flux simulation error controlled within ±12%; (2) Set up two sets of comparative scenario simulations: Scenario A is a photovoltaic scenario, and the CLM_PV all photovoltaic exclusive modules are enabled to simulate the underlying surface of the photovoltaic power station; Scenario B is a non-photovoltaic scenario, and the photovoltaic exclusive modules are turned off to simulate the underlying surface of the natural desert before construction; fix the driving data, initial conditions, and time steps of the two scenarios, compare the output results, quantify the impact of the photovoltaic power station on the local microclimate, water and heat flux, and carbon flux, and analyze the radiation-temperature-moisture-carbon coupling contribution chain; (3) Carry out virtual simulation of carbon sink enhancement technology, simulate carbon flux response under different plate spacing, vegetation configuration, and soil microbial regulation measures, and output desert photovoltaic carbon sink enhancement optimization scheme.
6. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 2, characterized in that, The sensor on the main observation tower is installed at a height at least 5m above the highest point of the photovoltaic panel; at least 3 sets of repeated observations are set at the auxiliary observation points under the panel and between the panels, and the average value is used to represent the microhabitat observation results.
7. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 3, characterized in that, The four photovoltaic-specific parameterized modules are written as independent subroutines, and data transfer between modules is completed through the built-in clm_varpar and clm_varcon data structures of CLM.
8. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 4, characterized in that, Bayesian parameter inversion uses the DREAM or MCMC toolbox, and particle filtering uses the PF-Lab open-source library; single-site parameter calibration uses high-performance computing nodes with 32 cores or more, running continuously for 7-14 days, and is accelerated by parallel computing.
9. The method for refined construction and parameter optimization of photovoltaic land surface process models adapted to desert areas according to claim 5, characterized in that, 20%-30% of the time period is reserved from the original observation data as independent verification data, covering multiple seasons and multiple weather types.
