High space-time land net ecosystem productivity estimation method
By fusing multi-source remote sensing data and optimizing models, and utilizing STNFM, STARFM, CASA, and random forest models, the spatiotemporal resolution and accuracy issues of NEP estimation were resolved, and high spatiotemporal resolution NEP data generation was achieved.
Patent Information
- Application Number
- CN202511319678.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-16
- Publication Date
- 2026-01-09
AI Technical Summary
Existing methods struggle to achieve high spatiotemporal resolution and high accuracy in estimating terrestrial net ecosystem productivity (NEP) at small spatial scales. In-situ observation data is costly, and remote sensing satellite data cannot provide high spatial resolution and long-term continuous observation.
By fusing multi-source remote sensing data, and through preprocessing and data fusion techniques, the STNFM and STARFM algorithms were used to optimize NDVI data. Combined with the CASA model and soil heteroaerobic respiration model, a random forest model was trained to generate high spatiotemporal resolution NEP data.
It improves the spatiotemporal resolution and accuracy of NEP estimation, enables the generation of high spatial resolution and long-term series data, and reduces data acquisition costs.
Smart Images
Figure CN121302230A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of remote sensing technology, specifically to a method for estimating the net ecosystem productivity of terrestrial ecosystems at a high spatiotemporal level. Background Technology
[0002] NEP is an indicator that measures the net carbon accumulation of an ecosystem over a certain period of time. It is the remaining part of NPP after deducting the organic matter consumed by soil microbial respiration. It can be used to assess the carbon sink capacity of an ecosystem and indicate the health status of the ecosystem. It has been widely used in carbon cycle research, land use planning, ecosystem management and protection, ecosystem restoration and repair and other fields.
[0003] Currently, methods for directly or indirectly estimating terrestrial non-energy production (NEP) at different spatiotemporal scales can be broadly categorized into two types. The first type estimates NEP based on in-situ observation data, including methods such as the quiet chamber method, direct harvesting method, biomass survey method, chlorophyll estimation method, and eddy covariance method. These methods rely on long-term, continuous in-situ measurement data and can provide high-precision in-situ-scale estimates; however, data acquisition and updating costs are high, and they cannot fully consider the heterogeneity of ecosystems. The second type estimates NEP by integrating meteorological and remote sensing data, mainly including geostatistical models, ecosystem process-based models, and solar energy utilization models. These methods integrate remote sensing data and meteorological observation data for mathematical modeling, offering advantages such as low data acquisition costs, wide coverage, high spatiotemporal accuracy, and well-developed parameterization methods.
[0004] Satellite remote sensing data provides a valuable opportunity to study the spatial patterns and dynamic changes of NEP. However, to achieve accurate NEP estimation and monitor its spatiotemporal dynamics over a long period at a small spatial scale, remote sensing data with high spatial resolution and long-term continuous observation is required. Currently, no remote sensing satellite data meets this requirement. Specifically, existing methods for estimating terrestrial net ecosystem productivity (NEP) have the following problems:
[0005] 1. Methods based on field observation data (such as the quiet room method and the direct harvesting method) are highly accurate, but the cost of data acquisition and updating is high, and they cannot fully take into account the heterogeneity of the ecosystem.
[0006] 2. Although remote sensing data-based methods have a wide coverage and low cost, no remote sensing satellite can currently provide data with both high spatial resolution and long-term continuous observation. Therefore, it is impossible to accurately estimate NEP and monitor its spatiotemporal dynamics over a long period of time at a small spatial scale.
[0007] This makes it difficult for existing methods to simultaneously achieve high spatiotemporal resolution and high accuracy in NEP estimation.
[0008] Therefore, a new solution is needed to address the above problems. Summary of the Invention
[0009] The purpose of this invention is to provide a high-spatial-temporal method for estimating net ecosystem productivity (NEP) on land, which integrates multi-source remote sensing data to estimate NEP, thereby improving the spatiotemporal resolution and accuracy of the estimation results and solving the technical problems mentioned in the background art.
[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for estimating net terrestrial ecosystem productivity at a high spatiotemporal level, comprising at least the following steps:
[0011] S101: Acquire boundary vector data, multi-source remote sensing NDVI data, digital elevation model (DEM), vegetation type and meteorological data within the area to be evaluated, and output high spatial resolution vegetation type data, NDVI, solar radiation, temperature and precipitation long- and short-term time series data within the area to be evaluated after preprocessing the acquired data.
[0012] Specifically, the data will first be preprocessed to obtain multi-source NDVI time series data for the area to be evaluated. The preprocessing will also yield long- and short-term time series data for vegetation type, solar radiation, temperature, and precipitation, providing reference data for subsequent model training and prediction. At the same time, the data preprocessing steps will also filter out erroneous data in the remote sensing and meteorological data, improving the accuracy and practicality of the various types of data output in the future.
[0013] S102: Based on the preprocessed NDVI data, the STNFM spatiotemporal interpolation nearest neighbor algorithm and the STARFM spatiotemporal adaptive fusion algorithm are used to output high spatial resolution NDVI long-short time series data, thereby improving the NEP estimation accuracy of the region to be evaluated.
[0014] Specifically, after acquiring the basic data, the STNFM spatiotemporal nearest neighbor interpolation algorithm is first used to fill missing values and clean up outliers in the NDVI data. Then, the STARFM spatiotemporal adaptive fusion algorithm is used to fuse the NDVI data from different spatial scales. This solves the problem of "gap" in the basic NDVI observation data caused by the "missing spatiotemporal data" of a single sensor, improves the quality of NDVI data, and further optimizes parameters such as the maximum and minimum values of NDVI.
[0015] S103: Based on long-term series data of vegetation type with high spatial resolution, NDVI, solar radiation, temperature, and precipitation, optimize the CASA model and the soil heteroaerobic respiration model Rs-Rh, and calculate the long-term series data of NEP force. Combine the long-term series data to train the random forest model, and input the short-term series data to calculate the NEP data at high time scales in the area to be evaluated.
[0016] Specifically, after optimizing parameters such as NDVI, the parameters of the CASA model and the soil heterogeneous respiration model Rs-Rh were updated. Based on the obtained long-term NPP and heterogeneous respiration data, the long-term NEP data was calculated. At the same time, the random forest model was trained using long-term NDVI, solar radiation, temperature, and precipitation data. The trained model was improved in terms of data sensitivity and estimation accuracy. Then, the short-term NEP data was estimated using the optimized model and short-term NDVI, solar radiation, temperature, and precipitation data.
[0017] Furthermore, step S101 includes at least the following steps:
[0018] Based on the vector boundary data of the assessment area;
[0019] The multi-source remote sensing NDVI data were declouded, mosaicked, cropped, and projected to obtain NDVI data at coarse and fine spatial scales, which prepared for the next step of missing value filling and data fusion, thereby improving the quality of NDVI data.
[0020] Vegetation types were trimmed, reclassified, and projected to refine them into 17 types, which were then used to optimize the CASA model and the soil heterogeneous respiration model Rs-Rh.
[0021] Interpolation calculations are performed based on elevation DEM and meteorological data to divide the processed solar radiation, temperature, and precipitation data into high spatial resolution long-short time series data according to time scale.
[0022] To improve the reliability and usability of the basic data, the estimation method provided in this invention will perform missing value filling and data fusion on the preprocessed multi-source NDVI data before model optimization, training and prediction, thereby optimizing the relevant parameters of the CASA model.
[0023] Furthermore, step S102 includes at least the following steps:
[0024] The Spatiotemporal Nearest Neighbor (STNFM) interpolation algorithm is designed to filter Normalized Difference Vegetation Index (NDVI) data from both temporal and spatial perspectives, thereby filling missing values and filtering outliers in the NDVI data and outputting NDVI data that better reflects the actual situation.
[0025] The STARFM algorithm is used to fuse high spatial resolution and coarse spatial resolution normalized vegetation indices to output more accurate NDVI long-short time series data. The maximum and minimum values of NDVI are reasonably optimized, and spatial nearest neighbor interpolation and spatial fusion are performed on the preprocessed NDVI to generate high-quality NDVI long-term and short-term time series data. Then, the maximum and minimum values of vegetation indices, the maximum and minimum values of the ratio vegetation index, and the minimum and maximum values of the ratio vegetation index are calculated. The optimized parameters are obtained by integrating the calculated data.
[0026] Furthermore, the optimization of the CASA model and the soil heteroaerobic respiration model Rs-Rh, and the calculation of long-term NEP force data, includes at least the following steps:
[0027] The photosynthetically active radiation ratio is optimized by optimizing the maximum and minimum values of NDVI and the maximum and minimum values of the ratio vegetation index, thereby optimizing the light energy conversion rate.
[0028] The parameters of the initial CASA model were optimized using the optimized photosynthetically active radiation ratio and the optimized light energy conversion rate. The long-term series data of solar radiation, temperature, and precipitation were then input into the optimized CASA model to obtain the long-term series data of NPP.
[0029] Furthermore, the soil heterogeneous respiration model Rs-Rh calculation to obtain long-term heterogeneous respiration data includes at least the following steps:
[0030] Based on the subdivision of 17 vegetation types, soil organic carbon density data and soil heterogeneous respiration model Rs-Rh model were optimized to calculate instantaneous soil respiration rate data, thereby improving the output of high-quality long-term heterogeneous respiration data from the model.
[0031] The instantaneous soil respiration rate data is input into the soil heterogeneous respiration Rs-Rh empirical model for calculation, and the long-term heterogeneous respiration data is output.
[0032] Furthermore, training a random forest model using long-term time-series data and inputting the short-term time-series data to calculate NEP data at high time scales within the region to be evaluated includes at least the following steps:
[0033] Based on long-term NPP and heterotrophic respiration data, long-term NEP data was calculated, and a random forest model was trained by combining high-resolution NDVI, solar radiation, temperature and precipitation long-term data.
[0034] High spatial resolution NDVI, solar radiation, temperature, and precipitation short-time series data are input into the random forest model, which outputs high spatial resolution NEP short-time series data, i.e., high spatiotemporal NEP data.
[0035] As a preferred example, a random forest model is built using a machine learning framework, initializing the number of base evaluators and basic parameter data for the random forest model.
[0036] Input vegetation type, NDVI, solar radiation, temperature, precipitation, and NEP long-term series data to train a random forest model and optimize the model parameters.
[0037] Input short-time series data on vegetation type, NDVI, solar radiation, temperature, and precipitation, and generate NEP short-time series data using a random forest model.
[0038] This invention obtains high spatiotemporal NEP data by training a random forest model using high spatial and long temporal sequence data and optimizing various parameters of the model.
[0039] Compared with the prior art, the beneficial effects of the present invention are:
[0040] 1. This invention proposes the fusion of multi-source remote sensing data. By preprocessing and fusing multi-source NDVI data, it solves the problem that a single remote sensing data source cannot simultaneously provide high spatial resolution and long-term series data.
[0041] 2. This invention proposes an optimized model, which improves the accuracy of NPP and heterogeneous respiration data by optimizing the CASA model and the soil heterogeneous respiration model;
[0042] 3. This invention utilizes a random forest model to apply feature learning from long-term NEP data to the estimation of short-term time series data, thereby achieving the generation of high spatiotemporal resolution NEP data. Attached Figure Description
[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0044] Figure 1 This is a schematic diagram of the implementation process of the present invention;
[0045] Figure 2 This is a structural diagram of the technical framework of the present invention. Detailed Implementation
[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0047] See Figure 1 and Figure 2 A method for estimating high-spatial-temporal terrestrial net ecosystem productivity includes steps 101 to 103, each step of which is detailed below:
[0048] Step 101: Obtain vector boundary data, multi-source NDVI remote sensing data, digital elevation model (DEM), vegetation type, and meteorological data of the area to be evaluated, and preprocess the data to output vegetation type data, high spatiotemporal resolution NDVI data, and high spatiotemporal resolution meteorological data of the area to be evaluated.
[0049] In this example, we first acquire various types of data. Table 1 shows all dataset information, mainly including vector boundary data, NDVI data, meteorological data, land cover type, and digital elevation model (DEM) data.
[0050] The vector boundary creation process begins by downloading JSON data from Tianditu (a Chinese mapping website), then converting it to SHP format using the MapShader editor, and finally performing projection transformation using ArcMap software to obtain an SHP file in the target coordinate system, which is used for subsequent data target area clipping operations.
[0051] The NDVI dataset MCD19A3CMG, provided by NASA, is a global daily resolution product from 2000 to the present. It is then processed by cloud removal, mosaicking, cropping, and projection to obtain the MODIS NDVI daily dataset with a spatial resolution of 0.05°.
[0052] The obtained Landsat NDVI datasets, including Landsat 5 TM, Landsat 7 ETM, and Landsat 8 OLI datasets, have a spatial resolution of 30 meters. After a series of processing steps, including mosaicking, cropping, and projection, a monthly Landsat NDVI dataset with a spatial resolution of 30 meters is obtained.
[0053] The vegetation type data comes from the GLC_FCS30D dataset, a global 30-meter land cover time-series dynamic remote sensing product released by the Aerospace Information Research Institute of the Chinese Academy of Sciences. After downloading, cropping, reclassifying, and projecting, 17 vegetation types were obtained.
[0054] The DEM data is an SRTM DEM product obtained from the U.S. Geological Survey, with a spatial resolution of 30 meters. The dataset has undergone mosaicking, cropping, and projection transformation processes to obtain DEM data suitable for the study area.
[0055] Meteorological data were primarily obtained from the National Meteorological Science Data Center, including daily and monthly observational data such as daily average temperature, total daily precipitation, and total daily solar radiation from 2000 to 2020. Then, using a DEM dataset with a spatial resolution of 30 m as covariates, the Anusplin meteorological interpolation software was used to spatially interpolate the meteorological station data within the study area, generating 30 m resolution gridded meteorological data.
[0056] Through the above preprocessing, daily Landsat NDVI data and monthly MODIS NDVI data of the target area at different spatial resolutions were obtained, as well as daily temperature, precipitation and solar radiation data and monthly temperature, precipitation and solar radiation data at high spatial resolution, providing data support for subsequent processing.
[0057] Step 102: Based on the preprocessed MODIS NDVI and Landsat NDVI data, use the spatiotemporal nearest neighbor interpolation algorithm STNFM and the spatiotemporal adaptive fusion algorithm STARFM to output high spatial resolution normalized vegetation index long-short time series data, thereby improving the NEP estimation accuracy of the area to be evaluated.
[0058] Considering the variation pattern of NDVI, the Temporal Nearest Neighbor (STNFM) algorithm is used in this example to fill in missing values. Even when the proportion of missing values is high, especially when consecutive adjacent values are missing, it still maintains good interpolation performance. The STNFM algorithm formula is shown below:
[0059]
[0060] Where (xw / 2, yw / 2) represents the center pixel of the moving window, and wijt is the weight of the predicted value of the center pixel by other pixels in the moving window. This weight is determined by the temporal and spatial distance between pixels.
[0061] The obtained multi-source remote sensing NDVI datasets were fused using the STARFM fusion algorithm. STARFM fusion simulates the high spatial resolution image at time t0 by utilizing one or two pairs of high spatial resolution and coarse spatial resolution images at time t1, and the coarse spatial resolution image at time t2. This method considers information such as pixel weights, spectral distances, and spatial distances between the input data. The STARFM algorithm formula is as follows:
[0062]
[0063] Where L(x,y,t) and M(x,y,t) represent the pixels of the Landsat and MODIS images at time t1, respectively; (x,y,t) is the pixel position of the given Landsat or MODIS image at time t; w represents the size of the moving window; n represents the logarithm of the input Landsat and MODIS images; (xw / 2,yw / 2) represents the center pixel of the moving window; and wijk is the weight of the predicted value of the center pixel by other pixels within the moving window, which is determined by the spectral, temporal, and spatial distances between pixels.
[0064] Through the above preprocessing, high spatial resolution NDVI long-short time series data of the target area are obtained, which solves the problem of missing data and noise interference that cause gaps and frequent fluctuations in the NDVI time series curve. This allows for reasonable optimization of the maximum and minimum values of NDVI, thereby improving the estimation accuracy of subsequent NPP.
[0065] Step 103: Based on the high spatial resolution vegetation type, NDVI, solar radiation, temperature, and precipitation monthly data, optimize the CASA model and the soil heteroaerobic respiration model Rs-Rh to calculate and obtain the NEP monthly data at high spatial resolution. Combine the NEP monthly data to train the random forest model, and input the NDVI, solar radiation, temperature, and precipitation daily data to calculate the NEP data at high spatiotemporal scales in the area to be evaluated.
[0066] Based on the CASA model proposed by Potter et al. in 1993, the aforementioned remote sensing data NDVI was used as the model input data, and combined with remote sensing data such as temperature, rainfall, soil environmental variables, vegetation physiological parameters, APAR data products, and light energy utilization rate to calculate NPP. The calculation formula is shown below:
[0067]
[0068] Where APAR(x,t) is the photosynthetically active radiation absorptivity of pixel x in month t. The value is calculated as pixel x actual light energy utilization rate. It is mainly based on the CASA model, and is obtained by inputting parameters such as NDVI, temperature, rainfall, and vegetation type, and then calculating the NPP monthly data at high spatial resolution.
[0069] In this example, combining meteorological data and soil organic carbon density (SOCD) data, the instantaneous soil respiration rate (Rs) is estimated using the geostatistical model of soil respiration (GSMSR). The estimation formula is shown below:
[0070]
[0071] In the formula, the constants are RDs=0=0.588, M=0.118, a=1.830, b=0.006, P0=2.972, and K=5.657, respectively.
[0072] Then, the Rh of the target region is calculated based on the Rs-Rh relationship model. Input data includes temperature T(x,t), precipitation P(x,t), and soil organic carbon density (SOCD), with SOCD corresponding to vegetation type. The data and organizational structure used in this example are shown below:
[0073] LAND_COVER_TYPES = {
[0074] 1: {'type': 'ENF', 'socd': 3.77, 'info': 'Evergreen needleleafforest'},
[0075] 2: {'type': 'EBF', 'socd': 4.7, 'info': 'Evergreen broadleafforest'},
[0076] 3: {'type': 'DNF', 'socd': 3.77, 'info': 'Deciduous needleleafforest'},
[0077] 4: {'type': 'DBF', 'socd': 4.7, 'info': 'Deciduous broadleafforest'},
[0078] 5: {'type': 'MXF', 'socd': 4.235, info': 'Mixed forest'},
[0079] 6: {'type': 'DS', 'socd': 2.56, 'info': 'Dense shurb'},
[0080] 7: {'type': 'OS', 'socd': 2.56, 'info': 'Open shurb'},
[0081] 8: {'value': 'ENF', 'socd': 3.77, 'info': 'Evergreen needleleafforest'},
[0082] 9: {'value': 'EBF', 'socd': 4.7, 'info': 'Evergreen broadleafforest'},
[0083] 10: {'type': 'Grassland', 'socd': 1.82, 'info': 'Grassland'},
[0084] 11: {'type': 'Wetland', 'socd': 0, 'info': 'Wetland'},
[0085] 12: {'type': 'Cropland', 'socd': 2.56, 'info': 'Cropland'},
[0086] 13: {'value': 'Grassland', 'socd': 0, 'info': 'Grassland'},
[0087] 14: {'value': 'Cropland', 'socd': 2.56, 'info': 'Cropland'},
[0088] 15: {'value': 'Other', 'socd': 0, 'info': 'nonvegetation'},
[0089] 16: {'value': 'Other', 'socd': 0, 'info': 'nonvegetation'},
[0090] 17: {'value': 'Other', 'socd': 0, 'info': 'nonvegetation'},
[0091] }
[0092] According to the GSMSR model described above, Rh is calculated based on the empirical relationship between Rs and Rh. In this example, the Rs-Rh relationship model shown in the following formula is used to calculate the Rh of the target region:
[0093]
[0094] The NPP data obtained from the CASA model and Rs-Rh model are used to calculate monthly net ecosystem productivity (NEP) data at high spatial resolution. The calculation formula is as follows:
[0095]
[0096] NEP represents the net carbon storage on a larger scale, and its value can be positive or negative. When NEP is greater than 0, it indicates that the ecosystem is a CO2 sink, and vice versa.
[0097] Random forest regression is a combined algorithm of decision tree regression. In this example, the scikit-learn machine learning framework is used to build a random forest model and the number of base evaluators for the random forest model is initialized.
[0098] The high spatial resolution data on vegetation type, monthly NDVI, monthly solar radiation, monthly average temperature, monthly precipitation, and monthly net ecosystem productivity (NEP) were organized into a training dataset to train the random forest model.
[0099] The vegetation type, daily NDVI, daily solar radiation, daily average temperature, and daily precipitation P data are organized into a prediction dataset. The random forest model is used to estimate the daily NEP data, and finally the prediction data are combined into raster image data.
[0100] Using the CASA model, the soil heterogeneous respiration empirical model Rs-Rh, and the random forest model, and based on monthly and daily data of NDVI, vegetation type, solar radiation, mean temperature, and precipitation at high spatial resolution, net ecosystem productivity (NEP) data at high spatial resolution was finally generated, with a spatial resolution of 30m and a temporal resolution of daily data.
[0101] In summary, the net ecosystem productivity (NEP) data obtained by estimating using the above-described method in this embodiment improves the spatial resolution to 30m and the temporal resolution to daily. This means that users can use the method provided in this embodiment to estimate the net ecosystem productivity data of the area to be evaluated one day later.
[0102] In summary:
[0103] This invention utilizes multi-source remote sensing product data from MODIS and Landsat, employing the Spatiotemporal Nearest Neighbors Filling Method (STNFM) and the Spatiotemporal Adaptive Reflectance Fusion Model (STARFM) to obtain high spatial resolution vegetation types, as well as long-term and short-term time series data of normalized vegetation index (NDVI), solar radiation, temperature, and precipitation within the area to be evaluated. Based on the vegetation types, the CASA (Carnegie-Ames-Stanford Approach) model and the soil heteroaerobic respiration model Rs-Rh are optimized. The invention outputs and calculates long-term time series data of net primary productivity (NPP) and heteroaerobic respiration within the area to be evaluated. Finally, the invention uses the vegetation index NDVI (Normalized Difference) to further refine the data. Using Vegetation Index (Vegetation Index), surface solar radiation, temperature, precipitation, and long-term NEP data, a random forest model was trained. Short-term data of vegetation index, solar radiation, temperature, and precipitation were then input into the training model to estimate high-temporal-scale NEP data. Based on high-temporal-scale multi-source remote sensing data, the CASA model, GSMR model, and random forest model were integrated to estimate high-temporal-scale land NEP, improving the accuracy and practicality of the data.
[0104] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. A method for estimating net terrestrial ecosystem productivity at high spatiotemporal levels, characterized in that: At least the following steps are included: S101: Acquire boundary vector data, multi-source remote sensing NDVI data, digital elevation model (DEM), vegetation type and meteorological data within the area to be evaluated, and output high spatial resolution vegetation type data, NDVI, solar radiation, temperature and precipitation long- and short-term time series data within the area to be evaluated after preprocessing the acquired data. S102: Based on the preprocessed NDVI data, the STNFM spatiotemporal interpolation nearest neighbor algorithm and the STARFM spatiotemporal adaptive fusion algorithm are used to output high spatial resolution NDVI long-short time series data, thereby improving the NEP estimation accuracy of the region to be evaluated. S103: Based on long-term series data of vegetation type with high spatial resolution, NDVI, solar radiation, temperature, and precipitation, optimize the CASA model and the soil heteroaerobic respiration model Rs-Rh, and calculate the long-term series data of NEP force. Combine the long-term series data to train the random forest model, and input the short-term series data to calculate the NEP data at high time scales in the area to be evaluated.
2. The method for estimating net terrestrial ecosystem productivity at high temporal and spatial levels according to claim 1, characterized in that: S101 includes at least the following steps: Based on the vector boundary data of the assessment area; The multi-source remote sensing NDVI data were declouded, mosaicked, cropped, and projected to obtain NDVI data at coarse and fine spatial scales, which prepared for the next step of missing value filling and data fusion. Vegetation types were trimmed, reclassified, and projected to refine them into 17 types, which were then used to optimize the CASA model and the soil heterogeneous respiration model Rs-Rh. Interpolation calculations are performed based on elevation DEM and meteorological data to divide the processed solar radiation, temperature, and precipitation data into high spatial resolution long-short time series data according to time scale.
3. The method for estimating net terrestrial ecosystem productivity at high temporal and spatial levels according to claim 2, characterized in that: S102 includes at least the following steps: The Spatiotemporal Nearest Neighbor (STNFM) interpolation algorithm is designed to filter Normalized Difference Vegetation Index (NDVI) data from both temporal and spatial perspectives, thereby filling missing values and filtering outliers in the NDVI data and outputting NDVI data that better reflects the actual situation. The STARFM algorithm is used to fuse high spatial resolution and coarse spatial resolution normalized vegetation indices to output more accurate NDVI long-short time series data. The maximum and minimum values of NDVI are reasonably optimized, and spatial nearest neighbor interpolation and spatial fusion are performed on the preprocessed NDVI to generate high-quality NDVI long-term and short-term time series data. Then, the maximum and minimum values of vegetation indices, the maximum and minimum values of the ratio vegetation index, and the minimum and maximum values of the ratio vegetation index are calculated. The optimized parameters are obtained by integrating the calculated data.
4. The method for estimating net terrestrial ecosystem productivity at high temporal and spatial levels according to claim 3, characterized in that: The optimization of the CASA model and the soil heteroaerobic respiration model Rs-Rh, and the calculation of long-term NEP force data, includes at least the following steps: The photosynthetically active radiation ratio is optimized by optimizing the maximum and minimum values of NDVI and the maximum and minimum values of the ratio vegetation index, thereby optimizing the light energy conversion rate. The parameters of the initial CASA model were optimized using the optimized photosynthetically active radiation ratio and the optimized light energy conversion rate. The long-term series data of solar radiation, temperature, and precipitation were then input into the optimized CASA model to obtain the long-term series data of NPP.
5. The method for estimating net terrestrial ecosystem productivity at high temporal and spatial levels according to claim 4, characterized in that: The soil heterogeneous respiration model Rs-Rh calculation to obtain long-term heterogeneous respiration data includes at least the following steps: Based on the subdivision of 17 vegetation types, soil organic carbon density data and soil heteroaerobic respiration model Rs-Rh model were optimized to calculate instantaneous soil respiration rate data. The instantaneous soil respiration rate data is input into the soil heterogeneous respiration Rs-Rh empirical model for calculation, and the long-term heterogeneous respiration data is output.
6. The method for estimating net terrestrial ecosystem productivity at high temporal and spatial levels according to claim 5, characterized in that: Training a random forest model using long-term time-series data and inputting the short-term time-series data to calculate NEP data at high time scales within the region to be evaluated includes at least the following steps: Based on long-term NPP and heterotrophic respiration data, long-term NEP data was calculated, and a random forest model was trained by combining high-resolution NDVI, solar radiation, temperature and precipitation long-term data. High spatial resolution NDVI, solar radiation, temperature, and precipitation short-time series data are input into the random forest model, which outputs high spatial resolution NEP short-time series data, i.e., high spatiotemporal NEP data.