Multi-source heterogeneous data fusion tree cauliflower growth environment risk early warning method and system

Abstract

The invention provides a tree cauliflower growth environment risk early warning method and system based on multi-source heterogeneous data fusion, and relates to the technical field of smart agriculture, and the method comprises the steps: obtaining the microclimate, matrix environment and other multi-source environment monitoring data of a cultivation region, and determining an environment risk evolution sequence in combination with external prediction data; preliminarily determining an early warning output parameter based on the sequence; calculating a first sub-index representing a wetting favorable factor and a second sub-index representing a waterlogging adverse factor, and determining a conflict constraint index based on a confrontation relationship between the first sub-index and the second sub-index; and performing conflict constraint processing on the early warning output parameter by using the conflict constraint index, and finally outputting early warning containing risk time sequence information when a condition is met. According to the method, the physiological favorable signal and the disaster-causing risk signal in the high-humidity environment are decoupled through the conflict constraint mechanism according to the physiological characteristics that the tree cauliflower likes moisture and feels waterlogging, the problem that the early warning false alarm rate is high due to the fact that the two signal characteristics are overlapped is effectively solved, and the risk identification accuracy is remarkably improved.

Description

Technical Field

[0001] This application relates to the field of smart agriculture technology, and more specifically, to a method and system for early warning of environmental risks in cauliflower growth based on the fusion of multi-source heterogeneous data. Background Technology

[0002] Tree flower vegetable, a woody plant with significant development and utilization value, boasts tender leaves and flower buds rich in various amino acids and vitamins, possessing high nutritional value and health benefits. In the cultivation and promotion of tree flower vegetable, its biological characteristics are characterized by cold resistance, preference for moist conditions, and adaptation to short days, with its distribution primarily occurring in mountainous areas or hilly slopes along rivers at altitudes of 500-1200m. This preference for moist conditions means that it requires maintaining high soil moisture content and air humidity during its vigorous growth period.

[0003] However, the root system of tree cauliflower is quite sensitive to soil aeration. Although it prefers moist conditions, it is also susceptible to waterlogging. If there is too much soil moisture and it cannot drain in time, it will lead to oxygen deficiency in the root zone, which can easily cause root rot or even plant death. Existing agricultural environmental monitoring and early warning technologies usually use a single-dimensional linear threshold discrimination method. That is, when the monitored soil moisture content or rainfall exceeds a set threshold, the system mechanically judges it as a risk of waterlogging and issues an alarm.

[0004] This traditional monitoring logic overlooks the dual nature of environmental factors' effects on plant physiology. For cauliflower, the environmental parameter of high water content has drastically different physiological meanings under different spatiotemporal contexts, such as varying temperatures, micro-topographic slopes, and water flow dynamics: it can be a "physiologically favorable state" that promotes rapid shoot growth, or a "disaster-risk state" that leads to root suffocation. Current technology struggles to distinguish between these two states, often issuing false alarms when environmental parameters are high but actually beneficial to crop growth, causing growers to take unnecessary drainage measures that actually disrupt the suitable humid microclimate; or failing to report when environmental parameters are not extremely high but physiological stress has already occurred.

[0005] Therefore, how to solve the problem of high false alarm rate and poor identification accuracy in environmental risk warning of crops with complex physiological characteristics such as "preferring moisture and fearing waterlogging" has become an urgent technical problem to be solved in the field of precision agriculture. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application provides a method and system for early warning of environmental risks in tree cauliflower growth by fusing multi-source heterogeneous data.

[0007] Firstly, this application provides a method for early warning of environmental risks in *Caulis broccoli* based on the fusion of multi-source heterogeneous data, including:

[0008] Acquire multi-source environmental monitoring data and external prediction data for the target cultivation area, wherein the external prediction data is used to characterize the environmental change sequence within the target prediction time window;

[0009] Based on the multi-source environmental monitoring data and the external prediction data, an environmental risk evolution sequence within the target prediction time window is determined, which is used to characterize the changes of environmental risk indicators over time within the target prediction time window.

[0010] The early warning output parameters are determined based on the environmental risk evolution sequence, and the early warning output parameters include the early warning risk value and the risk time series information corresponding to the early warning risk value;

[0011] Conflict constraint indicators are determined based on the multi-source environmental monitoring data and the external prediction data; conflict constraint processing is then performed on the early warning output parameters based on the conflict constraint indicators.

[0012] When the warning risk value after conflict constraint processing meets the preset warning conditions, a warning information is generated and output. The warning information includes the risk type, risk level, and risk time sequence information after conflict constraint processing.

[0013] Optionally, determining the environmental risk evolution sequence within the target prediction time window includes:

[0014] The multi-source environmental monitoring data and the external prediction data are processed according to the target prediction time window to obtain the target time series data.

[0015] Based on the target time series data, environmental risk indicators are calculated within the target prediction time window at a preset time step to obtain the environmental risk evolution sequence.

[0016] Optionally, the computational environment risk indicators include:

[0017] The temperature prediction sequence is extracted from the external prediction data, and the oxygen consumption intensity coefficient sequence is determined based on the temperature prediction sequence through a preset temperature response function. The preset temperature response function is used to map the temperature prediction values ​​in the temperature prediction sequence to the oxygen consumption intensity coefficient. The parameters of the preset temperature response function are pre-stored in a temperature response parameter table. The temperature response parameter table records several temperature breakpoints and the coefficient values ​​corresponding to each temperature breakpoint. For the temperature prediction value at each time step, breakpoint location and interpolation calculation are performed based on the temperature breakpoints to obtain the corresponding oxygen consumption intensity coefficient.

[0018] Based on the matrix environment data in the target time series data, the oxygen supply channel coefficient sequence is determined by a preset ventilation mapping function. The preset ventilation mapping function is used to map the matrix water content state included in the matrix environment data in the target time series data to oxygen supply channel coefficients. The parameters of the preset ventilation mapping function are pre-stored in a ventilation mapping parameter table. The ventilation mapping parameter table records the water content state breakpoints and the ventilation coefficient values ​​corresponding to each water content state breakpoint. For the matrix water content state at each time step, breakpoint location and interpolation calculation are performed based on the water content state breakpoints to obtain the corresponding oxygen supply channel coefficients.

[0019] A cumulative hypoxia stress index sequence is determined based on the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence.

[0020] The environmental risk indicators are determined based on the cumulative hypoxia stress index sequence, and the environmental risk evolution sequence is formed.

[0021] Optionally, the index for determining conflict constraints includes:

[0022] Based on the multi-source environmental monitoring data and the external prediction data, determine the matrix water content sequence and air humidity sequence within the target prediction time window;

[0023] Obtain a table of tree cauliflower growth threshold parameters, which includes a moisture threshold, a saturation threshold, and a threshold for tolerance to dehydration time.

[0024] Based on the tree cauliflower growth threshold parameter table, the moisture content sequence of the substrate and the air humidity sequence are evaluated for moisture advantage to obtain a moisture advantage score sequence, and the first sub-index is obtained by performing time-series cumulative calculation on the moisture advantage score sequence.

[0025] Based on the tree cauliflower growth threshold parameter table, the waterlogging adverseness evaluation of the substrate water content state sequence is carried out. The waterlogging adverseness evaluation includes calculating the saturation maintenance time and the decline time and obtaining the waterlogging adverseness score sequence accordingly. The waterlogging adverseness score sequence is subjected to time-series cumulative calculation to obtain the second sub-index.

[0026] Normalization is performed on the first sub-index and the second sub-index to obtain a first normalized value and a second normalized value. The first normalized value and the second normalized value are then multiplied to obtain the conflict constraint index.

[0027] Optional, also includes:

[0028] Acquire micro-topographic slope data of the target cultivation area, wherein the micro-topographic slope data is the environmental data of the target cultivation area and is used to determine the second sub-index; determine the baseline drainage curve based on the micro-topographic slope data;

[0029] The actual decay curve of the matrix water content is fitted based on the target time series data, and the morphological deviation of the actual decay curve from the reference receding curve is calculated.

[0030] The risk weight corresponding to the second sub-indicator is adjusted based on the morphological deviation, and the risk weight is used to weight the second sub-indicator when determining the conflict constraint indicator.

[0031] Optionally, determining the cumulative hypoxia stress index sequence includes:

[0032] The hypoxia difference sequence is calculated based on the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence, wherein the hypoxia difference sequence is the difference between the oxygen consumption intensity coefficient and the oxygen supply channel coefficient at each time step.

[0033] The cumulative hypoxia stress index sequence is obtained by performing a recursive cumulative calculation on the hypoxia difference sequence. The recursive cumulative calculation adds the cumulative hypoxia stress index of the previous time step to the cumulative increment obtained by weighting the hypoxia difference of the current time step according to the preset time step length at each time step, so as to obtain the cumulative hypoxia stress index of the current time step.

[0034] Optionally, the morphological deviation is the cumulative amount of the difference between the actual decay rate and the reference dewatering rate at each time step;

[0035] Wherein, the actual decay rate is the rate of change of the water content state of the actual decay curve at the corresponding time step, and the reference water loss rate is the rate of change of the water content state of the reference water loss curve at the corresponding time step.

[0036] The adjustment includes: when the morphological deviation exceeds a preset deviation threshold, setting the risk weight corresponding to the second sub-indicator to a first weight value; and when the morphological deviation does not exceed the preset deviation threshold, setting the risk weight corresponding to the second sub-indicator to a second weight value.

[0037] Optionally, determining the baseline recession curve based on the micro-topographic slope data includes:

[0038] Obtain matrix permeability parameters and a table of tree cauliflower growth threshold parameters that characterize matrix permeability and porosity. The matrix permeability parameters include at least one of permeability coefficient and porosity, and the tree cauliflower growth threshold parameter table includes saturation threshold, wetting threshold, and tolerance to dehydration time threshold.

[0039] A reference drainage curve is calculated based on the micro-topographic slope data and the matrix permeability parameters. The reference drainage curve characterizes the drainage process in which the matrix water content drops from the saturation threshold to the wetting threshold.

[0040] The reference drainage curve is subjected to threshold constraint processing based on the tree cauliflower growth threshold parameter table to obtain the baseline drainage curve.

[0041] Optionally, the threshold constraint processing includes:

[0042] A reference fallback time is determined based on the reference fallback curve, wherein the reference fallback time is the time it takes for the matrix water content, as represented by the reference fallback curve, to fall from the saturation threshold to the wetting threshold.

[0043] The weighting coefficient is determined based on the micro-topographic slope data, and a supplementary weighting coefficient is determined. The supplementary weighting coefficient is 1 minus the weighting coefficient. The weighting coefficient takes a value between 0 and 1 and increases as the micro-topographic slope increases.

[0044] The reference fallback duration is multiplied by the weighting coefficient to obtain the first weighting term, the tolerance fallback duration threshold is multiplied by the supplementary weighting coefficient to obtain the second weighting term, and the first weighting term and the second weighting term are added to obtain the weighted fallback duration.

[0045] The reference recession curve is reconstructed based on the weighted recession duration to obtain the baseline recession curve.

[0046] Secondly, this application provides a risk warning system for the growth environment of *Caulis broccoli* based on the fusion of multi-source heterogeneous data, including:

[0047] The acquisition module is used to acquire multi-source environmental monitoring data and external prediction data of the target cultivation area. The external prediction data is used to characterize the environmental change sequence within the target prediction time window.

[0048] The first processing module is used to determine the environmental risk evolution sequence within the target prediction time window based on the multi-source environmental monitoring data and the external prediction data, and is used to characterize the change of environmental risk indicators over time within the target prediction time window.

[0049] The early warning module is used to determine early warning output parameters based on the environmental risk evolution sequence. The early warning output parameters include an early warning risk value and risk time sequence information corresponding to the early warning risk value.

[0050] The second processing module is used to determine a first sub-indicator representing the favorable factors of humidity and a second sub-indicator representing the unfavorable factors of waterlogging based on the multi-source environmental monitoring data and the external prediction data, and to determine a conflict constraint index based on the first sub-indicator and the second sub-indicator; and to perform conflict constraint processing on the early warning output parameters based on the conflict constraint index.

[0051] The output module is used to generate and output early warning information when the early warning risk value after conflict constraint processing meets the preset early warning conditions. The early warning information includes risk type, risk level, and risk time sequence information after conflict constraint processing.

[0052] Compared with the prior art, the beneficial effects achieved by this application are as follows:

[0053] This application abandons the traditional linear early warning logic based on a single threshold and constructs a dual-channel evaluation system based on physiological duality. Specifically, this application can independently quantify the two dimensions of "favorable moisture" and "unfavorable waterlogging" based on the same set of multi-source environmental data, calculating a first sub-indicator characterizing growth promotion and a second sub-indicator characterizing disaster stress. This application determines a conflict constraint index based on the game relationship between these two sub-indicators and uses this index to dynamically process the conflict constraint of the original risk warning value. This processing method simulates the complex physiological response mechanism of plants. That is, when the system identifies the environment as being in a high-risk range, such as high water content, it will further verify the authenticity of the risk through "favorable moisture" factors, such as good micro-topographic drainage features and suitable respiration temperature. If favorable factors dominate, the system will use the conflict constraint index to significantly suppress the warning output, thereby accurately eliminating false waterlogging warnings caused by overlapping signal features. This application represents a leap from data monitoring to physiological state projection, improving the accuracy of early warning for moisture-loving but waterlogged-sensitive crops like tree cauliflower in complex mountainous environments. It avoids blind human intervention and provides strong technical support for the scientific and refined management of tree cauliflower. Attached Figure Description

[0054] Figure 1 A flowchart of a risk warning system for the growth environment of cauliflower based on multi-source heterogeneous data fusion provided in this application embodiment;

[0055] Figure 2 A flowchart illustrating a method for determining an environmental risk evolution sequence, provided in an embodiment of this application;

[0056] Figure 3 A flowchart illustrating a method for calculating environmental risk indicators provided in this application embodiment;

[0057] Figure 4 This is a schematic diagram of a risk warning system for the growth environment of cauliflower based on multi-source heterogeneous data fusion, provided in an embodiment of this application. Detailed Implementation

[0058] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0059] Example 1

[0060] See Figure 1 This embodiment provides a method for early warning of environmental risks in cauliflower growth based on multi-source heterogeneous data fusion, including steps S101 to S105, wherein:

[0061] S101: Acquire multi-source environmental monitoring data and external prediction data of the target cultivation area, wherein the external prediction data is used to characterize the environmental change sequence within the target prediction time window;

[0062] S102: Based on the multi-source environmental monitoring data and the external prediction data, determine the environmental risk evolution sequence within the target prediction time window, which is used to characterize the change of environmental risk indicators over time within the target prediction time window;

[0063] S103: Determine early warning output parameters based on the environmental risk evolution sequence, wherein the early warning output parameters include an early warning risk value and risk time series information corresponding to the early warning risk value;

[0064] S104: Based on the multi-source environmental monitoring data and the external prediction data, determine a first sub-indicator representing the favorable factors of humidity and a second sub-indicator representing the unfavorable factors of waterlogging, and determine a conflict constraint index based on the first sub-indicator and the second sub-indicator; perform conflict constraint processing on the early warning output parameters based on the conflict constraint index;

[0065] S105: When the warning risk value after the conflict constraint processing meets the preset warning conditions, a warning information is generated and output. The warning information includes the risk type, risk level, and risk time sequence information after the conflict constraint processing.

[0066] Regarding the above S101:

[0067] In one embodiment, the target cultivation area is divided into one or more monitoring micro-zones, each of which is equipped with at least one set of environmental acquisition nodes and one set of data aggregation gateways. The environmental acquisition nodes are used to collect multi-source environmental monitoring data, which includes at least micro-meteorological data and substrate environmental data.

[0068] To ensure that a continuous time series can be formed within the target prediction time window, the environmental acquisition node adds acquisition metadata to each monitoring record. The acquisition metadata includes at least a timestamp, micro-area identifier, and data source identifier. The timestamp can be sent and calibrated by the gateway through the network time synchronization service. The micro-area identifier is used to distinguish plots, ridges, or row / column partitions. The data source identifier is used to distinguish sensor types and device numbers.

[0069] In this embodiment, micrometeorological data is collected by a micrometeorological monitoring component deployed near the ground in the cultivation plot. The micrometeorological monitoring component can be an integrated micrometeorological station or a combination of multiple meteorological sensors, and its output fields include air temperature and relative humidity; in optional implementations, it also includes at least one of rainfall, wind speed, light intensity, and total solar radiation.

[0070] For example, air temperature and relative humidity can be collected by digital temperature and humidity probes, rainfall can be collected by tipping bucket rain gauges, wind speed can be collected by anemometers, and light intensity or radiation can be collected by light sensors or radiation meters. The sampling period for micrometeorological data can be set from 1 minute to 30 minutes; for example, a 10-minute sampling period is used.

[0071] In this embodiment, the matrix environment data is collected by a matrix monitoring component buried near the rhizosphere or in the matrix layer. The matrix monitoring component is used to output the matrix water content status, which can be characterized by volumetric water content, water content level, or matrix water potential. In an optional implementation, the matrix environment data also includes at least one of matrix temperature, electrical conductivity, redox potential, and dissolved oxygen-related quantities to reflect the changing trends of aeration and hypoxia risk.

[0072] For example, the substrate moisture content can be acquired using a capacitive moisture content probe or a frequency domain reflectance moisture content probe, the substrate temperature can be acquired using a thermistor probe, the conductivity can be acquired using a conductivity probe, and dissolved oxygen-related quantities can be obtained using a gas diffusion oxygen sensor or an indirect characterization quantity based on redox potential. For cultivation methods using bagged or layered substrates, the substrate monitoring component can be configured with multiple depth points within the same micro-region to form representative sampling. For example, sensors can be deployed at depths of 5 cm and 15 cm to reduce the impact of single-point measurements on subsequent risk extrapolation.

[0073] In this implementation, the environmental acquisition node and the data aggregation gateway can communicate via wired or wireless means.

[0074] For example, the environmental acquisition node outputs measurement values ​​to the gateway via an RS485 bus and transmits them using a register read method commonly used in industrial settings, or reports measurement frames to the gateway via a low-power wireless link. The data aggregation gateway can be an industrial IoT gateway or an embedded computing device, which runs an acquisition service process to periodically pull or receive monitoring data and write the monitoring data into a time-series data storage device.

[0075] For example, monitoring data is stored in a time series database table in the form of "timestamp - micro-zone identifier - field value". The table structure can be represented as: ts, zone_id, src_id, T_air, RH_air, Rain, VWC, T_sub, EC, DO_flag, quality_flag and other fields. Among them, quality_flag is used to mark states such as missing measurement, exceeding range, communication abnormality, etc., to prevent abnormal values ​​from directly entering subsequent risk inference.

[0076] In this implementation, external forecast data is used to characterize the environmental change sequence within the target forecast time window. External forecast data can come from meteorological service interfaces, gridded numerical weather prediction products, or environmental prediction model outputs, and is mapped to the target cultivation area via an early warning system.

[0077] For example, the external forecast data includes hourly forecast temperature sequences, hourly forecast relative humidity sequences, and hourly forecast rainfall sequences for the next 24 hours. When the external forecast data is grid forecast data, the early warning system can spatially interpolate the grid forecast data based on the geographic coordinates of the target cultivation area or the coordinates of the micro-area center to obtain a forecast sequence corresponding to the target cultivation area. The external forecast data is also supplemented with forecast metadata, which includes at least the forecast start time, forecast step size, and forecast timestamp sequence, enabling it to establish a time correspondence with multi-source environmental monitoring data within the same target forecast time window.

[0078] For ease of understanding, in an exemplary data representation, multi-source environmental monitoring data and external prediction data can be uniformly represented as time-stamped sequence data objects, for example, each time step forming a record:

[0079] {ts: “2025-12-30 10:00:00”, zone_id: “Z01”, T_air: 18.6, RH_air:92.1, Rain: 0.4, VWC: 0.36, T_sub: 17.2, EC: 1.1, forecast_T_air: 19.3, forecast_RH_air: 88.0, forecast_Rain: 1.2, quality_flag: 0}.

[0080] Regarding S102 above:

[0081] In one embodiment, after completing the collection and access of multi-source environmental monitoring data and external prediction data in S101, the early warning system determines the target prediction time window based on a preset early warning scale. The target prediction time window is used to limit the time range of risk simulation.

[0082] For example, a time period from the current moment to several hours or days in the future can be covered, such as the next 24 hours or the next 48 hours; the start and end times of the target prediction time window can be recorded as the window start time and the window end time, and a corresponding relationship can be established with the start time and prediction step size of the external prediction data.

[0083] In this implementation, the early warning system constructs a target time axis for risk simulation within the target prediction time window, and organizes multi-source environmental monitoring data and external prediction data based on this time axis.

[0084] Specifically, the system generates a set of discrete time points within the target prediction time window using a preset time step as the time benchmark for risk extrapolation. It also maps observation records from multi-source environmental monitoring data that fall within the target prediction time window and prediction records from external prediction data that cover the target prediction time window to discrete time points, so that an environmental state input data set corresponding to that time step can be formed at each time step.

[0085] For example, the set of input data corresponding to each time step can be represented by a structured record as: {timestamp, micro-area identifier, micro-meteorological observation field, matrix observation field, external prediction field}, wherein the micro-meteorological observation field includes at least one of air temperature, air relative humidity, and rainfall, the matrix observation field includes at least one of matrix water content, matrix temperature, and electrical conductivity, and the external prediction field includes at least one of predicted air temperature sequence value, predicted humidity sequence value, and predicted rainfall sequence value.

[0086] In this implementation, the early warning system calculates environmental risk indicators based on the input data set corresponding to each time step, and combines the environmental risk indicators of each time step in chronological order to form an environmental risk evolution sequence. The environmental risk indicators are used to characterize the risk level of the cauliflower growth environment under the environmental conditions of the corresponding time step, and can be continuous numerical risk scores, interval-based risk level values, or risk index values ​​with uniform dimensions.

[0087] To meet the needs of subsequent early warning output and constraint processing, environmental risk indicators can be unified to a preset numerical range, for example, corresponding to a risk intensity scale from 0 to 1, where larger values ​​correspond to higher risk levels. The resulting environmental risk evolution sequence can be represented as a sequence object sorted by timestamp, such as {(t1, r1), (t2, r2), …, (tn, rn)}, where n is the number of time steps included in the target prediction time window under the preset time step discreteness, i is the time step number and satisfies 1≤i≤n, ti is the timestamp corresponding to the i-th time step in the target prediction time window, and ri is the environmental risk indicator corresponding to ti.

[0088] This environmental risk evolution sequence can reflect the trend of environmental risk changes over time within the target prediction time window, providing a time-series basis for subsequently determining early warning output parameters and generating early warning information.

[0089] Optional, see Figure 2 The flowchart of a method for determining the evolution sequence of environmental risks provided in this embodiment includes steps S201-S202, wherein:

[0090] S201: The multi-source environmental monitoring data and the external prediction data are processed according to the target prediction time window to obtain the target time series data;

[0091] S202: Based on the target time series data, calculate the environmental risk indicators within the target prediction time window according to a preset time step to obtain the environmental risk evolution sequence.

[0092] In one embodiment, to avoid mismatch in risk projection caused by differences in sampling period, reporting time, and timestamp accuracy between multi-source environmental monitoring data and external prediction data, the early warning system first performs time correspondence processing on multi-source environmental monitoring data and external prediction data oriented towards the target prediction time window before generating the environmental risk evolution sequence. Then, it calculates environmental risk indicators within the target prediction time window according to a preset time step, thereby obtaining the environmental risk evolution sequence that changes over time.

[0093] In this implementation, the start time T0 and end time T1 of the target prediction time window are first determined, and a preset time step Δt is also determined. The preset time step can be configured according to the early warning timeliness requirements and the physiological response speed of the tree cauliflower to environmental stress. For example, given that the root oxygen consumption rate increases exponentially under high temperature, which may lead to acute hypoxia in a short period of time, a shorter time step (such as 10 minutes or 30 minutes) is preferred to capture transient risks; while during the low-temperature dormancy period, a longer step (such as 1 hour) can be selected to reduce the computational load. Based on this, the system generates a target time axis, which is a time step sequence {t1, t2, ..., tn} discretized by Δt within [T0, T1], where n is the number of time steps included in the target prediction time window under the preset time step, and ti is the timestamp of the i-th time step, satisfying 1≤i≤n.

[0094] To facilitate subsequent data alignment, the system can uniformly convert all data timestamps to the same time zone and the same precision, for example, to Beijing time with a precision of seconds, and correct the time drift of the acquisition device through the network time service or NTP service of the gateway.

[0095] In this embodiment, the time correspondence processing includes mapping multi-source environmental monitoring data to a target time axis and mapping external prediction data to the target time axis, thereby obtaining target time series data.

[0096] For multi-source environmental monitoring data, since the sampling frequency may be higher than Δt, the system can aggregate the monitoring records corresponding to each time step and generate the representative value of that time step by using the mean, median or last value hold method, for example. When the monitoring frequency is lower than Δt or there are interruptions, the system can use the forward hold method to fill the missing interval and write the data validity identifier at the same time to avoid missing values ​​directly causing risk misjudgment.

[0097] When the missing time exceeds a preset tolerance threshold (e.g., 1 / 10 of the tolerance time of tree cauliflower), the time step is marked as 'unreliable' or a sensor fault alarm is triggered to avoid misjudging the current risk by referencing outdated historical data.

[0098] For external forecast data, since it is typically output according to the reporting start time and forecast step size, the system can match the forecast sequence with the target time axis by timestamp. For example, adjacent interpolation or nearest neighbor matching can be used to obtain the forecast value for each time step. After time mapping is completed, the target time series data can be structured into records by time step, for example, each time step corresponds to one record: {ti, zone_id, micrometeorological field set, matrix field set, forecast field set, quality_flag}, and stored sorted by timestamp.

[0099] For example, in offline or near real-time implementations, Python data processing components can be used to resample and align the time series data. For instance, pandas' `resample` can be used to aggregate monitoring data by Δt, and `merge_asof` can be used to perform nearest neighbor matching between predicted and monitored data based on timestamps. In streaming implementations, Flink or Spark Streaming can be used to perform window aggregation on the monitoring stream within a sliding window, and the predicted sequences can be joined using time keys as broadcast variables or external dimension tables. The target time series data can be stored in a time series database for retrieval, for example in InfluxDB or TimescaleDB, with "timestamp + micro-region identifier" as the primary key index.

[0100] In this implementation, the system calculates environmental risk indicators based on target time series data and within a target prediction time window at preset time steps to obtain an environmental risk evolution sequence. Environmental risk indicators can be defined as risk measures of the environmental state at each time step, exemplarily as normalized risk scores or risk probability values.

[0101] The calculation of environmental risk indicators can be completed by the risk assessment component, which converts the set of input fields corresponding to each time step into a feature vector, performs risk inference, and outputs the risk value.

[0102] For example, the risk assessment component can use a trained machine learning model for inference. The model can be stored in ONNX format and loaded by the inference service, which can be accessed via an HTTP interface, such as POST / infer, with inputs of {ti, zone_id, feature_vector} and outputs of {ti, risk_score}. In another example, the risk assessment component can be implemented using a rule engine, mapping fields such as temperature, humidity, rainfall, and matrix moisture content to risk scores and combining them for output. The system combines the risk values ​​obtained at each time step in chronological order to form an environmental risk evolution sequence {(t1, r1), (t2, r2), ..., (tn, rn)}, where ri is the environmental risk index corresponding to time step ti, thus characterizing the trend of environmental risk change over time within the target prediction time window.

[0103] Optional, see Figure 3 This embodiment provides a method for calculating environmental risk indicators, including steps S301 to S304, wherein:

[0104] S301: Extract the temperature prediction sequence from the external prediction data, and determine the oxygen consumption intensity coefficient sequence based on the temperature prediction sequence using a preset temperature response function;

[0105] S302: Determine the oxygen supply channel coefficient sequence based on the matrix environment data in the target time series data through a preset ventilation mapping function;

[0106] S303: Determine the cumulative hypoxia stress index sequence based on the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence;

[0107] S304: Determine the environmental risk indicators based on the cumulative hypoxia stress index sequence, and form the environmental risk evolution sequence.

[0108] In one embodiment, to reduce the risk of misjudging between "physiologically favorable humid state" and "disastrous hypoxic waterlogging state" by triggering early warnings based solely on a single threshold such as matrix moisture content or air humidity, the early warning system introduces two observable computational chains based on the obtained target time series data: temperature-driven oxygen consumption intensity and matrix ventilation-driven oxygen supply channels. The cumulative hypoxia stress index sequence is formed by the temporal coupling results of the two chains, thereby upgrading the risk indicator from static threshold judgment to dynamic stress extrapolation oriented towards the target prediction time window.

[0109] In this implementation, the system extracts the temperature prediction sequence corresponding to the target prediction time window from external prediction data and generates an oxygen consumption intensity coefficient sequence through a preset temperature response function. The temperature response function is used to map the temperature prediction value to a dimensionless oxygen consumption intensity coefficient, reflecting the stress driving degree related to rhizosphere respiration intensity under that temperature condition.

[0110] For example, considering that the root respiration of cauliflower conforms to the general laws of biochemical reactions, that is, it is affected by... Temperature coefficient (the factor by which the respiration rate increases for every 10°C increase in temperature) control is used. The temperature response function preferably employs an exponential mapping or piecewise linear mapping: when the air temperature is within the suitable temperature range for cauliflower, a neutral intensity coefficient (e.g., 1.0) is output; when the air temperature deviates from the suitable range and moves towards higher temperatures, an exponentially increasing intensity coefficient is output. The parameters of the temperature response function can be stored in a temperature response parameter table, which provides at least several temperature breakpoints and their corresponding coefficient values. The system performs breakpoint location and interpolation calculations on the predicted air temperature at each time step to obtain the corresponding oxygen consumption intensity coefficient.

[0111] For example, in the Python implementation, the early warning system calls the temperature response calculation process at each time step to map the predicted air temperature value to obtain the corresponding oxygen consumption intensity coefficient; for example, it can execute temp_response(T, table_id) to return the oxygen consumption intensity coefficient. Furthermore, in a service-oriented deployment, the early warning system can call the temperature response calculation service via an HTTP interface, submitting request data containing a timestamp, micro-area identifier, predicted air temperature value, and parameter table identifier to the temperature response interface, and obtaining the oxygen consumption intensity coefficient corresponding to that time step.

[0112] In this embodiment, the system generates an oxygen supply channel coefficient sequence based on matrix environment data in the target time series data through a preset aeration mapping function. The aeration mapping function maps the matrix water content and related observations to dimensionless oxygen supply channel coefficients, reflecting the effectiveness of oxygen reaching the rhizosphere through matrix pore channels under the given matrix conditions. For example, the aeration mapping function uses the matrix water content as the main input and can be modified by incorporating fields such as porosity, permeability, and conductivity: a higher oxygen supply channel coefficient is output when the matrix water content is in the moist but unsaturated range, and a significantly lower oxygen supply channel coefficient is output when the matrix water content approaches saturation and remains at that level. The aeration mapping function can also be implemented using a lookup table interpolation method. The lookup parameters can be stored in a "ventilation mapping parameter table," which records the water content breakpoints and aeration coefficient values, and different entries can be configured for different matrix types.

[0113] For example, in an edge gateway or cloud computing node, the early warning system calls the ventilation mapping calculation process at each time step to map the matrix water content and optional fields such as porosity and permeability coefficient to oxygen supply channel coefficients; for example, the ventilation_map(vwc, porosity, ksat, table_id) can be executed to obtain the oxygen supply channel coefficients.

[0114] Furthermore, in the microservice implementation, the early warning system can call the ventilation mapping service through the HTTP interface, submit request data including timestamp, micro-area identifier, matrix water content status and parameter table identifier to the ventilation mapping interface, and obtain the oxygen supply channel coefficient.

[0115] In this implementation, the system determines the cumulative hypoxia stress index sequence based on the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence. The cumulative hypoxia stress index sequence is used to characterize the degree of accumulation of the adverse state of "strong oxygen consumption drive and weak oxygen supply channel" over time within the target prediction time window, avoiding triggering early warning based solely on instantaneous anomalies at a certain time step.

[0116] For example, at each time step, the system first calculates the hypoxia driving force corresponding to that time step. This driving force is determined by the antagonistic relationship between the oxygen consumption intensity coefficient and the oxygen supply channel coefficient. Subsequently, the driving force is cumulatively updated over time steps to obtain the cumulative hypoxia stress index for the current time step. The cumulative update can be implemented by a "stress recursion module," which maintains the cumulative value of the previous time step as a state variable on the software side and completes the update by combining the driving force of the current time step with a preset time step length.

[0117] For example, in the implementation, the function hypoxia_accumulate(prev_index, demand_coeff, supply_coeff, delta_t) is used to output curr_index and write curr_index to the sequence storage associated with the timestamp.

[0118] In this implementation, the system determines environmental risk indicators based on the cumulative hypoxia stress index sequence and forms an environmental risk evolution sequence. Environmental risk indicators can be defined as normalized risk scores, with a value range of 0 to 1; higher values ​​correspond to increased risk. For example, the system maps the cumulative hypoxia stress index to risk scores using preset risk mapping rules: a low-risk score is output when the cumulative hypoxia stress index is in a low range, and a significantly increased risk score is output when the cumulative hypoxia stress index exceeds a threshold range, which can be further mapped to a risk level. The risk mapping rules can be configured in a "Risk Mapping Parameter Table" using a lookup table, where entries record the correspondence between stress index ranges and risk scores and risk levels.

[0119] The system provides a "POST / risk_map" interface, with inputs of {timestamp, zone_id, hypoxia_index, map_id} and outputs of {timestamp, risk_score, risk_level}. The system combines the risk scores output at each time step in timestamp order to obtain the environmental risk evolution sequence corresponding to the target prediction time window, providing time-series input for subsequent early warning output parameter generation and conflict constraint handling.

[0120] Optionally, to avoid the risk of being overly sensitive to short-term fluctuations or insufficiently responding to persistent hypoxia stress by calculating the risk based solely on the instantaneous state at a certain time step, the early warning system first constructs a hypoxia difference sequence after obtaining the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence. Then, the hypoxia difference is weighted according to a preset time step and recursively accumulated to form a cumulative hypoxia stress index sequence that can simultaneously characterize "intensity and duration".

[0121] In this implementation, the hypoxia difference sequence is used to characterize the relative difference between the oxygen consumption drive level and the oxygen supply channel capacity at each time step. For each time step within the target prediction time window, the system reads the corresponding oxygen consumption intensity coefficient and oxygen supply channel coefficient, and calculates the hypoxia difference for that time step, ensuring that the hypoxia difference equals the difference between the oxygen consumption intensity coefficient and the oxygen supply channel coefficient. The resulting hypoxia difference sequence can be organized sequentially by time step as {d1, d2, ..., dn}, where n is the number of time steps included in the target prediction time window within a preset time step length, and di is the hypoxia difference corresponding to the i-th time step. To ensure that this sequence can be directly used for subsequent recursive accumulation, the system can store the hypoxia difference sequence in a one-to-one correspondence with the timestamp sequence, for example, writing it into the time series storage in the form of (timestamp ti, hypoxia difference di).

[0122] In this implementation, recursive accumulation is used to integrally accumulate the hypoxia difference over time to form a cumulative hypoxia stress index sequence. The system sets an initial cumulative hypoxia stress index at the beginning time step of the target prediction time window. For example, the initial value can be set to zero, or in a continuous operation scenario, the cumulative value at the end of the previous prediction cycle can be used as the starting value to maintain the continuity of the stress state. Subsequently, the system calculates a "cumulative increment" for each time step. The cumulative increment is determined by the hypoxia difference at the current time step and a preset time step length. The preset time step length is used to convert the difference into an incremental contribution related to the duration. For example, the preset time step length can be 10 minutes, 30 minutes, or 1 hour. The system can weight the hypoxia difference according to this step length to obtain the cumulative increment at the current time step.

[0123] In this implementation, when the system updates the cumulative hypoxia stress index at the i-th time step, it superimposes the cumulative hypoxia stress index of the previous time step with the cumulative increment of the current time step to obtain the cumulative hypoxia stress index of the current time step, thereby forming a cumulative hypoxia stress index sequence {h1, h2, ..., hn} that is updated recursively by time step. Here, hi is the cumulative hypoxia stress index at the i-th time step, h1 is obtained by superimposing the initial value with the cumulative increment of the first time step, and hi is obtained by superimposing h(i-1) with the cumulative increment of the i-th time step.

[0124] In one optional implementation, when the hypoxia difference is negative or indicates that the oxygen supply capacity is relatively dominant, the system can use the hypoxia difference to suppress the cumulative growth or perform a slow decline on the cumulative value to reflect the recovery trend after the stress is relieved. In another optional implementation, the system can also set upper and lower limits on the cumulative hypoxia stress index to avoid abnormal divergence of the cumulative value due to abnormal measurements.

[0125] Through the above-mentioned hypoxia difference construction and recursive accumulation process, the cumulative hypoxia stress index sequence can reflect the cumulative evolution of hypoxia stress within the target prediction time window with a unified time step update rule, providing a stable time series input for subsequent risk indicator mapping and early warning output.

[0126] Regarding the above S103:

[0127] In practical implementation, after obtaining the environmental risk evolution sequence within the target prediction time window, the early warning system executes an early warning output parameter generation process to compress the time-varying risk sequence into a set of output parameters that facilitates alarm decision-making and response scheduling. These output parameters include at least the early warning risk value and the corresponding risk time-series information.

[0128] In this implementation, the environmental risk evolution sequence is stored as an ordered pair {(t1,r1),(t2,r2),…,(tn,rn)} indexed by time steps, where ti is the timestamp of the i-th time step within the target prediction time window, and ri is the risk index value at that time step, which is, for example, a continuous value between 0 and 1. The system simultaneously records the current time t0, the preset time step size Δt, and the start and end times T0 and T1 of the target prediction time window. The environmental risk evolution sequence can be stored as a time series object in a time series library, or it can be cached in memory as an array, for example, represented by two equal-length arrays time_list=[t1…tn] and risk_list=[r1…rn].

[0129] In this implementation, the system first determines the warning risk value. For example, the warning risk value can be defined as the peak value of the risk indicator within the target prediction time window, reflecting the most severe risk level within the future time window. The system performs a peak search on the risk_list to obtain the maximum value r_max and records its corresponding peak time t_peak. When multiple identical peaks exist, the system can take the earliest peak time as t_peak to facilitate early intervention. The system can also introduce an "effective warning threshold" r_th, exemplarily set to 0.7, to distinguish between "high-risk segments requiring warning" and "background fluctuations." Under this threshold, if r_max is less than r_th, the warning risk value can still be output as r_max, but the system will mark the risk timing information as "no high-risk segment formed" to avoid false alarms.

[0130] In this implementation, the system further determines risk timing information. Risk timing information is used to characterize the start and end times and duration of high-risk segments, and includes, for example, the risk start time t_start, risk end time t_end, peak time t_peak, and lead time. To avoid false alarms caused by single-point spikes, the system uses a "continuous exceedance threshold persistence criterion" to locate high-risk segments: a minimum duration step m_min is set, for example, 3 time steps. The system scans the risk_list in chronological order, searching for time periods that satisfy ri being continuously not less than r_th and having a continuous length not less than m_min, and records each time period that meets the conditions as a candidate risk segment. If multiple candidate risk segments exist, the system can select the risk segment with the largest peak value within the segment as the primary risk segment; when multiple risk segments have the same peak value within the segment, the system can select the risk segment with the earliest start time as the primary risk segment, thus favoring earlier warnings. Once the main risk segment is determined, t_start takes the timestamp corresponding to the first time step of the segment, and t_end takes the timestamp corresponding to the last time step of the segment; the lead time is determined by the time difference between t_start and the current time t0, and can be expressed in hours or time steps.

[0131] For example, if the target prediction time window is the next 24 hours, Δt is 1 hour, then n is 24. In a certain calculation, the system obtains the peak value of risk_list as r_max = 0.86, corresponding to t_peak at 16:00 on the current day. Assuming r_th is 0.7 and m_min is 3, and the scan shows that for five consecutive time steps from 14:00 to 18:00, the value is not less than 0.7, then t_start = 14:00, t_end = 18:00, and t_peak = 16:00. If the current time t0 is 10:00, then lead_time is 4 hours. Therefore, the warning output parameters can be organized into a structured record, such as {warning risk value: 0.86, risk start time: 14:00, risk end time: 18:00, peak time: 16:00, lead time: 4 hours}, and used as input for subsequent conflict constraint processing and warning information generation.

[0132] The above process can be executed by the early warning parameter generation module. In the Python implementation, the environmental risk evolution sequence can be represented as a pandas Series object, with the index being a timestamp and the value being a risk indicator; the peak and peak time are obtained through max and idxmax, and the high-risk segment is obtained by grouping the Boolean sequence of "whether it exceeds the threshold" into continuous segments and filtering the continuous length.

[0133] Regarding S104 above:

[0134] Base sequence data for wet and waterlogging assessment are extracted within the target prediction time window.

[0135] For example, the system obtains the matrix moisture state sequence M(ti) and the air humidity sequence H(ti) from the target time series data, and the rainfall prediction sequence P(ti) from external prediction data; where ti is the timestamp of the i-th time step within the target prediction time window. The system also reads the cauliflower growth threshold parameter table as an evaluation benchmark. The threshold parameter table includes at least: a wetness threshold W_wet, a saturation threshold W_sat, and a tolerance for receding water duration threshold D_tol.

[0136] Threshold parameter tables can be stored in the form of configuration files or database tables. For example, they can be stored in the parameter library as records of {crop number, W_wet, W_sat, D_tol} and retrieved and loaded according to crop number and cultivation mode.

[0137] In this embodiment, the system constructs a first sub-index characterizing the factors favorable to humidification based on the basic sequence data.

[0138] Exemplarily, the system calculates a wetting favorable score s_pos(ti) at each time step ti, which is used to characterize the degree of "wetting degree meeting physiological favorability". Its inputs at least include: the substrate water content state M(ti) and the air humidity H(ti), and are mapped based on the wetting threshold W_wet.

[0139] To avoid the score being sensitive to single-point noise, the system performs time accumulation on the wetting favorable score to obtain the first sub-index I_pos; for example, I_pos is obtained by weighted summing the wetting favorable scores of each time step according to a preset time step Δt, or by accumulating the duration of continuous wetting conditions within the target prediction time window.

[0140] Exemplarily, when the suitable wetting interval corresponding to W_wet is that the substrate water content state is not less than 0.28 and the air humidity is not less than 85%, the system can count the time steps that meet the conditions as effective wetting steps, multiply the number of effective wetting steps by Δt to obtain the wetting duration, and then convert it into I_pospos accordingly.

[0141] In this embodiment, the system constructs a second sub-index representing waterlogging adverse factors based on the basic sequence data. Exemplarily, the system identifies the time period when the substrate water content state reaches or exceeds the saturation threshold W_sat within the target prediction time window, and calculates the saturation holding duration D_sat; at the same time, it identifies the falling process when the substrate water content state falls back from the saturation threshold to the wetting threshold W_wet, and calculates the falling duration D_rec. The system then calculates the waterlogging adverse score s_neg(ti) or the waterlogging adverse cumulative amount I_neg, so that the waterlogging adversity reflects both "whether saturation occurs" and "saturation duration and falling lag".

[0142] Exemplarily, the time steps that satisfy M(ti)≥W_sat can be counted as saturation steps and accumulated to obtain D_sat; during the post-rainfall falling stage, the system records the moment when M(ti)<W_sat is first satisfied and the moment when M(ti)≤W_wet is first satisfied, and the difference between the two is used as D_rec. Subsequently, the system uses the tolerance falling water duration threshold D_tol as a comparison benchmark. If the combined index of D_sat and D_rec is close to or exceeds D_tol, the waterlogging adverse score is increased, and time accumulation is performed on it to obtain the second sub-index I_neg.

[0143] In this embodiment, to enable the two types of sub-indices to form conflict constraints under the same dimension, the system performs normalization processing on the first sub-index and the second sub-index to obtain the normalized wetting favorability N_pos and the normalized waterlogging adversity N_neg. [[ID=​For example, normalization can be scaled according to the target prediction time window length by dividing I_pos and I_neg by the maximum possible cumulative amount within the window or the upper limit of the window duration, so that N_pos and N_neg fall within the range of 0 to 1. The system then determines the conflict constraint index C_conf based on N_pos and N_neg. For example, C_conf is in a product form so that it reflects the coexistence of "wetness is significantly beneficial" and "waterlogging is significantly detrimental". That is, when one of them is low, the conflict level automatically decreases, and when both are high, the conflict level increases significantly.

[0145] In this implementation, the system performs conflict constraint processing on the early warning output parameters based on conflict constraint indices. The goal of conflict constraint processing is to increase the evidence requirements for early warning triggering when favorable wet conditions and unfavorable waterlogging conditions strongly conflict within the same prediction time window, suppress false alarms driven by a single signal, and make the risk time series information closer to the period when "unfavorable waterlogging conditions are dominant or their persistence is insufficient for recovery".

[0146] For example, the system sets a linkage rule between the conflict level threshold C_th and the risk trigger threshold: when C_conf does not exceed C_th, the warning risk value and risk timing information remain in their original output; when C_conf exceeds C_th, the system raises the trigger threshold of the warning risk value to a higher threshold r_th_high, and limits the risk start time to the earliest time step that simultaneously satisfies "the risk index is not lower than r_th_high and N_neg is not lower than the preset unfavorable threshold", thereby avoiding premature triggering of alarms in the stage where "wetness is favorable but waterlogging is unfavorable and has not yet become dominant".

[0147] For example, when the basic trigger threshold r_th is 0.70, the boosted threshold r_th_high is 0.80, and C_th is 0.60, if the risk peak within a certain time window is 0.78 but C_conf is 0.72, the system will not output a valid warning start time. Only when the risk rises to 0.82 in subsequent time steps and N_neg reaches the unfavorable threshold will the system determine that time step as the risk start time and update the risk timing information. Through the above conflict constraint processing, the warning output parameters, while maintaining sensitivity to high-risk events, can reduce the false alarm rate and improve the identification accuracy for the signal overlap characteristics of crops that "prefer moisture and are afraid of waterlogging".

[0148] Optionally, to ensure that the assessment of waterlogging adverseness reflects the environmental constraint that "water should recede faster" under sloping cultivation conditions, the early warning system introduces micro-topographic slope data when constructing the second sub-indicator, and uses the slope data to form a benchmark receding curve for comparison. Then, the matrix water content sequence in the target time series data is fitted to the actual decay curve, and the risk weight corresponding to the second sub-indicator is adjusted by the morphological deviation between the two, so that the second sub-indicator has a differentiated weight response under different slope conditions when calculating subsequent conflict constraints.

[0149] In this embodiment, the micro-topographic slope data can be obtained using digital elevation data of the cultivated plot or field survey data.

[0150] For example, the system can use a digital surface model (DSM) or a digital elevation model (DEM) generated by UAV aerial survey as a terrain data source, and crop the DEM with the micro-area boundary of the target cultivation area to obtain the elevation raster of each micro-area; alternatively, RTK surveying equipment can be used to collect elevation points at key points in the micro-area, and local slopes can be generated by interpolation.

[0151] The system calculates slope at the micro-region scale, for example using the three-point method or the grid gradient method. The slope value is then written into the environmental data storage in the form of "micro-region identifier - slope value," serving as environmental data for the target cultivation area in the determination of the second sub-indicator. To improve stability, the system can perform noise reduction processing on the slope value, for example, by taking the mean or median of the slope from multiple sampling points in the same micro-region as the representative slope value for that micro-region.

[0152] In this embodiment, the baseline drainage curve is used to characterize the reference trajectory of the matrix moisture content decreasing over time under given slope conditions and the combined effects of gravity-driven and conventional infiltration conditions. The baseline drainage curve can be generated in a sequence format consistent with the target time axis according to a preset time step. The baseline drainage curve can be generated from a "slope-drainage curve library" or a "slope-drainage parameter library": the system looks up the corresponding slope grade based on the micro-area slope value, reads the drainage parameters of that grade, and generates the baseline curve.

[0153] For example, a baseline curve can be generated using an exponential decline method. This involves setting a baseline initial water level at an initial moment and generating a decreasing sequence over time based on the time constant or decay factor corresponding to different slope gradients. The steeper the slope, the smaller the corresponding time constant or the larger the decay factor, resulting in a faster decline in the baseline curve. The aforementioned curve or parameter library can be pre-configured based on historical monitoring data, experimental data, or empirical models during system deployment, and multiple versions can be created separately for different regions and substrate types to adapt to engineering projects.

[0154] In this implementation, the system fits the actual decay curve based on the target time series data. Specifically, the system reads the matrix water content sequence from the target time series data and identifies the "receding water segment" within the target prediction time window to avoid mixing in the fitting of continuous rainfall or irrigation phases.

[0155] For example, the system can combine rainfall prediction sequences or monitored rainfall data to determine several time steps after rainfall stops as the starting point of the receding water stage, and use the matrix water content sequence within the receding water stage as the fitting sample. The actual decay curve can be generated using a smoothing fitting method to suppress the interference of sensor noise on morphological judgment. For example, a moving average smoothing method can be used followed by exponential fitting or spline fitting to obtain the actual receding sequence at the same time step as the baseline curve. For different micro-regions, the system generates corresponding actual decay curves and records the fitting residuals or the proportion of effective samples as data validity indicators to prevent curve morphology distortion caused by missing measurements.

[0156] In this implementation, the system calculates the morphological deviation of the actual attenuation curve relative to the benchmark recession curve and adjusts the risk weight corresponding to the second sub-index accordingly. The morphological deviation is used to measure the degree of deviation of the actual recession process from the benchmark recession process, and can be characterized by, for example, the area of ​​curve difference, the weighted cumulative amount of point-to-point difference, or a similarity index based on attenuation rate characteristics.

[0157] The system limits the calculation of morphological deviation to the same recession period and accumulates or statistically analyzes it step by step in accordance with a preset time step. Subsequently, the system maps the morphological deviation to a risk weight adjustment: when the morphological deviation indicates that the actual recession is significantly slower than the baseline recession, the system increases the risk weight corresponding to the second sub-indicator, increasing the proportion of waterlogging adverseness in the conflict constraint calculation; when the morphological deviation indicates that the actual recession is consistent with or faster than the baseline recession, the system decreases the risk weight corresponding to the second sub-indicator, decreasing the proportion of waterlogging adverseness in the conflict constraint calculation. For example, the system can limit the risk weight to a preset range and divide the morphological deviation into several levels using a segmented mapping rule, assigning corresponding weight values ​​to ensure consistency with the risk level system of the early warning system.

[0158] In this way, the second sub-index not only reflects the waterlogging disadvantage of the substrate moisture content itself, but also explicitly incorporates the physical constraint of drainage represented by the slope. This makes it easier to amplify and identify the anomaly of "slow drainage" in sloping or mountainous cultivation scenarios and enter into subsequent conflict constraint processing, thereby reducing false alarms under the favorable moist conditions and improving the ability to distinguish waterlogging risks.

[0159] In one embodiment, to ensure that the deviation of the actual decay curve from the reference receding curve can be used in an operable numerical form to participate in the risk weight adjustment of the second sub-index, the system constructs the morphological deviation as the cumulative difference between the actual decay rate and the reference receding rate at each time step. The actual decay rate reflects the rate of change of the matrix moisture content within the receding section over time, while the reference receding rate reflects the rate of change of the reference receding curve at the same time step. The difference between the two is used to characterize the hysteresis of the actual receding rate relative to the reference receding rate, and then time accumulation is used to suppress misjudgments caused by single-point noise.

[0160] In this embodiment, the system acquires curve values ​​for adjacent time steps within the receding water section at a preset time step Δt, and calculates the rate of change for the corresponding time step. For example, the system uses the fitted value of the water-bearing state as the curve value, and the actual decay rate can be calculated by the difference between adjacent time steps: at time step ti, the change in water-bearing state fitted value of the actual decay curve at ti and ti-1 is obtained, and then divided by Δt to get the rate; in this embodiment, i is the time step index, used to identify the i-th sampling time after discretization at a preset time step Δt within the target prediction time window. Similarly, the baseline receding water rate is obtained by dividing the difference between the baseline receding water curve values ​​at ti and ti-1 by Δt. To improve the stability of the rate calculation, the system can perform smoothing processing on the water-bearing state curve before differencing, for example, using a moving average with a window of 3 to 5 time steps, or using median filtering to remove isolated spikes; smoothing only applies to the curve value level and does not change the time boundary of the receding water section.

[0161] In this embodiment, the system calculates the rate difference at each time step and accumulates the rate difference within the receding water section to obtain the morphological deviation. For example, at time step ti, the system calculates the rate difference Δv(ti) = v_act(ti) - v_base(ti), where v_act(ti) is the actual decay rate and v_base(ti) is the baseline receding water rate. Subsequently, the system accumulates Δv(ti) within the receding water section sequentially according to the time steps to obtain the morphological deviation D. Since the water content typically decreases during receding water, if the rate of change is calculated as "water content difference divided by Δt," then when the actual receding water is slower, the actual decay rate is closer to zero, thus relatively increasing Δv(ti). The accumulated morphological deviation increases with the increase in "receding water hysteresis," thereby maintaining consistency with the subsequent threshold discrimination direction.

[0162] In this implementation, the system sets a preset deviation threshold D_th and adjusts the risk weight corresponding to the second sub-indicator in a binarized manner accordingly. For example, when the morphological deviation D exceeds the preset deviation threshold D_th, the system sets the risk weight corresponding to the second sub-indicator to a first weight value w1; when the morphological deviation D does not exceed the preset deviation threshold D_th, the system sets the risk weight corresponding to the second sub-indicator to a second weight value w2. The first and second weight values ​​can be configured in the system parameters, and are typically set to w1 greater than w2, so as to increase the influence of waterlogging adverse factors in the conflict constraint calculation when "actual water recedes significantly behind the baseline water recedes"; conversely, to reduce this influence when "actual water recedes in line with or faster than the baseline water recedes". The above weight switching results then participate in the weighting processing of the second sub-indicator, enabling the anomaly of water receding delay under slope conditions to be more sensitively reflected in the conflict constraint stage.

[0163] In one embodiment, to ensure that the baseline drainage curve can simultaneously reflect the topographic drainage conditions of the target cultivation area, the infiltration capacity of the substrate, and the tolerance boundary of the cauliflower to the "saturation-decline" process, the system, when determining the baseline drainage curve, in addition to using micro-topographic slope data, also obtains substrate permeability parameters characterizing substrate permeability and porosity, and loads the cauliflower growth threshold parameter table; subsequently, based on the slope data and substrate permeability parameters, a reference drainage curve is first calculated, and then the reference drainage curve is subjected to threshold constraint processing using the cauliflower growth threshold parameter table, thereby obtaining the baseline drainage curve used for subsequent morphological deviation calculation.

[0164] In this embodiment, the matrix permeability parameters are used to characterize the drainage capacity and water migration resistance of the matrix under saturated or near-saturated conditions, and include at least one of the permeability coefficient and porosity. The matrix permeability parameters can be obtained through a combination of field and laboratory testing. For example, the permeability coefficient can be determined through a double-ring infiltration test, a constant head or variable head infiltration test; in facility cultivation or substrate bag cultivation scenarios, it can also be measured using a permeameter after sampling at representative sampling points. Porosity can be calculated based on the conversion between matrix bulk density and particle density. Bulk density can be determined by ring sampling and weighing, and particle density can be determined using a specific gravity bottle or known material parameters. The system associates and stores the permeability coefficient and porosity corresponding to the same micro-region with the micro-region identifier, enabling it to participate in curve calculations together with the slope data of that micro-region; when multiple sampling points exist in the same micro-region, the mean or median can be taken as the representative parameter of that micro-region, and the sampling quantity and dispersion are recorded for quality control.

[0165] In this embodiment, the tree cauliflower growth threshold parameter table is used to provide key thresholds related to the tree cauliflower's characteristics of "beneficial moisture, harmful saturation, and limited tolerance duration," including at least a saturation threshold, a moisture threshold, and a tolerance duration threshold for water loss. These thresholds can be determined from cultivation trial data, regional agronomical specifications, or historical planting experience, and stored in a structured parameter database. For example, the saturation threshold and moisture threshold can correspond to threshold ranges of substrate volumetric moisture content or substrate water potential, and the tolerance duration threshold for water loss is used to constrain the maximum allowable time for the water content to drop from the saturation threshold to the moisture threshold. The threshold parameter table can be configured with multiple entries according to variety, cultivation season, or growth stage. Before calculating the baseline water loss curve, the system selects the corresponding entry based on the crop identification and cultivation mode of the target cultivation area.

[0166] In this implementation, the system calculates a reference drainage curve based on micro-topographic slope data and matrix permeability parameters. The reference drainage curve characterizes the trajectory of the matrix water content falling from the saturation threshold to the wetting threshold under the current slope and matrix drainage conditions. The system can generate this trajectory using a parametric drainage model. The key to the drainage model is determining the "drainage rate" or "drainage timescale," which is jointly determined by the gravity drainage driven by the slope and the channel capacity represented by matrix permeability / porosity. For example, the system can first calculate the equivalent gravity component along the slope direction based on the slope, then combine the permeability coefficient and porosity to determine the equivalent drainage capacity parameters, and accordingly determine the drainage timescale. Subsequently, using the saturation threshold as the initial water content level of the reference drainage curve, the water content level is updated recursively at a preset time step within the target prediction time window, generating a reference drainage curve that monotonically declines over time until it falls back to near the wetting threshold or reaches the end of the target prediction time window. To avoid unreasonable fluctuations in the curve caused by abnormal parameters, the system can impose basic constraints on the decline process, such as limiting the water content level to not be higher than the saturation threshold and not lower than the wetting threshold, and keeping the decline direction consistent.

[0167] In this embodiment, the system performs threshold constraint processing on the reference drainage curve based on the tree cauliflower growth threshold parameter table to obtain the baseline drainage curve.

[0168] The threshold constraint processing includes at least a threshold verification of the fall time of the reference fall curve: the system identifies the fall time corresponding to the fall of the reference fall curve from the saturation threshold to the wetting threshold, and compares it with the tolerance fall time threshold.

[0169] When the duration of the fall of the reference fall curve is within the threshold of the tolerance fall duration, the system can directly determine the reference fall curve as the baseline fall curve; when the duration of the fall of the reference fall curve exceeds the threshold of the tolerance fall duration, the system performs boundary constraint processing on the reference fall curve so that the baseline fall curve reflects the tolerance boundary of the tree cauliflower to "saturation persistence and fall hysteresis".

[0170] Boundary constraint processing can be manifested by limiting key parameters of the receding process, ensuring that the receding time of the baseline receding curve within the "saturation threshold - wetting threshold" range does not exceed the tolerance receding time threshold, while maintaining the monotonic receding characteristics of the curve consistent with the threshold boundary. The baseline receding curve obtained after this threshold constraint processing serves as the reference basis for subsequent comparison with actual attenuation curves and calculation of morphological deviation, thereby ensuring that slope, permeability, and crop tolerance boundaries are consistently reflected in the same reference curve.

[0171] Optionally, after generating the reference drainage curve and loading the tree cauliflower growth threshold parameter table, a threshold constraint processing is performed on the reference drainage curve so that the benchmark drainage curve used for subsequent morphological deviation calculations simultaneously reflects two types of constraints: "the steeper the slope, the stricter the drainage requirements" and "tree cauliflower has an upper limit to its tolerance to the saturation-decline process".

[0172] The core operation of this threshold constraint processing is as follows: first, extract the reference receding time from the reference receding curve; then, determine the weight coefficient and its supplementary weight coefficient based on the micro-topography slope; then, integrate the reference receding time with the tolerance receding time threshold according to the weight to obtain the weighted receding time; finally, reconstruct the reference receding curve on a time scale using the weighted receding time to form the baseline receding curve.

[0173] In this embodiment, the reference setback duration is determined by the time it takes for the matrix moisture content to fall from the saturation threshold to the wetting threshold. The system locates the setback start and end points from the reference setback curve: the setback start point can be the timestamp corresponding to the starting time step of the reference setback curve, or the time step when the curve first satisfies the moisture content not exceeding the saturation threshold; the setback end point is the time step when the curve first satisfies the moisture content not exceeding the wetting threshold.

[0174] If the reference receding water curve crosses the wetting threshold in a discrete time step, the system can perform a linear interpolation on the crossed interval to obtain a more accurate endpoint timestamp. That is, when the water content crosses the wetting threshold at two adjacent time steps with one high and one low water content, the endpoint timestamp is obtained according to the proportional position of the threshold between the two points. The reference receding time is obtained by subtracting the receding starting timetamp from the receding endpoint timestamp, and is expressed in hours or minutes to maintain consistency with the tolerable receding time threshold.

[0175] In this implementation, the weighting coefficient is determined by micro-topographic slope data and its value is limited to between 0 and 1, increasing with the slope. For example, the system first performs segmented normalization on the micro-slope values ​​of the target cultivation area: Let the lower slope limit S_low and the upper slope limit S_high be system configuration parameters, corresponding to the representative slopes of "gentle slope" and "steep slope," respectively. When the micro-slope S is not greater than S_low, the weighting coefficient is set to 0; when the micro-slope S is not less than S_high, the weighting coefficient is set to 1; when S is between (S_low, S_high), the weighting coefficient is set to (S-S_low) / (S_high-S_low). To avoid weight jumps caused by slope measurement noise, the system can first smooth the slope values ​​or take the median of the slopes at multiple points in the same micro-area before substituting them into the above mapping to calculate the weighting coefficient. The supplementary weighting coefficient is determined by subtracting the weighting coefficient from 1, ensuring that the sum of the two is always 1.

[0176] In this implementation, the system multiplies the reference receding time by a weighting coefficient to obtain a first weighted term, multiplies the tolerance receding time threshold by a supplementary weighting coefficient to obtain a second weighted term, and adds the first and second weighted terms to obtain a weighted receding time. This weighted receding time is used to reflect the adjustment of slope on receding constraints: in steep slope micro-regions, the weighting coefficient is larger, and the weighted receding time is closer to the reference receding time, thus being more sensitive to "slow receding"; in gentle slope micro-regions, the supplementary weighting coefficient is larger, and the weighted receding time is closer to the tolerance receding time threshold, thus avoiding overly stringent misjudgments due to weak terrain-driven factors.

[0177] In this implementation, the system reconstructs the reference recession curve based on the weighted recession time to obtain the baseline recession curve. The reconstruction can be achieved by time scale scaling, that is, keeping the recession shape of the reference recession curve on the water content axis unchanged, and adjusting the time axis so that the time taken for the curve to recession from the saturation threshold to the wetting threshold is changed from the reference recession time to the weighted recession time.

[0178] For example, the system calculates the time scale coefficient k = T_ref / T_w, where T_ref is the reference fall time and T_w is the weighted fall time. Then, for each discrete time step of the baseline fall curve, the system uses the relative time τ from the fall start point of that time step as the independent variable to calculate the corresponding reference curve sampling time τ′ = k·τ, and reads the water content at τ′ on the reference fall curve by interpolation as the water content at τ of the baseline fall curve.

[0179] In this way, when T_w is less than T_ref, the baseline recession curve is compressed on the time axis, resulting in a faster decline; when T_w is greater than T_ref, the baseline recession curve is stretched on the time axis, resulting in a slower decline.

[0180] To ensure threshold consistency, the system applies boundary constraints to the curve endpoints during the reconstruction process: the water content corresponding to the fall-off start point is not higher than the saturation threshold, the water content corresponding to the fall-off end point is not higher than the wetting threshold, and the curve remains monotonically constant over time within the fall-off interval, so that the baseline receding curve can serve as a stable reference for subsequent comparison of actual attenuation curves.

[0181] Optionally, to address the issue of unstable early warning triggering caused by the overlap of "moisture-favorable" and "waterlogging-unfavorable" conditions for tree cauliflower within the same environmental parameter space, the system constructs a moisture-favorable evaluation link and a waterlogging-unfavorable evaluation link in parallel within the target prediction time window. The cumulative evaluation results of the two links are then normalized and coupled to obtain a conflict constraint index. This conflict constraint index characterizes the degree of coexistence of both physiologically favorable moisture conditions and waterlogging-related risks within the same time window, thus providing a basis for subsequent constraint processing of early warning output parameters.

[0182] In this implementation, the system first forms a matrix water content sequence and an air humidity sequence within the target prediction time window based on multi-source environmental monitoring data and external prediction data. Specifically, the system determines the start time t1 and the end time tn of the target prediction time window, and discretizes the time window into n time steps according to a preset time step size Δt. The time step index is i, which increases from 1 to n, and ti represents the timestamp corresponding to the i-th time step, satisfying ti=t1+(i-1)·Δt.

[0183] The system reads the matrix moisture content value M(i) and the air humidity value H(i) at each time step ti. M(i) can be obtained by a matrix moisture content sensor, a dielectric moisture probe or a tensiometer, and H(i) can be obtained by the temperature and humidity sensor of a micro-weather station. When the external forecast data contains humidity-related corrections or rainfall forecast information, the system can use the forecast information as a compensation term to participate in interpolation alignment when generating the sequence, but without changing the attributes of M(i) and H(i) as basic evaluation inputs.

[0184] For missing time steps, the system can use adjacent time steps to preserve or linear interpolation to fill in the missing time steps, and record the missing time step as a marker to reduce the score contribution of that time step in subsequent evaluations.

[0185] In this implementation, the system acquires a table of growth threshold parameters for cauliflower and reads the moisture threshold W_wet, saturation threshold W_sat, and tolerance for receding water duration threshold D_tol from it. The threshold parameter table can be configured with entries by variety, cultivation season, and cultivation method. The system can load corresponding entries based on the crop identifier and management parameters of the target cultivation area. The moisture threshold is used to define the lower limit or interval boundary of "favorable moisture," the saturation threshold is used to define the trigger boundary of "waterlogging risk," and the tolerance for receding water duration threshold is used to define the tolerable upper limit of "saturation persistence and delayed receding water."

[0186] In this embodiment, the system performs a wetting advantage evaluation on the matrix water content state sequence and the air humidity sequence based on the threshold parameter table to obtain a wetting advantage score sequence, and performs time-series cumulative calculation on the score sequence to obtain the first sub-index.

[0187] For example, the system calculates a favorable humidity score S_pos(i) at each time step ti, where S_pos(i) reflects the degree to which M(i) and H(i) are satisfied relative to the humidity threshold. For ease of engineering implementation, S_pos(i) can be piecewise mapped: when M(i) is lower than W_wet, S_pos(i) is set to 0; when M(i) is not lower than W_wet and H(i) is not lower than the preset humidity lower limit H_wet, S_pos(i) is set to 1; when only one of them is satisfied, S_pos(i) is set to an intermediate value between 0 and 1 to reflect the case where "the humidity condition is partially satisfied". Subsequently, the system performs time-series cumulative calculation on S_pos(i) to form the first sub-index I_pos. For example, a weighted accumulation method can be used: I_pos=∑(i=1 to n) S_pos(i)·Δt, so that I_pos represents the effective duration of the favorable wet state within the target prediction time window. In order to suppress the rise caused by a single short-term satisfaction, S_pos(i) can also be smoothed before accumulation. For example, the moving average of S_pos(i) for k consecutive time steps can be taken and then included in the accumulation.

[0188] In this implementation, the system performs a waterlogging adverse assessment on the matrix moisture state sequence based on a threshold parameter table. The assessment process includes calculating the saturation maintenance duration and the decline duration, and obtaining a waterlogging adverse score sequence accordingly. Then, a time-series cumulative calculation is performed on the score sequence to obtain a second sub-index. Specifically, the system identifies saturation segments within the target prediction time window: when M(i) is not lower than W_sat, the time step is determined to be a saturated state; the system counts the time steps of continuous saturation states to obtain the saturation maintenance duration D_sat, which can be obtained by multiplying the number of continuous saturation time steps by Δt. The system further identifies the decline process after the saturation segment ends: taking the end time of the saturation segment as the decline start point, the system finds the moment in the subsequent time steps when M(i) is first not higher than W_wet as the decline end point, and the time difference between the two determines the decline duration D_rec; if the wetting threshold is not reached within the target prediction time window, the decline end point is temporarily set as tn, and tn - the decline start point is used as the current value of D_rec. Based on D_sat and D_rec, the system generates a waterlogging adverse score S_neg(i) at each time step. For example, the system can associate S_neg(i) with "saturation duration" and "recession lag": when in the saturation segment, S_neg(i) increases with the duration of saturation; when entering the recession stage and the recession is not yet complete, S_neg(i) increases with the duration of recession; and using the tolerance recession duration threshold D_tol as the segment boundary, when the combined amount of saturation duration and recession duration is close to or exceeds D_tol, the growth slope of S_neg(i) is increased to reflect the sensitivity of cauliflower to prolonged saturation and recession lag. Subsequently, the system performs time-series cumulative calculation on S_neg(i) to obtain the second sub-index I_neg. For example, I_neg=∑(i=1 to n) S_neg(i)·Δt is used to make I_neg characterize the cumulative intensity of adverse waterlogging conditions within the target prediction time window.

[0189] In this implementation, the system performs normalization on the first and second sub-indicators to obtain a first normalized value and a second normalized value, and multiplies the two to obtain the conflict constraint index. The purpose of normalization is to eliminate the influence of differences in scoring scales and time window lengths on the coupling results. For example, the system divides I_pos by the target prediction time window length T_win (T_win=(n-1)·Δt or n·Δt, consistently defined by the system) to obtain the first normalized value N_pos, and divides I_neg by the same time window length and the preset maximum unfavorable scoring upper limit S_neg_max to obtain the second normalized value N_neg, so that both N_pos and N_neg fall within the range of 0 to 1; when the value range of S_pos(i) and S_neg(i) is already limited to 0 to 1, the system can directly obtain N_pos and N_neg by dividing I_pos and I_neg by T_win respectively. Finally, the system multiplies N_pos and N_neg to obtain the conflict constraint index C_conf = N_pos·N_neg. This makes C_conf significantly increase when both the wetness advantage and the waterlogging disadvantage are high, and automatically decrease when one of them is low. This achieves a quantitative characterization of the coexistence of advantages and disadvantages, and provides a directly usable constraint quantity for subsequent conflict constraint processing.

[0190] Regarding the above S105:

[0191] In one embodiment, after obtaining the warning risk value after conflict constraint processing, the system enters the warning output stage. The system is pre-configured with preset warning conditions, which may be composed of a risk value threshold, a risk level threshold, a continuous satisfaction duration threshold, and suppression rules.

[0192] For example, the system compares the warning risk value at each time step. When the warning risk value is not less than the preset risk threshold and the cumulative duration of the condition is not less than the preset duration threshold, the system determines that the warning triggering condition is met. When the warning risk value exceeds the threshold but is at a short-term peak and has not reached the duration threshold, the system may not trigger the warning to reduce the prompting interference caused by transient fluctuations.

[0193] In this implementation, the system determines the risk type based on the dominant risk component or scoring source in the environmental risk evolution sequence. For example, if the risk component related to cumulative hypoxia stress contributes more, the risk type is determined to be waterlogging-hypoxia; if the scoring component related to insufficient moisture contributes more, the risk type is determined to be drought stress; when external forecast data indicates events such as high temperature or heavy rainfall and is consistent with the transition in the risk evolution sequence, the risk type can also be marked as high temperature superimposed or heavy rainfall superimposed, so as to prompt managers to pay attention to the triggers.

[0194] In this implementation, the system maps warning risk values ​​to risk levels. Risk levels can be determined using a segmented threshold mapping method, for example, mapping risk values ​​falling into different intervals to low, medium, high, or more granular levels respectively, and applying level corrections in conjunction with confidence levels or data validity markers: when data validity is low, the system can lower the risk level by one level or add a "data quality prompt" to the warning information to reduce unnecessary actions caused by false triggers.

[0195] In this implementation, the system generates risk time-series information to characterize the temporal location and persistence of the risk within the target prediction time window.

[0196] For example, the system extracts the time step from the environmental risk evolution sequence where the risk value first exceeds a preset risk threshold as the risk start time, and extracts the time step where the risk value falls back below the threshold as the risk end time, thus obtaining the risk duration. When the risk has not yet fallen back, the risk end time can be tentatively set as the end time of the target prediction time window. The system can also extract the peak time, peak risk value, and risk rise slope to help determine the urgency of the risk.

[0197] In this implementation, the system assembles and outputs early warning information. The early warning information includes at least the risk type, risk level, and risk timing information, and may include additional fields such as target cultivation area identifier, micro-area identifier, trigger time, and suggested handling parameters. The output method can be one or more of the following: local display terminal push, mobile terminal push, or background service push.

[0198] For example, the system sends alert messages to the management platform via MQTT topic publishing or an HTTPS interface, and records alert event entries in local storage. These event entries include a snapshot of the triggering conditions and key input data fragments for later tracing. To avoid duplicate alerts, the system can set a silent window for the same risk type in the same micro-area, updating only the risk level and risk timing information within the silent window without issuing new alert notifications repeatedly.

[0199] Example 2

[0200] Based on the same inventive concept, this embodiment provides a risk warning system for the growth environment of cauliflower that is a result of multi-source heterogeneous data fusion, corresponding to the risk warning method for the growth environment of cauliflower that is a result of multi-source heterogeneous data fusion. Since the principle of the system in this embodiment is similar to the risk warning method for the growth environment of cauliflower that is a result of multi-source heterogeneous data fusion described in Embodiment 1, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0201] Reference Figure 4 The multi-source heterogeneous data fusion-based early warning system for the growth environment risk of cauliflower provided in this embodiment includes:

[0202] The acquisition module 10 is used to acquire multi-source environmental monitoring data and external prediction data of the target cultivation area. The external prediction data is used to characterize the environmental change sequence within the target prediction time window.

[0203] The first processing module 20 is used to determine the environmental risk evolution sequence within the target prediction time window based on the multi-source environmental monitoring data and the external prediction data, and is used to characterize the change of environmental risk indicators over time within the target prediction time window.

[0204] Early warning module 30 is used to determine early warning output parameters based on the environmental risk evolution sequence, wherein the early warning output parameters include an early warning risk value and risk time sequence information corresponding to the early warning risk value;

[0205] The second processing module 40 is used to determine a first sub-indicator representing the favorable factors of humidity and a second sub-indicator representing the unfavorable factors of waterlogging based on the multi-source environmental monitoring data and the external prediction data, and to determine a conflict constraint index based on the first sub-indicator and the second sub-indicator; and to perform conflict constraint processing on the early warning output parameters based on the conflict constraint index.

[0206] The output module 50 is used to generate and output early warning information when the early warning risk value after conflict constraint processing meets the preset early warning conditions. The early warning information includes risk type, risk level and risk time sequence information after conflict constraint processing.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for early warning of environmental risks in *Cauliflower* growth based on multi-source heterogeneous data fusion, characterized in that... include: Acquire multi-source environmental monitoring data and external prediction data for the target cultivation area, wherein the external prediction data is used to characterize the environmental change sequence within the target prediction time window; Based on the multi-source environmental monitoring data and the external prediction data, an environmental risk evolution sequence within the target prediction time window is determined, which is used to characterize the changes of environmental risk indicators over time within the target prediction time window. The early warning output parameters are determined based on the environmental risk evolution sequence, and the early warning output parameters include the early warning risk value and the risk time series information corresponding to the early warning risk value; Conflict constraint indicators are determined based on the multi-source environmental monitoring data and the external prediction data; and conflict constraint processing is performed on the early warning output parameters based on the conflict constraint indicators. When the warning risk value after conflict constraint processing meets the preset warning conditions, a warning information is generated and output. The warning information includes the risk type, risk level, and risk time sequence information after conflict constraint processing.

2. The method for early warning of risks to the growth environment of *Caulis broccoli* based on multi-source heterogeneous data fusion according to claim 1, characterized in that, Determining the environmental risk evolution sequence within the target prediction time window includes: The multi-source environmental monitoring data and the external prediction data are processed according to the target prediction time window to obtain the target time series data. Based on the target time series data, environmental risk indicators are calculated within the target prediction time window at a preset time step to obtain the environmental risk evolution sequence.

3. The method for early warning of environmental risks in *Cauliflower* growth based on multi-source heterogeneous data fusion according to claim 2, characterized in that, The computational environment risk indicators include: The temperature prediction sequence is extracted from the external prediction data, and the oxygen consumption intensity coefficient sequence is determined based on the temperature prediction sequence through a preset temperature response function. The preset temperature response function is used to map the temperature prediction values ​​in the temperature prediction sequence to the oxygen consumption intensity coefficient. The parameters of the preset temperature response function are pre-stored in a temperature response parameter table. The temperature response parameter table records several temperature breakpoints and the coefficient values ​​corresponding to each temperature breakpoint. For the temperature prediction value at each time step, breakpoint location and interpolation calculation are performed based on the temperature breakpoints to obtain the corresponding oxygen consumption intensity coefficient. Based on the matrix environment data in the target time series data, the oxygen supply channel coefficient sequence is determined by a preset ventilation mapping function. The preset ventilation mapping function is used to map the matrix water content state included in the matrix environment data in the target time series data to oxygen supply channel coefficients. The parameters of the preset ventilation mapping function are pre-stored in a ventilation mapping parameter table. The ventilation mapping parameter table records the water content state breakpoints and the ventilation coefficient values ​​corresponding to each water content state breakpoint. For the matrix water content state at each time step, breakpoint location and interpolation calculation are performed based on the water content state breakpoints to obtain the corresponding oxygen supply channel coefficients. A cumulative hypoxia stress index sequence is determined based on the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence. The environmental risk indicators are determined based on the cumulative hypoxia stress index sequence, and the environmental risk evolution sequence is formed.

4. The method for early warning of risks to the growth environment of *Caulis broccoli* based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The criteria for determining conflict constraints include: Based on the multi-source environmental monitoring data and the external prediction data, determine the matrix water content sequence and air humidity sequence within the target prediction time window; Obtain a table of tree cauliflower growth threshold parameters, which includes a moisture threshold, a saturation threshold, and a threshold for tolerance to dehydration time. Based on the tree cauliflower growth threshold parameter table, the moisture content sequence of the substrate and the air humidity sequence are evaluated for moisture advantage to obtain a moisture advantage score sequence. Then, time-series cumulative calculation is performed on the moisture advantage score sequence to obtain the first sub-index, which is used to characterize the moisture advantage factors. Based on the tree cauliflower growth threshold parameter table, the waterlogging adverseness evaluation of the substrate water content state sequence is carried out. The waterlogging adverseness evaluation includes calculating the saturation maintenance time and the decline time and obtaining the waterlogging adverseness score sequence accordingly. The waterlogging adverseness score sequence is subjected to time-series cumulative calculation to obtain a second sub-index, which is used to characterize the waterlogging adverse factors. Normalization is performed on the first sub-index and the second sub-index to obtain a first normalized value and a second normalized value. The first normalized value and the second normalized value are then multiplied to obtain the conflict constraint index.

5. The method for early warning of risks to the growth environment of *Caulis broccoli* based on multi-source heterogeneous data fusion according to claim 4, characterized in that, Also includes: Acquire micro-topographic slope data of the target cultivation area, wherein the micro-topographic slope data is the environmental data of the target cultivation area and is used to determine the second sub-index; determine the baseline drainage curve based on the micro-topographic slope data; The actual decay curve of the matrix water content is fitted based on the target time series data, and the morphological deviation of the actual decay curve from the reference receding curve is calculated. The risk weight corresponding to the second sub-indicator is adjusted based on the morphological deviation, and the risk weight is used to weight the second sub-indicator when determining the conflict constraint indicator.

6. The method for early warning of risks to the growth environment of *Caulis broccoli* based on multi-source heterogeneous data fusion according to claim 3, characterized in that, The sequence for determining the cumulative hypoxia stress index includes: The hypoxia difference sequence is calculated based on the oxygen consumption intensity coefficient sequence and the oxygen supply channel coefficient sequence, wherein the hypoxia difference sequence is the difference between the oxygen consumption intensity coefficient and the oxygen supply channel coefficient at each time step. The cumulative hypoxia stress index sequence is obtained by performing a recursive cumulative calculation on the hypoxia difference sequence. The recursive cumulative calculation adds the cumulative hypoxia stress index of the previous time step to the cumulative increment obtained by weighting the hypoxia difference of the current time step according to the preset time step length at each time step, so as to obtain the cumulative hypoxia stress index of the current time step.

7. The method for early warning of risks to the growth environment of *Caulis broccoli* based on multi-source heterogeneous data fusion according to claim 5, characterized in that, The morphological deviation is the cumulative difference between the actual decay rate and the reference dewatering rate at each time step. Wherein, the actual decay rate is the rate of change of the water content state of the actual decay curve at the corresponding time step, and the reference water loss rate is the rate of change of the water content state of the reference water loss curve at the corresponding time step. The adjustment includes: when the morphological deviation exceeds a preset deviation threshold, setting the risk weight corresponding to the second sub-indicator to a first weight value; and when the morphological deviation does not exceed the preset deviation threshold, setting the risk weight corresponding to the second sub-indicator to a second weight value.

8. The method for early warning of environmental risks in *Cauliflower* growth based on multi-source heterogeneous data fusion according to claim 5, characterized in that, The determination of the baseline recession curve based on the micro-topographic slope data includes: Obtain matrix permeability parameters and a table of tree cauliflower growth threshold parameters that characterize matrix permeability and porosity. The matrix permeability parameters include at least one of permeability coefficient and porosity, and the tree cauliflower growth threshold parameter table includes saturation threshold, wetting threshold, and tolerance to dehydration time threshold. A reference drainage curve is calculated based on the micro-topographic slope data and the matrix permeability parameters. The reference drainage curve characterizes the drainage process in which the matrix water content drops from the saturation threshold to the wetting threshold. The reference drainage curve is subjected to threshold constraint processing based on the tree cauliflower growth threshold parameter table to obtain the baseline drainage curve.

9. The method for early warning of environmental risks in *Cauliflower* growth based on multi-source heterogeneous data fusion according to claim 8, characterized in that, The threshold constraint processing includes: A reference fallback time is determined based on the reference fallback curve, wherein the reference fallback time is the time it takes for the matrix water content, as represented by the reference fallback curve, to fall from the saturation threshold to the wetting threshold. The weighting coefficient is determined based on the micro-topographic slope data, and a supplementary weighting coefficient is determined. The supplementary weighting coefficient is 1 minus the weighting coefficient. The weighting coefficient takes a value between 0 and 1 and increases as the micro-topographic slope increases. The reference fallback duration is multiplied by the weighting coefficient to obtain the first weighting term, the tolerance fallback duration threshold is multiplied by the supplementary weighting coefficient to obtain the second weighting term, and the first weighting term and the second weighting term are added to obtain the weighted fallback duration. The reference recession curve is reconstructed based on the weighted recession duration to obtain the baseline recession curve.

10. A risk early warning system for the growth environment of *Cauliflower* based on multi-source heterogeneous data fusion, characterized in that: include: The acquisition module is used to acquire multi-source environmental monitoring data and external prediction data of the target cultivation area. The external prediction data is used to characterize the environmental change sequence within the target prediction time window. The first processing module is used to determine the environmental risk evolution sequence within the target prediction time window based on the multi-source environmental monitoring data and the external prediction data, and is used to characterize the change of environmental risk indicators over time within the target prediction time window. The early warning module is used to determine early warning output parameters based on the environmental risk evolution sequence. The early warning output parameters include an early warning risk value and risk time sequence information corresponding to the early warning risk value. The second processing module is used to determine conflict constraint indicators based on the multi-source environmental monitoring data and the external prediction data; and to perform conflict constraint processing on the early warning output parameters based on the conflict constraint indicators. The output module is used to generate and output early warning information when the early warning risk value after conflict constraint processing meets the preset early warning conditions. The early warning information includes risk type, risk level, and risk time sequence information after conflict constraint processing.

Images