Polar region winter phytoplankton net primary productivity monitoring method and system and computer program

Abstract

The invention belongs to the technical field of ocean remote sensing monitoring, particularly relates to a polar region winter phytoplankton net primary productivity dynamic monitoring method and system and a computer program, and is suitable for biogeochemical process monitoring under the weak light condition of polar region sea areas such as the south pole and the north pole. The method comprises the following steps: multi-source data acquisition and space-time matching; satellite-borne laser radar data signal correction, including transient response elimination and polarization crosstalk correction; satellite-borne laser radar data quality control; constructing a physical bio-optical parameter database; training a double-branch deep learning model; step-by-step parameter reconstruction is carried out; carrying out light adaptation modeling on a vertical section; calculating carbon-based productivity; and multi-source data verification is carried out, so that the accurate net primary productivity of the polar region winter sea area phytoplankton is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine remote sensing monitoring, and particularly relates to a dynamic monitoring method, system and computer program for the net primary productivity of polar winter phytoplankton based on spaceborne lidar, which is applicable to the monitoring of biogeochemical processes under low light conditions in polar seas such as the Antarctic and Arctic. Background Art

[0002] As an important part of the global carbon cycle, the net primary productivity (NPP) of marine phytoplankton plays a crucial role in global climate change and carbon sink assessment. Especially in polar seas, due to special climate conditions and unique ecosystems, they have an undeniable impact on global climate regulation and marine biogeochemical processes. Polar seas not only show relatively high phytoplankton productivity in summer, but may still maintain certain ecological functions in winter. Although the light is extremely weak, organisms in extreme environments may adopt special survival strategies, enabling polar phytoplankton to still play an ecological role under polar night conditions. However, due to the harsh polar winter environment and continuous extremely low light conditions day and night, it brings great difficulties to the monitoring of phytoplankton NPP.

[0003] Currently, the global monitoring of marine phytoplankton NPP mainly relies on two major methods: passive remote sensing methods and in-situ observations by ships or buoys. Passive remote sensing methods mainly obtain the reflection signals of ocean water color, such as parameters like chlorophyll concentration, particulate backscattering coefficient, and water body diffuse attenuation coefficient, to indirectly invert the productivity of phytoplankton. The Chinese patent application for invention (Publication No.: CN109490270A, Publication Date: March 19, 2019) discloses a measuring device and method for the primary productivity of phytoplankton based on chlorophyll fluorescence. Under light-shielded conditions, the algal sample is irradiated by a simulated light source for more than 60 s, and the single-cycle and relaxation fluorescence kinetic curves under light conditions are measured during the 50 ms short-term shutdown interval of the simulated light, taking advantage of the light history dependence of algae, to obtain the fluorescence kinetic parameters under light-adapted conditions. A laser diode array is used as the induced excitation light source, and a photomultiplier tube is used as the fluorescence detector to measure the chlorophyll fluorescence kinetic curve of algae; a crown-shaped optical collector and a multi-band light detector array are used to design an underwater photosynthetically active radiation measurement unit to measure the natural environment spectrum corresponding to the absorption characteristics of algae. This method is relatively convenient for data acquisition and has a high spatial coverage rate under daylight conditions with sufficient light. However, in polar winter, due to the extremely low solar altitude angle and weak sea surface reflection signals, the effective data coverage rate of passive remote sensing products is usually less than 20%, and it is even insufficient to provide continuous spatio-temporal data sequences, thus unable to truly reflect the ecological dynamic changes in polar seas.

[0004] On the other hand, although shipborne observations and buoy in-situ observations can provide detailed local data in polar waters, they are limited by the extremely harsh climate and ice distribution. The spatial coverage of these methods in the vast polar region is very limited, and data collection is restricted by factors such as weather, and the temporal resolution is low. In addition, the optical and biochemical parameters collected by in-situ equipment can often only provide local representative information and are difficult to reflect the true situation of the entire polar waters. Therefore, how to achieve large-scale, continuous, and high-temporal-resolution phytoplankton NPP monitoring in the polar winter has become a technical problem that needs to be solved urgently in the fields of marine science and atmospheric environment research.

[0005] Based on the above background, in recent years, the scientific community has begun to explore the feasibility of using spaceborne lidar technology to conduct ocean observations. Spaceborne lidar has the characteristics of active light detection and does not rely on solar radiation, so it can still effectively obtain target information in polar night or low light conditions. Spaceborne lidar systems represented by CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) can work continuously during the day and night, detecting scattered signals from the sea surface and below through laser emission and echo reception technology. The introduction of this technology provides new possibilities for filling the data gap in polar winter and has attracted widespread attention from researchers at home and abroad. Summary of the invention

[0006] In view of the above problems, the present invention provides a novel method for dynamic monitoring of polar winter phytoplankton net primary productivity based on space-borne lidar. The method includes a two-branch two-step deep learning model, which can accurately invert multiple ocean water optical and chlorophyll parameters from the space-borne lidar signal, and then reconstruct the net primary productivity of phytoplankton in high-latitude waters, thus realizing the dynamic monitoring of polar winter phytoplankton net primary productivity. This provides more reliable data support for marine scientific research and polar carbon sink assessment.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions: A method for monitoring the net primary productivity of polar winter phytoplankton based on space-borne laser radar comprises the following steps: S1: Multi-source remote sensing data acquisition and time-space matching, registering satellite-borne lidar data with ocean color remote sensing data through set time windows and space windows to ensure spatial coverage and time consistency; S2: Perform signal correction on the raw echo data of the space-borne lidar, including transient response correction and polarization crosstalk correction; S3: LiDAR data quality control is performed based on water body depolarization ratio, sea ice concentration and wind speed data to eliminate sea ice interference and mirror reflection anomalies; S4: Construct a parameter database containing various physical-bio-optical characteristics, and normalize the characteristic variables in the database. The various physical-bio-optical characteristics include the echo signal after spaceborne lidar calibration, the water body depolarization ratio, the water body backscattering coefficient, and the ocean color remote sensing products; S5: Based on the database, train a dual-branch deep neural network model. The first branch extracts the echo waveform characteristics based on a convolutional neural network, and the second branch fuses the optical parameters for physical feature learning, and then outputs the optical chlorophyll concentration after splicing; S6: Implement parameter reconstruction based on a two-step modeling strategy. First, output the monthly average climatology, and then reconstruct the anomaly values through differences to improve the polar inversion accuracy; S7: Based on the reconstructed parameters, perform vertical profile light adaptation modeling to obtain the photosynthetically active radiation distribution in the depth direction; S8: Using the reconstructed optical parameters and light distribution, calculate the net primary productivity of phytoplankton based on the carbon-based model; S9: Use buoy data and passive remote sensing data for multi-source cross-validation to evaluate and optimize the productivity inversion accuracy.

[0008] Preferably, the transient response correction in step S2 includes deconvolving the original echo signal β(z) using the detector response function F to obtain the corrected signal: β′(z)=F - ¹β(z) where β′(z) is the signal after detector transient response correction, β(z) is the lidar measurement signal, and F is the detector's transient response function; The polarization crosstalk correction is achieved through the following formula: , , where CT is the crosstalk coefficient, and are the signals of the parallel channel and the vertical channel after polarization crosstalk correction respectively, and are the signals of the parallel channel and the vertical channel after transient response correction respectively.

[0009] Preferably, the total water body depolarization ratio δ T is calculated as follows: , where δ T is the total water body depolarization ratio, and p is the position of the peak.

[0010] Preferably, the water body backscattering coefficient γ in step S4 is calculated according to the following formula: , where γ is the backscattering coefficient of the water body, β s is the backscattering coefficient of the sea surface: , where θ is the nadir angle of the laser beam, and σ 2 is the mean square deviation of the sea wave slope: , where W is the sea surface wind speed; The normalization process is as follows: , where X norm is the normalized parameter variable, X is the original variable, is the variable mean, and SD(X) is the variable standard deviation.

[0011] Preferably, the photosynthetically active radiation PAR(z) in the vertical direction in step S7 is estimated according to the following model: PAR(z) = PAR(0) × exp(-2K d z), where PAR(z) is the photosynthetically active radiation at depth z, PAR(0) is the photosynthetically active radiation at the sea surface, and K d is the diffuse attenuation coefficient obtained by inversion.

[0012] Preferably, the net primary productivity NPP of phytoplankton in step S8 is calculated according to the following formula: NPP = C × μ × Z eu , where NPP is the net primary productivity of phytoplankton, C is the phytoplankton carbon biomass, μ is the growth rate, and Z eu is the euphotic layer depth; The carbon biomass C is calculated according to the following formula: C = 13000 × (b bp - 0.00035), where b bp is the particulate backscattering coefficient; The growth rate μ is calculated according to the following formula: , where Chl is the chlorophyll concentration and PAR(z) is the photosynthetically active radiation.

[0013] Furthermore, the present invention also provides a monitoring system for the net primary productivity of polar winter phytoplankton, which implements the above method and includes: A multi-source remote sensing data acquisition module, which is used to receive spaceborne lidar data and ocean color remote sensing data; A spatio-temporal matching module, which is used to accurately register multi-source remote sensing data according to a preset spatial distance window and time window to ensure the consistency of data in spatial coverage and time series; A signal correction module, which is used to perform the following two corrections on the spaceborne lidar data: Transient response correction, which obtains the corrected signal by deconvolving the original echo signal β(z) using the detector transient response function F; Polarization crosstalk correction, which is achieved by converting the signals of the corrected parallel channel and vertical channel; A data quality control module, which is used to detect and filter abnormal data based on the total water depolarization ratio, sea ice concentration, and wind speed information; A physical-biogeooptical parameter database construction module, which is used to integrate the corrected lidar echo signal, depolarization ratio, seawater backscattering coefficient, and remote sensing products, and perform normalization processing on the integrated parameters; A dual-branch deep learning model module, including: The first branch, which uses a convolutional neural network (CNN) to extract features from the lidar echo waveform; The second branch, which fuses physical parameters such as the water body depolarization ratio and backscattering coefficient output by the data quality control module and the parameter database construction module, The dual-branch features are spliced by a fully connected regression network and then the water body optical chlorophyll parameters are output to realize waveform-environment collaborative modeling under physical constraints; A parameter reconstruction module, which is used to further reconstruct the inversion parameters with a two-step optimization strategy: first, output the monthly average climatological parameters, and then reconstruct the outliers based on the difference between the observed values and the climatological state, so as to improve the parameter inversion accuracy; A vertical profile light adaptation modeling module, which is used to calculate the photosynthetically active radiation (PAR) in the depth direction according to the reconstructed water body optical parameters; A carbon-based productivity calculation module, which is used to calculate the net primary productivity (NPP) of phytoplankton according to the light adaptation modeling results and water body parameters, and a multi-source data verification module, which is used to verify the optical parameters and NPP results output by the above modules using buoy data, ocean color remote sensing products, and other measured data, so as to ensure the accuracy and stability of the operation of the entire system.

[0014] Preferably, the transient response correction and polarization crosstalk correction formulas used by the signal correction module are implemented as preset modules in the system hardware or software to automatically correct the echo data of the spaceborne lidar; and / or the data quality control module can automatically identify and filter abnormal data affected by sea ice, specular reflection, and foam interference based on the water body depolarization ratio δT and auxiliary sea ice concentration and wind speed data.

[0015] Furthermore, the present invention also provides a computer device, including: a processor; a memory; and a computer program stored in the memory, which is executed under the control of the processor to implement the described method.

[0016] Furthermore, the present invention also provides a computer-readable storage medium, on which program instructions are stored, and when the instructions are executed by a computer, they are used to implement the described method.

[0017] Due to the adoption of the above technical solutions, the present invention integrates the active detection technology of spaceborne lidar with multi-source data fusion, deep learning algorithms, and physical model constraints, realizing the full-process automation, precision, and continuity of the monitoring of the net primary productivity (NPP) of polar winter phytoplankton, and achieving the following remarkable technical effects: 1. Greatly improve data coverage and spatio-temporal continuity: Traditional passive remote sensing methods have extremely low effective data coverage in polar winter due to low light, polar day / night, etc., and can only cover a very small area or sample intermittently; while the present invention utilizes the active laser detection ability of spaceborne lidar, which is not restricted by solar radiation, and can obtain sea area echo data in real time even under polar night conditions. By setting a strict spatio-temporal matching mechanism, precise registration of multi-source data in time and space is achieved, thereby greatly improving the coverage and spatio-temporal continuity of polar data collection and effectively filling the observation blind area in polar winter.

[0018] 2. Precise correction and reliable signal recovery: Aiming at the common detector transient response effect and polarization crosstalk problems in lidar signals, the present invention proposes a correction scheme based on deconvolution and crosstalk correction formulas, effectively restoring the true physical characteristics of the original echo signal. By removing the interference of the non-ideal response of the detector and polarization signals, the accuracy and stability of the data are improved, providing a reliable basis for subsequent parameter inversion, and further improving the accuracy of the estimation of ocean optical parameters and NPP.

[0019] 3. Enhance data quality and intelligent parameter inversion: In view of various interference factors such as sea ice, specular reflection, and foam existing in the polar environment, the present invention designs a multi-dimensional data quality control module based on the total depolarization ratio of water bodies, sea ice concentration, and wind speed data, effectively identifying and eliminating abnormal data to ensure high-quality data input. At the same time, by using the physical-biological optical parameter database construction module, the multi-source data fusion is normalized, improving the distribution state of the data in the neural network model and providing solid data support for subsequent deep learning parameter inversion.

[0020] 4. Breakthrough deep learning model for multi-parameter collaborative inversion: Adopting a dual-branch deep learning model structure, the present invention makes full use of the convolutional neural network to extract the spatio-temporal features of lidar echo waveforms, and at the same time fuses physical parameters such as the water body depolarization ratio and backscattering coefficient to establish a waveform-environment collaborative inversion model under physical constraints. Through feature splicing and fully connected regression, the model effectively improves the accuracy of inverting chlorophyll concentration and other water body optical parameters, laying a foundation for accurately calculating the NPP of phytoplankton. In addition, through the step-by-step parameter reconstruction strategy, using the climatological output as the initial estimate first and then correcting the outliers, the seasonal bias problem caused by sparse polar data is further overcome.

[0021] 5. Construct a scientific, rigorous and general NPP calculation model: On the basis of inverting the water body optical parameters, the present invention constructs an NPP estimation scheme with a carbon-based model as the core, and by accurately calculating the carbon biomass of phytoplankton, the chlorophyll growth rate, and the euphotic layer depth, it ensures that the net primary productivity of phytoplankton can still be accurately evaluated under low-light conditions in the polar region. This scheme makes full use of the complementary advantages of lidar data and auxiliary water color remote sensing data, making up for the problem of underestimated productivity of traditional models in the polar environment and providing more comprehensive and accurate data support for global carbon cycle research and climate change prediction.

[0022] 6. Multi-source verification to ensure the stability and reliability of the system: The present invention introduces a multi-source data verification module, using buoy observations, passive water color remote sensing products, and other measured data to cross-compare and verify the optical parameters and NPP results inverted by lidar, so as to comprehensively evaluate and optimize the overall performance of the monitoring system. This multi-source verification method not only enhances the stability and reliability of the system operation, but also provides sufficient experimental and data basis for practical applications, making the system suitable for long-term continuous monitoring in the complex polar environment.

[0023] 7. Promote the progress of polar ocean remote sensing technology and carbon cycle science: The technical solution of the present invention not only breaks through the limitations of passive remote sensing data acquisition under low light conditions in polar winters, but also realizes the collaborative inversion of multiple parameters in the polar environment through the deep integration of lidar technology and advanced data processing and deep learning methods. This technological progress greatly improves the accuracy and temporal resolution of the monitoring of polar ocean phytoplankton NPP, which has great theoretical and practical significance for strengthening global carbon sink research and accurately assessing the impact of climate change. At the same time, it also provides new ideas and new platforms for the development and application of subsequent related technologies.

[0024] In summary, through the collaborative work of multiple modules and innovative data processing, deep learning models, and physical constraint mechanisms, the present invention forms a systematic, complete, reasonable in structure, and stable in performance monitoring system for the net primary productivity of polar winter phytoplankton. This system not only achieves a breakthrough in active sensing and multi-source data fusion technically, but also significantly improves the data quality and spatial coverage rate of polar ecological monitoring in practical applications, providing reliable and advanced technical support for global carbon cycle modeling, climate change prediction, and polar ecological environmental protection, and having broad application prospects and profound scientific significance. Brief Description of the Drawings

[0025] Figure 1 is the flow chart of the method for monitoring the net primary productivity of polar winter phytoplankton based on spaceborne lidar.

[0026] Figure 2 is the double-branch two-step neural network structure.

[0027] Figure 3 is the comparison between the spaceborne lidar inversion parameters and passive water color remote sensing products.

[0028] Figure 4 is the comparison between the spaceborne lidar inversion parameters and the measured results.

[0029] Figure 5 is the comparison between the spaceborne lidar net primary productivity and the passive water color remote sensing net primary productivity.

[0030] Figure 6 is the comparison between the spaceborne lidar net primary productivity and the buoy lidar net primary productivity.

[0031] Figure 7 is the comparison between the spaceborne lidar winter coverage and the passive water color remote sensing coverage. Detailed Description of the Invention

[0032] Combined with the embodiments of the present invention, the technical solutions in the embodiments will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the protection scope of the present invention.

[0033] As Figure 1 shown, the method for monitoring the net primary productivity of polar winter phytoplankton based on spaceborne lidar of the present invention includes the following steps: Step 1: Multi-source remote sensing data acquisition and spatio-temporal matching; Step 2: Spaceborne lidar data signal correction; Step 3: Spaceborne lidar data quality control; Step 4: Construction of a physical-biological optical parameter database; Step 5: Training of a dual-branch deep learning model; Step 6: Step-by-step parameter reconstruction; Step 7: Vertical profile light adaptation modeling; Step 8: Carbon-based productivity calculation; Step 9: Multi-source data verification. The following will elaborate on each step in detail.

[0034] Step 1: Multi-source remote sensing data acquisition and spatio-temporal matching In multi-source remote sensing data acquisition and spatio-temporal matching, spaceborne lidar data and passive ocean color remote sensing products are obtained, and data matching is achieved based on a certain spatial distance window and time window to ensure strict consistency of multi-source data in spatial coverage and time series. In this example, CALIOP spaceborne lidar data and MODIS chlorophyll Chl, particulate backscattering coefficient bbp, and diffuse attenuation coefficient remote sensing products are used. The spatial distance window is 25 km, and the time window is 12 h; that is, data within 25 km on the same day are regarded as valid matching points. This example takes the Antarctic Sea area as an example to elaborate on the method for monitoring the net primary productivity of polar winter phytoplankton based on spaceborne lidar.

[0035] Step 2: Spaceborne lidar data signal correction For the spaceborne lidar data signal correction, the influence of the non-ideal characteristics of the detector on the data is removed, including the transient response correction of the detector: β′(z)=F - ¹β(z), where β′(z) is the signal after the transient response correction of the detector, β(z) is the lidar measurement signal, and F is the transient response function of the detector.

[0036] And the polarization crosstalk correction of the polarization beam splitter: , , where and are the signals of the parallel channel and the vertical channel after the polarization crosstalk correction, respectively, and are the signals of the parallel channel and the vertical channel after transient response correction, respectively.

[0037] Step 3: The quality control of spaceborne lidar data is as follows: Remove the influence of sea ice based on the total depolarization ratio of water bodies, remove the ice-water mixed area based on sea ice concentration data, and limit the wind speed to avoid specular reflection and foam interference. Among them, the total depolarization ratio of water bodies is: , where δ T is the total depolarization ratio of water bodies, and p is the position of the peak.

[0038] Step 4: Construct a physical-bio-optical parameter database Integrate the echo signals after spaceborne lidar calibration, the total depolarization ratio of water bodies, the backscattering coefficient of water bodies, and ocean color remote sensing products, and normalize all parameters to improve the stability of model training. Among them, the backscattering coefficient of water bodies is: , where is the backscattering coefficient of water bodies, is the backscattering coefficient of the sea surface: , where is the nadir angle of the laser beam, is the mean square deviation of the sea wave slope: , where W is the sea surface wind speed.

[0039] The normalization process is as follows: , where X norm is the normalized parameter variable, X is the original variable, is the variable mean, and SD(X) is the variable standard deviation.

[0040] Step 5: Training of the dual-branch deep learning model The training of the dual-branch deep learning model is to construct a dual-branch network combining waveforms and physical parameters through deep learning (as Figure 2 shown, where CNN is the convolutional layer and FC is the fully connected layer).

[0041] 1. Branch 1 - Feature extraction of lidar echo waveforms (CNN branch): Input layer: Receive one-dimensional (or two-dimensional, depending on the data preprocessing results) lidar echo waveform data.

[0042] Convolutional Layer (CNN): Design at least 2 - 3 convolutional layers, using different filter sizes to extract local features. After each convolutional layer, batch normalization and a non - linear activation function (such as ReLU) can be followed to accelerate convergence.

[0043] Pooling Layer: Use max - pooling or average - pooling after some convolutional layers to reduce the size of the feature map, making the model somewhat robust to noise.

[0044] Global Pooling / Flattening Layer: Finally, flatten the feature map output by the convolutional layer or use global average pooling to generate a feature vector of a fixed size.

[0045] 2. Branch 2 - Physical Parameter Fusion Branch: Input Layer: Receive a set of physical parameter vectors, mainly including the total depolarization ratio δ of the water body T and the backscattering coefficient γ (other relevant parameters can be extended if necessary).

[0046] Fully - Connected Layer (FC): Design 1 - 2 fully - connected layers (Dense layers). In each layer, use non - linear activation (such as ReLU) to transform the input vector, extract physical features, and output a feature vector of a fixed dimension.

[0047] 3. Feature Concatenation and Regression Output Feature Fusion: Concatenate the waveform feature vector obtained from Branch 1 and the physical parameter feature vector obtained from Branch 2 to form a joint feature representation that combines waveform information and environmental physical parameters.

[0048] Fully - Connected Regression Layer: Perform regression on the concatenated joint features through one or more fully - connected layers until the target parameter - the optical chlorophyll parameter of the water body is output. Use a linear activation function in the output layer to directly predict the continuous value of the target.

[0049] 4. Loss Function and Training Strategy 1) Loss Function: Use the mean squared error loss function (Mean Squared Error, MSE) to measure the difference between the predicted value and the true chlorophyll parameter. You can choose the Adam or SGD optimizer and perform end - to - end training with an appropriate learning rate (such as 1e - 3 or lower, adjusted according to the data characteristics).

[0050] 2) Training Process: Divide the dataset into training / validation / test sets to ensure the generalization performance of the model. During training, implement the Early Stopping strategy or learning rate decay for the model to prevent overfitting and improve the convergence speed. Meanwhile, data augmentation techniques (such as adding slight noise to the echo data) can be adopted to enhance the robustness of the model.

[0051] 5. Model Output: After training, the model can output accurate water optical chlorophyll parameters based on the input lidar waveforms and corresponding physical parameters, providing accuracy guarantee for subsequent NPP inversion and parameter reconstruction.

[0052] Step Six: The step-by-step parameter reconstruction is divided into two-step optimized inversion: To address the problems of polar data sparsity and seasonal bias, a two-step optimized inversion strategy is adopted, dividing the inversion process into a "climatology estimation stage" and an "outlier correction stage".

[0053] 1. First Step: Climatology Parameter Estimation Objective: Use historical data and training samples to train the model to output monthly average climatology parameters as the reference state. Climatology parameters usually refer to the average state over a relatively long time scale (such as monthly), which includes the basic optical characteristics of the ocean environment, such as monthly average chlorophyll concentration, particulate backscattering coefficient (bbp), and water body diffuse attenuation coefficient (Kd).

[0054] Implementation Method: 1) Data Division: Use the average values of each parameter for each month statistically over a relatively long time period (such as multi-year data) to establish the target output dataset.

[0055] 2) Model Structure: Based on the dual-branch model trained in Step Five, expand its output layer to achieve the estimation of monthly average climatology parameters.

[0056] 3) Training Process: The input data is the preprocessed CALIOP lidar data and the corresponding physical parameters, and the target output is the historical average value (climatology) for the corresponding month. The mean squared error loss function is also used to fit the model output and the monthly average climatology parameters. After training, the model can provide a reference estimation result for a given month, which has strong stability and statistical representativeness.

[0057] 2. Second Step: Outlier Reconstruction Objective: Based on the reference climatology parameters, model the deviation between the current observations and the climatology to correct the abnormal deviations caused by data sparsity and extreme environmental changes, and achieve more refined parameter inversion.

[0058] Implementation Method: 1) Difference Calculation: For each actual observation point, calculate the difference (residual / anomaly) between the observed value and the monthly mean climatological parameter output in the first step, denoted as Δ, i.e., Δ = y obs- y climate ; where y obs is the current observed value, and y climate is the climatological parameter of the same month.

[0059] 2) Reconstruction model design: Construct an outlier reconstruction network, which can adopt relatively simple fully connected layers or convolutional layers. The input is the original input data (or the intermediate features after the first-step dual-branch model), and the calculated difference Δ, and the output is the corrected anomaly component. In the design of this network, the idea of residual learning can be utilized to make the network focus on fitting the non-linear part between the observed value and the reference climatology. The MSE loss function is also used during training to make the output outlier closest to the true difference (provided by high-quality observed data).

[0060] 3) Integrated output: The final inversion result is: y final = y climate+ Δp redy ; where Δ pred is the correction value output by the outlier reconstruction network. In this way, on the one hand, the stability of the monthly mean climatology is maintained, and on the other hand, effective compensation for seasonal and local anomalies is carried out, thus effectively solving the problems of data sparsity and seasonal bias.

[0061] 3. Joint training strategy Cascade training: To improve the overall inversion accuracy, a cascade training strategy can be adopted, that is, first train the climatology estimation model alone. After convergence, freeze some weights, and then train the outlier reconstruction model. This can not only make the reference output stable but also optimize the abnormal part in detail.

[0062] Joint optimization: Another implementation method is to design an end-to-end joint network structure, where the output of the first stage is directly connected in series with the anomaly correction part, and the parameter updates of the two stages are guided by the joint loss function at the same time. The joint loss function can be composed of the climatology loss and the outlier reconstruction loss, and the weights of each part can be adjusted according to the experimental results.

[0063] 4. Model verification and parameter monitoring Using cross - validation methods and auxiliary verification with multi - source data (such as buoy data, passive water - color remote - sensing data) to ensure the prediction accuracy of the climatology and anomaly correction models separately and as a whole. Comparing the prediction results at each stage with the actual observed data, using metrics (such as correlation coefficient R, mean absolute percentage error MAPD, root mean square error RMSD, etc.) to quantitatively evaluate the model performance, and dynamically adjusting the training strategy during model updates.

[0064] 5. Final output parameters After step six, the model can simultaneously retrieve multiple key water - body optical parameters, including: chlorophyll concentration (Chl); backscattering coefficient of water - body particulate matter (bbp); diffuse attenuation coefficient of water - body (Kd). The high - precision retrieval of these parameters will serve as the basis for subsequent NPP calculations and will be further optimized and corrected in multi - source data verification to ensure the reliability and accuracy of the overall monitoring system.

[0065] Step seven: Modeling of vertical profile light adaptation In step seven, the modeling of vertical profile light adaptation is to calculate the vertical profile of photosynthetically active radiation according to the retrieved water - body optical parameters: PAR(z) = PAR(0) × exp(-2K d z), where PAR(z) is the photosynthetically active radiation at depth z, PAR(0) is the photosynthetically active radiation at the sea surface, and K d is the retrieved diffuse attenuation coefficient.

[0066] Step eight: Calculation of carbon - based productivity Based on carbon - based productivity, calculate the net primary productivity of phytoplankton: NPP = C × μ × Z eu , where NPP is the net primary productivity of phytoplankton, C is the carbon biomass of phytoplankton, μ is the growth rate, and Z eu is the euphotic layer depth; The carbon biomass C is calculated according to the following formula: C = 13000 × (b bp - 0.00035), where b bp is the backscattering coefficient of particulate matter; The growth rate μ is calculated according to the following formula: , where Chl is the chlorophyll concentration and PAR(z) is the photosynthetically active radiation.

[0067] Step nine: Multi - source data verification In multi-source data verification, various measured data are used to verify the water optical chlorophyll parameters retrieved by lidar, and the NPP calculated from buoy parameters is compared to ensure the model accuracy. After calculation, the net primary productivity in the Antarctic Ocean in winter estimated by lidar is 54.5 Tg C, which is 6 times that of the passive ocean color remote sensing estimate (8.5 Tg C).

[0068] Figure 3 It is a comparison between the retrieved parameters of spaceborne lidar and passive ocean color remote sensing products. The correlation coefficient R is above 0.8, the mean absolute percentage error MAPD is within 30%, and the root mean square error RMSD is also very low, indicating the accuracy of the retrieval in Sydney Harbour.

[0069] Figure 4 It is a comparison between the retrieved parameters of spaceborne lidar and the measured results. For marine environmental monitoring, MAPD within 40% is acceptable.

[0070] Figure 5 It is a comparison between the net primary productivity of spaceborne lidar and that of passive ocean color remote sensing. Here, the monthly net primary productivity (Monthly NPP) in different latitude bands of the high-latitude sea area in the Southern Hemisphere is shown. The values in summer are basically the same, and the lidar LiDAR in winter is significantly higher than the passive ocean color remote sensing MODIS results, indicating that lidar makes up for the lack of winter data.

[0071] Figure 6 It is a comparison between the net primary productivity of spaceborne lidar and that of buoy lidar. MAPD is within 20%, indicating the accuracy of the results.

[0072] Figure 7It is a comparison of the winter coverage of spaceborne lidar and the coverage of passive ocean color remote sensing. It shows the comparison of the effective remote sensing observation coverage of passive ocean color remote sensing MODIS and spaceborne lidar CALIOP during the Antarctic winter, presenting the ratio of the remote sensing coverage area to the ice-free sea area and the distribution of the climate net primary productivity in June, July, and August. The comparison of CALIOP (blue) and MODIS (red) data for the entire winter months and for June, July, and August is given. The effective coverage ratios are as follows: winter (80.7% for CALIOP and 12.5% for MODIS), June (61.85% for CALIOP and 0.2% for MODIS), July (87.6% for CALIOP and 0.3% for MODIS), August (92.6% for CALIOP and 37.1% for MODIS). The gray solid line on the map represents the sea ice edge line. It shows that by collaborating the CALIOP lidar with a physically constrained deep learning model, the data coverage rate in the polar winter can be increased from 12.5% of the traditional method to 80.7%, thus achieving all-weather continuous monitoring in the polar night period and the sea ice edge area and filling the gap in winter NPP data in the global carbon cycle model.

[0073] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0074] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows or multiple flows and / or Figure 1 blocks or multiple blocks.

[0075] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in the process Figure 1 a process or processes and / or blocks Figure 1 a block or blocks.

[0076] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the process Figure 1 a process or processes and / or blocks Figure 1 a block or blocks.

[0077] The foregoing is a description of embodiments of the present invention. By the foregoing description of the disclosed embodiments, those skilled in the art can implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for monitoring the net primary productivity of polar winter phytoplankton, characterized in that, It includes the following steps: S1: Multi-source remote sensing data acquisition and spatio-temporal matching. Register the spaceborne lidar data and ocean color remote sensing data through the set time window and space window to ensure spatial coverage and time consistency; S2: Signal correction for the original echo data of the spaceborne lidar, including transient response correction and polarization crosstalk correction; S3: Based on the water body depolarization ratio, sea ice concentration and wind speed data, conduct lidar data quality control to exclude sea ice interference and specular reflection anomalies; S4: Construct a parameter database containing various physical-bio-optical characteristics, and normalize the characteristic variables in the database. The various physical-bio-optical characteristics include the echo signal after correction of the spaceborne lidar, the water body depolarization ratio, the water body backscattering coefficient, and the ocean color remote sensing products; S5: Train a dual-branch deep neural network model based on the database. The first branch extracts echo waveform features based on a convolutional neural network, and the second branch fuses optical parameters for physical feature learning, and outputs the optical chlorophyll concentration after splicing; S6: Implement parameter reconstruction based on a two-step modeling strategy. First, output the monthly average climate state, and then reconstruct the outliers through difference to improve the polar inversion accuracy; S7: Based on the reconstructed parameters, conduct vertical profile light adaptation modeling to obtain the photosynthetically active radiation distribution in the depth direction; S8: Use the reconstructed optical parameters and light distribution to calculate the net primary productivity of phytoplankton based on a carbon-based model; S9: Use buoy data and passive remote sensing data for multi-source cross-validation to evaluate and optimize the productivity inversion accuracy.

2. The method according to claim 1, characterized in that, The transient response correction in step S2 includes deconvolving the original echo signal β(z) using the detector response function F to obtain the corrected signal: β′(z)=F - ¹β(z) where β′(z) is the signal after detector transient response correction, β(z) is the lidar measurement signal, and F is the detector's transient response function; The polarization crosstalk correction is achieved through the following formula: ， ， where CT is the crosstalk coefficient, and are the signals of the parallel channel and the vertical channel respectively after polarization crosstalk correction, and are the signals of the parallel channel and the vertical channel respectively after transient response correction.

3. The method according to claim 1, characterized in that, The depolarization ratio δ of the water body in step S3 T is calculated as follows: ， Among them, δ T is the depolarization ratio of the water body, and p is the position where the peak value is located.

4. The method according to claim 1, characterized in that In the step S4, the backscattering coefficient γ of the water body is calculated according to the following formula: , where γ is the backscattering coefficient of water body, β s is the backscattering coefficient of the sea surface: , where θ is the nadir angle of the laser beam, and σ 2 is the mean square deviation of the sea wave slope: , where W is the sea surface wind speed; The normalization process is: , Among them, X norm is the normalized parameter variable, X is the original variable, is the variable mean, and SD(X) is the variable standard deviation.

5. The method according to claim 1, characterized in that The photosynthetically active radiation PAR(z) in the vertical direction in step S7 is estimated according to the following model: PAR(z) = PAR(0) × exp(-2K d z), PAR(z) is the photosynthetically active radiation at depth z, and PAR(0) is the photosynthetically active radiation at the sea surface K d is the diffuse attenuation coefficient obtained by inversion.

6. The method according to claim 1, wherein The net primary productivity NPP of phytoplankton in step S8 is calculated according to the following formula: NPP = C × μ × Z eu , where NPP is the net primary productivity of phytoplankton, C is the phytoplankton carbon biomass, μ is the growth rate, and Z eu is the euphotic layer depth; The carbon biomass C is calculated according to the following formula: C = 13000 × (b bp - 0.00035), where b bp is the particulate backscattering coefficient; The growth rate μ is calculated according to the following formula: ， where Chl is the chlorophyll concentration and PAR(z) is the photosynthetically active radiation.

7. A monitoring system for the net primary productivity of polar winter phytoplankton, characterized in that, This system implements the method described in any one of claims 1-6, including: a multi-source remote sensing data acquisition module for receiving spaceborne lidar data and ocean color remote sensing data; A spatio-temporal matching module for precisely registering multi-source remote sensing data according to a preset spatial distance window and time window to ensure the consistency of data in spatial coverage and time series; A signal correction module for performing the following two corrections on the spaceborne lidar data: Transient response correction, by deconvolving the original echo signal β(z) using the detector transient response function F to obtain the corrected signal; Polarization crosstalk correction is achieved by converting the corrected parallel-channel and vertical-channel signals; a data quality control module for detecting and filtering abnormal data based on the total depolarization ratio of water bodies, sea ice concentration, and wind speed information; a physical and bio-optical parameter database construction module for integrating the corrected lidar echo signals, depolarization ratio, seawater backscattering coefficient, and remote sensing products, and normalizing the integrated parameters; A dual-branch deep learning model module, including: A first branch that uses a convolutional neural network to extract features from the lidar echo waveform; A second branch that fuses physical parameters such as the water body depolarization ratio and backscattering coefficient output by the data quality control module and the parameter database construction module; The dual-branch features are spliced by a fully connected regression network and then the water body optical chlorophyll parameters are output, realizing waveform-environment collaborative modeling under physical constraints; A parameter reconstruction module for further reconstructing the inversion parameters with a two-step optimization strategy: first output the monthly average climatological parameters, and then reconstruct the outliers based on the difference between the observed values and the climatological values, thereby improving the parameter inversion accuracy; A vertical profile light adaptation modeling module for calculating the photosynthetically active radiation PAR in the depth direction based on the reconstructed water body optical parameters; A carbon-based productivity calculation module for calculating the net primary productivity NPP of phytoplankton according to the light adaptation modeling results and water body parameters, and a multi-source data verification module for verifying the optical parameters and NPP results output by the above-mentioned modules using buoy data, ocean color remote sensing products, and other measured data, thereby ensuring the accuracy and stability of the operation of the entire system.

8. The system according to claim 7, wherein The transient response correction and polarization crosstalk correction formulas used by the signal correction module are implemented as preset modules in the system hardware or software to automatically correct the echo data of the spaceborne lidar; and / or the data quality control module is based on the water body depolarization ratio δ T and the auxiliary sea ice concentration and wind speed data, and can automatically identify and filter abnormal data affected by sea ice, specular reflection and foam interference.

9. A computer device, characterized in that, Including: A processor; A memory; And A computer program stored in the memory, which is executed under the control of the processor to implement the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, on which program instructions are stored, and when the instructions are executed by a computer, they are used to implement the method according to any one of claims 1 to 6.

Images