A decision support system and method for agriculture

Field The present disclosure relates to a method and system for providing a decision support system and method for agriculture. More specifically, the decision support system and method provide an accurate nearcasting of hyperlocal weather and/or pests outbreaks for decision support for agricultural activities. Background

Farmers need accurate real-time and future weather forecasts at their farm location to make decisions at various stages in the growing season. Various systems exist to enable location-specific forecasts using interpolated data between stations. However, these differ between each other with regard to weather parameters and often do not always provide the accuracy required to make a decision to undertake an agricultural activity.

Further, farmers need higher resolution, location-specific information than what is available at present. Thus, decisions at any given point in time may require precise quantitative information, for example, the amount of expected rainfall within the next 6 hours or 12 hours or throughout longer periods up to a few days. A system which can provide such insight will be very valuable for making decisions to undertake an agricultural activity. Whilst there are technologies in existence that use physical sensors to measure data (e.g. Temperature Sensors, Rainfall Gauges), along with forecasting models to predict meteorological values (E.g. AROME, GFS), applying these technologies on a hyper-local, personalized farm basis is a critical problem in the field of precision agriculture. The integration and analysis of rainfall radar is common technology utilized in measurement of rainfall where sensors do not lie, however, mis-readings can (and do) occur due to factors like temporal delay (even 5 minutes can see a substantial change in rainfall volume, especially in a particularly in an unstable thunderstorm), variable wind direction, wind speed, cloud height. Further, substantial differences in actual rainfall amounts vs. modelled data even in highly short term (1 hour) temporal ranges occur. Similarly, there can be significant variances in meteorological phenomena (e.g. rainfall) in relatively short distances (<10km). These variances or errors in the prediction of weather concerning a farm is critical for economic feasibility or profitability of the farm. These problems are exacerbated by climate change, an increasing degree of volatility in weather patterns along with increased rainfall amounts and intensity in temperate climates and the challenges posed by big variations in precipitation over relatively short distances (5-10km). These changes present increasing challenges to arable agriculture and the time dependency of yield and other factors on narrow temporal windows.

Additionally, there is a need for a system or service which provides risk management and decision support in relation to weather events and agricultural operations and provides this in a context-aware and hyper-local manner. Such a system tailored to the specific needs of each field and farm and providing a tailored service within which the farmer can evaluate and manage weather- related risks more effectively would be very advantageous to farmers, agronomists and agricultural enterprises.

Certain commercial vendors do currently supply hyper-local weather forecasts for farmers. However such forecasts based as they are on interpolated data for a given GPS are not accurate and hence not adopted by the farming community. For example, US2017/0061050 A1 discloses a method and system for applying real-time field level weather simulation and prediction to one or more models to develop harvest advisory outputs. WO2015/017676 discloses a system which tracks weather and how it will impact organisations or individuals. US2019/0254242 discloses a system which receives wide-area meteorological data and local area sensor data.

Therefore, there is an unresolved need for a method and system for providing improved and accurate hyper-local weather prediction, where the prediction is for a short temporal range (1 -12 hours) and also for additional temporal ranges which may be dynamic in nature and length (e.g. a large zone of high-pressure, stable weather will be less subject to weather variance compared to that of a deep low- pressure depression.). Further, there is a strong dependency on the reliability of the measurements recorded by the sensors (e.g. reading frequency, uptime), accuracy tolerance, accuracy drift and susceptibility to errors (i.e. a leaf blocking a rain gauge). Whilst many modern sensors ensure tight tolerances in terms of their respective measurement accuracies, these sensors are merely static data collection entities For precision agriculture, accurate measurements of temperature and rainfall are the two most impactful weather variables, as both are essential in the growth of crops i.e. even a 0.5 degree Celsius offset from the ‘true’ value can cause a significant change in growth stage prediction over extended temporal windows. Therefore, it is essential to detect sensor failures or device malfunctions and filter out potentially erroneous data from said failed sensors or malfunctioning devices. Thus, there is a need for a filtering mechanism which filters out potentially erroneous data to maintain the prediction accuracy. While weather is fundamental to farm activity it is but one key information driver. It is also desirable to provide mechanisms for consideration and combining a number of pertinent knowledge sources. Examples of such sources include farm topography, farm soil characteristics, farmer fiscal circumstance and current commodity prices/trends. None of the prior arts considers farm topography and the soil characteristics for providing accurate advisories to farmers which are tailored to their specific farms and fields as is provided herein.

Therefore, there is a long-felt and unresolved need for a precision decision support system which provides advisories to farmers to initiate or avoid an agricultural activity. Thereby, preventing losses and maximizing yields for the farmers. Summary

The present invention, as set out in the appended claims, relates to a method and system for providing a decision support system and method for agriculture. More specifically, the decision support system and method provide an accurate nearcasting of hyperlocal weather and/or pests outbreaks for decision support for agricultural activities.

The method comprises discovering and receiving meteorological sensor data from a plurality of weather stations located in and around a geographical area. The confidence parameter for each measurement in said meteorological sensor data is determined. This step is carried out to ascertain the degree of accuracy of the received data and/or to filter out corrupted data or data received from faulty weather stations. In an embodiment, the corresponding confidence parameters are stored in a weather database, such that a weather prediction model may leverage the confidence parameters for making an accurate prediction.

The invention relies on discovering and effectively using large amounts of external data be it sensor data or model data. The effective conflation and adjudication of weather data and other agronomic data sets of a historic or real time nature so as to derive an output which is the most accurate in every context is a challenging technical problem. Preferably the system discovers relevant data streams and conflates such data obtained in real-time with historic data and calculates a confidence coefficient. The system adjudicates between different data sources so as to select those sensors or parameters which it has found to be most accurate in the current context. For example, the system can calculate a “ground impact factor” for a specific farm or area taking into account its soil and topography and history and finally it operates in a ubiquitous, hyperlocal data conflation process. In other words, the system and method access a much wider range of continuously revised hyperlocal agronomic data for this specific farm or field and personalise the output of its nearcast in ways that would not otherwise be possible.

The confidence parameter is determined based on one or more of the following: divergence between historical weather parameters for said geographical area and measured weather parameters received from a weather station located in said geographical area on the same day of the year and time of day in the past years; abrupt or anomalous change in received measured weather parameters from said weather station within a short-term predetermined temporal window for said geographical location; divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location, where the weather parameters measured in geographical locations adjoining said geographical location are adjusted for change in geological factor and where the divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location is weighted based on the distance between said geographical location and said adjoining geographical locations; divergence between: difference of weather parameters measured in said geographical location and weather parameters predicted in said geographical location; and difference of weather parameters measured in geographical locations adjoining said geographical location and predicted weather parameters in geographical locations adjoining said geographical location; and, statistical occurrence of previous inaccuracies in measured weather parameters received from said weather station. Further, a plurality of weather predictions is generated for said geographical area, using a plurality of weather prediction models, where said weather prediction models predict the weather based on present and past received meteorological sensor data and the corresponding confidence parameters associated with each measurement in said received present and past meteorological sensor data. In other words, the weather prediction models are used to predict weather parameters in the near future for said geographical area using the temporally stored meteorological sensor data in the weather database. In one embodiment the step of determining the confidence parameter further comprises discovering and receiving data representative of soil, topography and/or yield data and determining the confidence parameter based on said soil and/or yield data. The soil and/or yield data is obtained from one or more soil sensors configured to provide at least one of soil temperature, soil moisture, hydro-conductivity, texture, pH or nutrient status or micronutrient status.

Further, predicted weather data comprising weather predictions for said geographical area is received from one or more weather prediction servers. Thereafter the weather database cleans and temporally stores said meteorological sensor data along with corresponding confidence parameters and said predicted weather data.

Thereafter, performance of each of weather predictions, received from weather prediction servers and generated using said weather prediction models for said geographical area, is continuously evaluated. The evaluation is performed based on a deviation of each of predicted weather parameters of each of weather predictions with the meteorological sensor data.

Thereafter, each weather parameter for said geographical location is predicted based on said continuous evaluation by: providing a predicted weather parameter of a weather prediction selected from a plurality of weather predictions received from weather prediction servers or generated using said weather prediction models for said geographical area, where the selection of the weather prediction has a minimum deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter.

In an embodiment, a weighted predicted weather parameter is provided from ‘n’ weather predictions. Said ‘n’ weather predictions are selected from a plurality of weather predictions received from weather prediction servers and generated using said weather prediction models for said geographical area, where the selection of ‘n’ weather predictions are ranked in ascending order by their deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter, where the weight assigned to each weather prediction is based on said rank and where said selected ‘n’ weather predictions have similar evaluated performance. In a preferred embodiment, the number ‘n’ is two. In a preferred embodiment, said geographical area has a radius of less than 5 kilometers. Further, the prediction of weather parameters comprises of prediction of weather parameters for said geographical area in the following 1 hour to 6 hours. Further, the predicted weather parameters comprise of rainfall, temperature, relative humidity and average wind speed.

In a preferred embodiment, continuous adjudication between different weather models is provided by evaluating predictions and ground truth from said different weather providers. The application chooses between those parameters from different models which are most likely to be correct and provides a fused forecast wherein the most accurate parameter from each provider is used. For example, weather service “A” may have excellent wind prediction, weather service “B” may have excellent rainfall prediction while weather service “C” has both poor wind and rainfall prediction. This system is continuously validated and changes dynamically based on inputs of new data. This feature obviates the need for farmers to use more than one weather source outside of the application as is often the case, to inform themselves for decisions. The use of this approach also provides a more general non-local weather model which is based on “best in class” consensus data and may be applied elsewhere for non-hyper-local applications in agriculture-related meteorology. For example its use in highly accurate three-day regional forecasts for the farming community In the UK one may envisage forecasts for the South, Eastern Region, Midlands, Wales, the North and Scotland.

In a further embodiment, the system conflates and engages in continuous adjudication of ubiquitous data streams within a hyper-local context. Within a digital agricultural platform, there are provided access to a range of additional data from multiple sources both from its own database and also through a range of microservices which call external data providers. Such data may comprise, soil data, topographical data (slope, aspect, elevation), cropping data, data on fertiliser and fungicide or other applications, yield data and satellite imagery or radar. Thus the weather model herein and its recommendations linked to such a system and its data sources may be validated and made more field context- specific through this data conflation, Alternatively, on a stand-alone basis i.e. without comprising a component of a digital agriculture platform, the system is provided with microservices which call on certain additional data sets selected from the above.

The ubiquitous hyper-local data may be utilised to create a “ ground impact factor” wherein the specific history (e.g. history of flooding or poor drainage in certain areas), geographical location, farm topography, soil type and texture, soil hydro- conductivity, soil moisture and temperature levels are used to generate a ground impact factor parameter which predicts the severity of any weather event on a specific farm. This may result on occasion, in different recommendations for farms within a local area depending on their specific characteristics. Thus, the application provides a valuable, context-aware and hyper-local nearcasts for each field or farm. If the system is deployed in the form of a distributed architecture (e.g. an edge or fog node type deployment), each farm instance and its database evolves separately in time; local ones in a similar manner and distant ones with different soils and regional climates in a significantly different way. Such data is held both on the database of the local instance and also transmitted to a central cloud-based server for more detailed analytics. Accordingly, each farmer owns a unique resource; a personalised system adapted for their farm and a growing and unique record of how weather events have impacted crop performance on that farm.

In another embodiment, the system comprises a customisable context-aware risk management tool for use by farmers. This feature comprises a graphical user interface showing the risk levels of rain or other events over several days along with a confidence level. Hence this aspect of the invention allows the farmer to select which temporal window and which risk level to accept for the performance of an agricultural operation depending on other variables such as convenience, availability of labour, plant, availability of agri-inputs for spraying or any other factors.

The system provides a variety of configurable user inputs including; risk level, the context of the operation (drilling, application of fertiliser, applications of fungicide at time windows TO, T1, T2, T3, other anti-pest applications, harvesting), observations and user feedback on satisfaction with previous forecasts and recommendations.

It will be appreciated that the invention herein is not confined to the continuous discovery, conflation and adjudication of data for the provision of nearcasting or meteorological purposes alone but may also be extended as described herein to other agronomic data types, streams and models for other types of decision support in agronomy.

For example, in a further embodiment, the system provides methods for the management of disease and pest risk to combinable crops or other non- combinable crops such as fruit. Outbreaks of Septoria blight on wheat are particularly dependent on rainfall. Outbreaks of insect pests are also dependent on season and weather conditions. There exists already a range of disease prediction models for Septoria and other diseases and insect infestations that utilise climatic factors and parameters such as degree days, The invention herein provides a novel method to adjudicate between such models and also to conflate them with real-time sensor data, real-time forecast data and human observations derived from the weather sensors and models described above.

Such an application works as follows;

Continuously discover and access several captive pest or disease models and/or specialised disease detecting sensors, pest and disease services provided by external parties in each country or region, drone imagery using RGB or hyperspectral cameras, proximal RGB cameras connected to machine learning to which the system has access via microservices and human observations and also database stored labelled images which represent machine vision-based diagnosis of plant disease not visible or not yet visible to the human eye.

Derive and apply confidence coefficients to the data sources as per above using current and historical data and deviations as described above.

Adjudicate between models, historic data and sensor data o optimise the disease/pest prediction and risk level continuously.

Provide recommendation on a personalised hyperlocal basis to the grower for actions to mitigate threats.

In addition to the data sources and models above, the application also provides methods for the effective use of socially derived information on pests. For example, a user may be notified that another user, say 5km away and currently upwind, has just reported an outbreak of fungal disease or pest which will be an important input to decision making. Twitter feeds may be analysed here using semantic entity recognition for automation of analysis of tweets or a group of local instances may be facilitated to communicate with each other in this regard. In addition to the above embodiment for diseases and pests, additional embodiments are envisaged by this extensible, framework and data stack as follows; topographical data from a plurality of sources may be conflated and used to predict soil moisture which in turn forms an input into decision making around the amount and timing of fungicide or pesticide applications or decisions involving the use of heavy agricultural machinery.

Additionally, the invention herein may be employed on other data such as soil and yield data and its conflation. For soil, the data may comprise access to multiple online sources of data per GPS location and/or the discovery and subsequent conflation of a plurality of soil sensors which may include soil temperature, soil moisture, hydro-conductivity, texture, pH or nutrient status or micronutrient status. Also, the use of remote sensing may also provide an additional layer in which soil parameters can be sensed, acquired and visualised with the system.

Thus, the invention herein provides as well as a meteorological application, an intelligent, extensible data stack and methods for discovering agronomically relevant data, conflating, adjudicating, making recommendations and creating context-aware, hyperlocal personalisation of such data sets and recommendations for action.

The decision support system for agriculture, comprises at least a processor operatively coupled to a memory, and a transceiver, a scheduler and a dynamic query system wherein the scheduler collects data on a defined temporal basis and the dynamic query system continuously seeks the latest sensor and meteorological model data, one or more weather prediction servers, one or more weather stations operatively coupled to said processor via said transceiver; said memory storing a plurality of weather prediction models and computer-readable instructions to cause the processor to: a) receive meteorological sensor data from said plurality of weather stations located in and around a geographical area; determine the confidence parameter for each measurement in said meteorological sensor data, wherein the confidence parameter is determined based on one or more of: divergence between historical weather parameters for said geographical area and measured weather parameters received from a weather station located in said geographical area on the same day of the year and time of day in the past 10 years; abrupt or anomalous deviation in received measured weather parameters from said weather station within a predetermined short -term temporal window for said geographical location; divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location, where the weather parameters measured in geographical locations adjoining said geographical location are adjusted for change in geological factor and where the divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location is weighted based on the distance between said geographical location and said adjoining geographical locations; divergence between: difference of weather parameters measured in said geographical location and weather parameters predicted in said geographical location; and difference of weather parameters measured in geographical locations adjoining said geographical location and predicted weather parameters in geographical locations adjoining said geographical location; and, IB statistical occurrence of previous inaccuracies in measured weather parameters received from said weather station; b) receive predicted weather data comprising weather predictions for said geographical area from said one or more weather prediction servers; c) store said meteorological sensor data along with corresponding confidence parameters and said predicted weather data in a temporal database stored in said memory; d) generate a plurality of weather predictions for said geographical area using a plurality of weather prediction models, where said weather prediction models predict the weather based on present and past received meteorological sensor data and the corresponding confidence parameters associated with each measurement in said received present and past meteorological sensor data; e) continuously evaluate performance of each of weather predictions received in step a) and generated in step d) for said geographical area based on a deviation of each of predicted weather parameters of each of weather predictions with the meteorological sensor data; f) based on said performance evaluation, predict each weather parameter for said geographical location by: providing a predicted weather parameter of a weather prediction selected from a plurality of weather predictions received in a) or generated in d), where the selection of the weather prediction has a minimum deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter; or providing a weighted predicted weather parameter from ‘n’ weather predictions selected from a plurality of weather predictions received in step a) or generated in step d), where the selection of ‘n’ weather predictions are ranked in ascending order by their deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter, where the weight assigned to each weather prediction is based on said rank and where said selected ‘n’ weather predictions have similar evaluated performance. In a preferred embodiment the value of ‘n’ is two. g) generate an advisory for one or more agricultural activities based on said prediction of weather parameters, where said generation of the advisory for one or more agricultural activities is further based on ground impact factor, wherein said factor being derived from a model comprising information on soil moisture, flooding risk, soil hydro-connectivity and texture and topology of said geographical area. Said one or more agricultural activities comprises drilling or equivalent seeding, fertilizing, applying fungicides and pesticides or other crop applications and harvesting.

The system can adjudicate between different weather models and their predicted parameters by obtaining and analysing data on predictions and ground truth for those parameters. The system can be configured to choose between those parameters from different models which are most likely to be correct and provide a fused forecast wherein the most accurate parameter from each provider is used.

In one embodiment the system constantly adjudicates and revises scores based on the input of a plurality of real-time data and forecasts along with changing seasonal contexts. For example, one data source may be optimum for inclusion in the model in Winter and other in Summer.

In one embodiment the system constantly adjudicates and revises scores based on the recorded accuracy at the local level. For example, one data source may be optimum for inclusion in the model for one farm in one geographic area of the country while another is optimum elsewhere or even for nearby farms with radically different elevations, slopes, aspects.

In one embodiment the system conflates and adjudicates continuously streams of data within a data-rich hyper-local context. Such data comprises not only weather-related sensor data but also modelled data and other forms of data streams including vehicle telemetry, PV devices, loT devices and other types of data accessed via microservice calls such as soil and satellite imagery.

In a preferred embodiment, said geographical area has a radius of less than 5 kilometres. Further, the prediction of weather parameters comprises of prediction of weather parameters for said geographical area in the following 1 hour to 6 hours for nearcasting and in temporal windows up to three days for a more extended view that may be used for risk management as described herein. Further, the predicted weather parameters comprise of rainfall, temperature, relative humidity and average wind speed.

In a preferred embodiment the system is designed and constructed by employing the widely used enterprise software approach of microservices. This comprises an ensemble of coupled modules communicating via APIs to provide a range of advantages including flexibility regarding technology stack, extensibility, scalability and better fault isolation as well as speed of build and deployment as well as the ability to migrate microservices independently in different manners. In a related preferred embodiment, the microservices comprising the system are deployed within containers (e.g. Docker).

Available data sources types and microservice categories include: (1 ) proximal sensor derived, (2) remote sensor derived, (3) model derived, (4) human and social derived, (4) complex hyper-local services. The proximal sensor derived data may comprise sources such as data from proprietary weather stations owned by the provided of the Aura service, EPA rainfall telemetry, marine-derived data, that from a plurality of loT and PV device, and car telemetry such as activation of windscreen wipers. The remote sensor derived data may include data from DataPoint satellite cloud cover and rainfall radar, earth observation data from Sentinel 1 or 2, or data from commercial earth observation services. The model derived data may include data from national weather forecasting services and also from commercial providers supplying interpolated hyper-local forecasts or nearcasts.

Human and social derived data comprises a variety of sources as follows. Human derived data may be obtained by input from users (growers) of the system through observations and comments on the impact of weather events and is stored in the database for inclusion in future refinements of the working of the system such data may also be obtained by Citizen Science programmes from certain members of the general public or from co-operation among farmers in the same locality who are using the application. Social derived data may be obtained from Twitter and other social media technologies. Twitter feeds may be analysed using semantic entity recognition for weather-related events and incorporated automatically into the application database, Confidence coefficients may also be derived for Twitter feed sourced data and its components.

In the context of the present invention complex hyper-local services refers to the recursive utilisation of the other sources (1-4) above to provide hyper-local and context-aware services for each farmer. In part, this may be derived from integration with other data sources, analytics and models within an advanced digital agriculture platform.

Thereby, the precision decision support method and system provides precise advice to farmers to initiate or avoid an agricultural activity. Thereby, preventing losses and maximizing profit for the farmers.

Brief Description of the Drawings

The invention will be more clearly understood from the following description of an embodiment thereof, given by way of example only, with reference to the accompanying drawings, in which:- FIG. 1 exemplarily illustrates a high-level flow diagram of the agricultural decision support method in accordance with some of the embodiments of the present invention; FIG. 2 exemplarily illustrates the components of the agricultural decision support system in accordance with some of the embodiments of the present invention; and

FIG. 3 exemplarily illustrates the architecture of the various components of the agricultural decision support system in accordance with some of the embodiments of the present invention.

Detailed Description of the Drawings

The present invention relates to a method and system for providing a decision support system and method for agriculture. More specifically, the decision support system and method provides an accurate prediction for decision support for agricultural activities.

FIG. 1 exemplarily illustrates a high level flow diagram of the agricultural decision support method in accordance with some of the embodiments of the present invention. The method comprises receiving meteorological sensor data 102 from a plurality of weather stations located in and around a geographical area. The meteorological sensor data may also include present and historic meteorological sensor data derived from publicly available and private sources, comprising airport telemetry, ship telemetry, weather buoys, oil rigs, rainfall gauges, flood meter readings, rainfall radar analysis, satellite imagery analysis and distributed weather station network. Further, predicted weather data comprising weather predictions for said geographical area is received from one or more weather prediction data sources 102. In addition the method and system is adapted to discover and receive data representative of soil, topography and/or yield data. The soil and/or yield data is obtained from one or more soil sensors configured to provide at least one of soil temperature, soil moisture, hydro-conductivity, texture, pH or nutrient status or micronutrient status. Further, a confidence parameter 104 for each measurement in said received meteorological sensor data is determined. This step is carried out to ascertain the degree of accuracy of the received meteorological sensor data and/or to filter out corrupted meteorological sensor data or data received from faulty weather stations. In an embodiment, the corresponding confidence parameters are stored in a temporal weather database 103, such that weather prediction models 105 may leverage the confidence parameters for making an accurate prediction. In other words, lower confidence on possibly inaccurate or corrupt meteorological sensor data 102 may be filtered out or assigned a lower weight while predicting weather parameters in the near future by the weather prediction models 105. Thereby, providing highly accurate predictions for generating agricultural activity advisory 106. The confidence parameter further comprises utilising discovered and received data representative of soil, topography and/or yield data and the confidence parameter can be determined based on said soil and/or yield data.

The confidence parameter is determined 104 based on one or more of the following:

1. Divergence between historical weather parameters for said geographical area and measured weather parameters received from a weather station located in said geographical area on the same day of the year and time of day in the past years. For example, a weather station may report rainfall of about 2mm of rainfall for the given geographical location for a particular day, however, the same day in previous years may have recorded an average of 200mm rainfall. Thus such a reading is suspected to be inaccurate or due to a faulty or a blocked sensor. In other words, long term reading assessment (at a given latitude/longitude) is computed e.g. for the same date and time for the last 10 years historical temperature data is analysed to obtain a long term average of 16.7 degrees Celsius. In view of the long term average, if received meteorological sensor data (temperature) is significantly lower or higher than said long term average then said received meteorological sensor data (temperature) may be flagged to be inaccurate or labelled with a low confidence.

2. Abrupt change in received measured weather parameters from said weather station within a predetermined temporal window for said geographical location.

That is, short term reading assessment (at a given geographical area latitude/longitude) is carried out, where the most recent readings are analysed, e.g. the meteorological sensor data for a given geographical area for the past 4 hours may read 13.0, 12.7, 12.6, 12.0 degrees Celsius respectively which seems to be a consistent decline in temperature which is expected. However, in a similar instance, if the readings would have read 13.0, 12.7, -5.6, 12.0 respectively, then such meteorological sensor data may be flagged as inaccurate or tagged with a low confidence due to an abrupt variation in the short term (-5.6 Celsius) which is not expected.

3. Divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location, where the weather parameters measured in geographical locations adjoining said geographical location are adjusted for change in geological factor and where the divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location is weighted based on the distance between said geographical location and said adjoining geographical locations. For example, a weather station at a geographical location reports a temperature of 16 degree Celsius and the neighbouring weather stations report a temperature of 16.1, 15.9, 16 degree

Celsius respectively, then the reading of the weather station is in agreement with the readings of the neighbouring weather stations, hence such a reading may be tagged with high confidence. On the contrary, if the weather station reports a temperature of 19 degree Celsius then it is highly likely that the weather station may be reporting an inaccurate temperature reading. However, a person skilled in the art would appreciate that temperature (weather parameters) may vary with altitude and distance from the weather station. Hence, geological factors such as altitude and the distance from the weather station is factored while determining whether the temperature readings from the weather station agrees with the neighbouring weather stations. For example, a weather station may be located in a valley which reads 19 degree Celsius and a neighbouring weather station located on a mountain top may read 16 degree Celsius. In such a case, by factoring in the difference in height (geological factor) of the temperature recorded by the weather station may be considered to agree with the temperature of the neighbouring weather station. In another example, if a weather station ‘A’ reports 12 degree Celsius and two neighbouring stations ‘B’ and Ό’ report temperatures of 12.5 and 13 degree Celsius respectively, then the distance between weather station A and weather stations B and C respectively may be considered while computing the confidence of the temperature measured by weather station A. Accordingly, the variance in temperature between weather stations farther away may be assigned lesser weight than the variance in temperature between weather stations closer together.

In an embodiment, both the geological factors and the distance factor may be considered for computing the confidence coefficient for a reading of a weather parameter received from a weather station when compared with neighbouring weather stations.

4. The divergence between the difference of weather parameters measured in said geographical location and weather parameters predicted in said geographical location. For example, the predicted temperature for a geographical location is 20 degree Celsius and the received temperature reading is 24 degree Celsius, which is well beyond the prediction accuracy, then such a reading may be considered to be inaccurate. In another embodiment, the difference of weather parameters measured in geographical locations adjoining said geographical location and predicted weather parameters in geographical locations adjoining said geographical location may be considered to evaluate the confidence coefficient. 5. Statistical occurrence of previous inaccuracies in measured weather parameters received from said weather station.

A person skilled in the art would appreciate that one of the above factors or a combination of the above factors may be utilized to compute the confidence coefficient of weather parameters received from meteorological sensor data sources.

Thereafter, the weather database 103 temporally stores said meteorological sensor data along with corresponding confidence parameters and said predicted weather data.

Further, a plurality of weather predictions is generated for said geographical area, using a plurality of weather prediction models 105, where said weather prediction models predict the weather based on present and past received meteorological sensor data 102 and the corresponding confidence parameters associated with each measurement in said received present and past meteorological sensor data. In other words, the weather prediction models 105 are used to predict weather parameters in the near future for said geographical area using the temporally stored meteorological sensor data in the weather database 103.

Thereafter, performance of each of weather predictions, received from weather prediction data sources 101 and generated using said weather prediction models 105 for said geographical area, is continuously evaluated. The evaluation is performed based on a deviation of each of predicted weather parameters of each of weather predictions with the meteorological sensor data 102.

Thereafter, each weather parameter for said geographical location is predicted based on said continuous evaluation by: providing a predicted weather parameter of a weather prediction selected from a plurality of weather predictions received from weather prediction servers or generated using said weather prediction models for said geographical area, where the selection of the weather prediction has a minimum deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter. For example, weather prediction model ‘X’ predicts a weather parameter (e.g. rainfall) with least deviation among the prediction models, and where other prediction models have significantly larger deviation than model X”, then model ‘X’ is selected for prediction for said weather parameter. Similarly, a prediction model or a weather prediction data source is selected for each of the weather parameters.

In an embodiment, a weighted predicted weather parameter is provided from ‘n’ weather predictions. Said ‘n’ weather predictions are selected from a plurality of weather predictions received from weather prediction servers and generated using said weather prediction models for said geographical area, where the selection of ‘n’ weather predictions are ranked in ascending order by their deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter, where the weight assigned to each weather prediction is based on said rank and where said selected ‘n’ weather predictions have similar evaluated performance. In a preferred embodiment the number ‘n’ is two. For example weather prediction model ‘X’ and Ύ predicts a weather parameter (e.g. rainfall) with least deviation among the prediction models, and where other prediction models have significantly larger deviation than models ‘X” and Ύ, then models ‘X’ and Ύ are selected for prediction for said weather parameter. In such a case, a weighted prediction is provided, where the model ‘X’ having a lesser degree of deviation with respect to model Ύ’, then model ‘X’ is weighted proportionately more than the prediction provided by model Ύ. Similarly, prediction models or a weather prediction data sources are selected for each of the weather parameters.

In a preferred embodiment, said geographical area has a radius of less than 5 kilometres. Further, the prediction of weather parameters comprises of prediction of weather parameters for said geographical area in the following 1 hour to 6 hours. Further, the predicted weather parameters comprise of rainfall, temperature, relative humidity, atmospheric pressure, solar radiation, and average wind speed, wind direction. Finally, an advisory is generated for one or more agricultural activities based on said prediction of said weather parameters. In an embodiment the advisory for one or more agricultural activities is/are further based on ground impact factor, wherein said factor is derived from a model comprising information on soil moisture, flooding risk, soil hydro-connectivity and texture and topology of said geographical area. In an embodiment said one or more agricultural activities comprises spraying, seeding, fertilizing, and harvesting.

The agricultural activity may be one or more of spraying, seeding, fertilizing, and harvesting. For example, the decision support method provides advisories which suggest whether the weather is conducive for planting a crop in the near future or should a farmer avoid planting a crop in the near future.

In another example, the decision support method may indicate optimal spraying times for fungicide, herbicide, fertilizer and any other applications where the near- term weather is important. The decision support method may provide an advisory as to when and where to harvest i.e. based on grain dry down models a prediction may be made as to which fields of arable crops have optimum moisture for harvest. Such models are used to make decisions as to which farms and fields to harvest as there may be losses or cost incurred if there is too much grain moisture which would require drying, or in case of too little moisture there may be a loss in grain weight. Thus, the present invention provides decision support to a farmer before the harvesters are assigned for harvesting the crop.

A person skilled in the art would appreciate that the present invention takes the ground impact factor into consideration while generating advisories. For example, soil type, depth (and many other potential variables like; field slope, elevation, exposure, etc.) to assess potential impact a weather phenomenon may have (e.g., by predicting soil hydro-conductivity, potential rainfall impact with reference to; soil moisture, flooding risk, field accessibility) is used. For example, rice paddy requires a flooded parcel of land for planting and whether a particular amount of rainfall would flood the land depends on the soil characteristics and the lay of the land, therefore the present invention may advise cultivation where there is 20 millimeters of rainfall if the soil constitution is clay based (thus preventing water to percolate easily) and may not advise cultivation if the soil constitution is sandy or porous.

In other words, the method comprises prediction of weather parameters such as rainfall, temperature, relative humidity and average wind speed for a geographical area having a radius of less than 25 kilometres and preferably less than 5 kilometres. The weather parameters are predicted using weather prediction models which accepts one or more of historical weather data of said geographical location, and recent actual weather parameters received from weather stations and/or weather data from various databases. Further, forecast data from various weather forecast systems may also be directly utilized. FIG. 2 exemplarily illustrates the components of the agricultural decision support system in accordance with some of the embodiments of the present invention. The agricultural decision support system comprises a decision support server 201 , at least one weather prediction server 202, at least one weather station 203, and at least one client device 204. The decision support server 201 comprises a processor 201a operatively coupled to a memory 201b and a transceiver 201c. The client device 204 comprises a processor 204a operatively coupled to a memory 204b, a display 204c and a transceiver 204d. The decision support server 201 is operatively coupled to said at least one weather prediction server 202 and at least one weather station 203.

FIG. 3 exemplarily illustrates the architecture of the various components of the agricultural decision support system in accordance with some of the embodiments of the present invention. As illustrated the geographical areas are shown as 301a, 301b, and 301c. Each geographical area has at least one weather station 203a, 203b, 203c. There are one or more weather prediction servers 202a, 202b which provide predictions for weather parameters for geographical areas including geographical regions 301a, 301b, and 301c. The decision support server 201 is in communication with said weather stations 203a- c via said weather databases 202a-b or directly via the internet. The client device 204 is in communication with the decision support server 201 .

The memory 201 b of the decision support server 201 stores a temporal weather database, a plurality of weather prediction models and computer-readable instructions to cause the processor 201 a to: receive meteorological sensor data 102 from a plurality of weather stations located in and around a geographical area and predicted weather data comprising weather predictions for said geographical area is received from one or more weather prediction data sources 102.

Further, the processor 201a is configured to: compute a confidence parameter for each measurement in said received meteorological sensor data is determined. This step is carried out to ascertain the degree of accuracy of the received meteorological sensor data and/or to filter out corrupted meteorological sensor data or data received from faulty weather stations. In an embodiment, the corresponding confidence parameters are stored in a temporal weather database 201 b, such that weather prediction models may leverage the confidence parameters for making an accurate prediction. In other words, lower confidence on possibly inaccurate or corrupt meteorological sensor data may be filtered out or assigned a lower weight while predicting weather parameters in the near future by the weather prediction models. Thereby, providing highly accurate predictions for generating an advisory on agricultural activity.

The confidence parameter is determined by the processor 201a based on one or more of the following:

1 . Divergence between historical weather parameters for said geographical area and measured weather parameters received from a weather station located in said geographical area on the same day of the year and time of day in the past years. For example, a weather station may report rainfall of about 2mm of rainfall for the given geographical location for a particular day, however, the same day in previous years may have recorded an average of 200mm rainfall. Thus such a reading is suspected to be inaccurate or due to a faulty or a blocked sensor. In other words, long term reading assessment (at a given latitude/longitude) is computed e.g. for the same date and time for the last 10 years historical temperature data is analysed to obtain a long term average of 16.7 degrees Celsius. In view of the long term average, if received meteorological sensor data (temperature) is significantly lower or higher than said long term average then said received meteorological sensor data (temperature) may be flagged to be inaccurate or labelled with a low confidence.

2. Abrupt change in received measured weather parameters from said weather station within a predetermined temporal window for said geographical location.

That is, short term reading assessment (at a given geographical area latitude/longitude) is carried out, where the most recent readings are analysed, e.g. the meteorological sensor data for a given geographical area for the past 4 hours may read 13.0, 12.7, 12.6, 12.0 degrees Celsius respectively which seems to be a consistent decline in temperature which is expected. However, in a similar instance, if the readings would have read 13.0, 12.7, -5.6, 12.0 respectively, then such meteorological sensor data may be flagged as inaccurate or tagged with a low confidence due to an abrupt variation in the short term (-5.6 Celsius) which is not expected.

3. Divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location, where the weather parameters measured in geographical locations adjoining said geographical location are adjusted for change in geological factor and where the divergence between weather parameters measured in said geographical location and weather parameters measured in geographical locations adjoining said geographical location is weighted based on the distance between said geographical location and said adjoining geographical locations. For example, a weather station at a geographical location reports a temperature of 16 degree Celsius and the neighbouring weather stations report a temperature of 16.1, 15.9, 16 degree Celsius respectively, then the reading of the weather station is agreement with the readings of the neighbouring weather stations, hence such a reading may be tagged with a high confidence. On the contrary, if the weather station reports a temperature of 19 degree Celsius then it is highly likely that the weather station may be reporting an inaccurate temperature reading. However, a person skilled in the art would appreciate that temperature (weather parameters) may vary with altitude and distance from the weather station. Hence, such geological factors such as altitude and the distance from the weather station is factored while determining whether the temperature readings from the weather station agrees with the neighbouring weather stations. For example, a weather station may be located in a valley which reads 19 degree Celsius and a neighbouring weather station located on a mountain top may read 16 degree Celsius. In such case, by factoring in the difference in height (geological factor) of the temperature recorded by the weather station may be considered to agree with the temperature of the neighbouring weather station. In another example, if a weather station ‘A’ reports 12 degree Celsius and two neighbouring stations ‘B’ and Ό’ report temperatures of 12.5 and 13 degree Celsius respectively. Then the distance between weather station A and weather stations B and C respectively may be considered while computing the confidence of the temperature measured by weather station A. Accordingly, the variance in temperature between weather stations farther away may be assigned lesser weight than variance in temperature between weather stations closer together.

In an embodiment, both the geological factors and the distance factor may be considered for computing the confidence coefficient for a reading of a weather parameter received from a weather station when compared with neighbouring weather stations.

4. The divergence between difference of weather parameters measured in said geographical location and weather parameters predicted in said geographical location. For example, the predicted temperature for a geographical location is 20 degree Celsius and the received temperature reading is 24 degree Celsius, which is well beyond the prediction accuracy, then such a reading may be considered to be inaccurate. In another embodiment, difference of weather parameters measured in geographical locations adjoining said geographical location and predicted weather parameters in geographical locations adjoining said geographical location may be considered to evaluate the confidence coefficient. 5. Statistical occurrence of previous inaccuracies in measured weather parameters received from said weather station.

A person skilled in the art would appreciate that one of the above factors or a combination of the above factors may be utilized to compute the confidence coefficient of weather parameters received from meteorological sensor data sources.

Thereafter, the processor 201a is configured to temporally store said meteorological sensor data along with corresponding confidence parameters and said predicted weather data to the temporal weather database 201 b.

Further, the processor is configured to generate a plurality of weather predictions for said geographical area, using a plurality of weather prediction models stored in memory 201b, where said weather prediction models are implemented by the processor 201a to predict the weather based on present and past received meteorological sensor data and the corresponding confidence parameters associated with each measurement in said received present and past meteorological sensor data. In other words, the weather prediction models are used to predict weather parameters in the near future for said geographical area using the temporally stored meteorological sensor data in the temporal weather database.

Further, the processor 201a is configured to continuously evaluate the performance of each of weather predictions, received from weather prediction data sources and generated using said weather prediction models for said geographical area. The evaluation is performed based on a deviation of each of predicted weather parameters of each of weather predictions with the meteorological sensor data. Further the processor 201a is configured to predict each weather parameter for said geographical location based on said continuous evaluation by: providing a predicted weather parameter of a weather prediction selected from a plurality of weather predictions received from weather prediction servers or generated using said weather prediction models for said geographical area, where the selection of the weather prediction has a minimum deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter. For example, weather prediction model ‘X’ predicts a weather parameter (e.g. rainfall) with least deviation among the prediction models, and where other prediction models have significantly larger deviation than model X”, then model ‘X’ is selected for prediction for said weather parameter. Similarly, a prediction model or a weather prediction data source is selected for each of the weather parameters.

In an embodiment, a weighted predicted weather parameter is provided from ‘n’ weather predictions. Said ‘n’ weather predictions are selected from a plurality of weather predictions received from weather prediction servers and generated using said weather prediction models for said geographical area, where the selection of ‘n’ weather predictions are ranked in ascending order by their deviation from the meteorological sensor data for a predetermined temporal range for said predicted weather parameter, where the weight assigned to each weather prediction is based on said rank and where said selected ‘n’ weather predictions have similar evaluated performance. In a preferred embodiment the number ‘n’ is two. For example weather prediction model ‘X’ and Ύ predicts a weather parameter (e.g. rainfall) with least deviation among the prediction models, and where other prediction models have significantly larger deviation than models ‘X” and Ύ, then models ‘X’ and Ύ are selected for prediction for said weather parameter. In such a case, a weighted prediction is provided, where the model ‘X’ having a lesser degree of deviation with respect to model Ύ’, then model ‘X is weighted proportionately more than the prediction provided by model Ύ. Similarly, prediction models or a weather prediction data sources are selected for each of the weather parameters. BO

In a preferred embodiment, said geographical area has a radius of less than 5 kilometres. Further, the prediction of weather parameters comprises of prediction of weather parameters for said geographical area in the following 1 hour to 6 hours. Further, the predicted weather parameters comprise of rainfall, temperature, relative humidity, atmospheric pressure, solar radiation, and average wind speed, wind direction.

The processor 102a is configured to generate an advisory for one or more agricultural activities based on said prediction of said weather parameters. In an embodiment the advisory for one or more agricultural activities is/are further based on ground impact factor, wherein said factor being derived from a model comprising information on soil moisture, flooding risk, soil hydro-connectivity and texture and topology of said geographical area. In an embodiment said one or more agricultural activities comprises spraying, seeding, fertilizing, and harvesting.

Further, the processor 204a of the client device 204 transmits the location (global position coordinates) of the farm in question to the decision support servers 201 via said transceiver 204d. The decision support server 201 determines a geographical location 203a within which said farm resides. In response, the decision support server 201 transmits the weather prediction and advisory data for said geographical location 203a to the client device 204. The client device then receives the weather prediction and advisory data via the transceiver 204d and the processor 204a displays said weather prediction data and the advisory data on the display 204c of the client device.

Thereby, the precision decision support method and system provides precise advice to farmers to initiate or avoid an agricultural activity. Thereby, preventing losses and maximizing profits for the farmers.

Further, a person ordinarily skilled in the art will appreciate that the various illustrative logical/functional blocks, modules, circuits, and process steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or a combination of hardware and software. To clearly illustrate this interchangeability of hardware and a combination of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or a combination of hardware and software depends upon the design choice of a person ordinarily skilled in the art. Such skilled artisans may implement the described functionality in varying ways for each particular application, but such obvious design choices should not be interpreted as causing a departure from the scope of the present invention.

The process described in the present disclosure may be implemented using various means. For example, the apparatus described in the present disclosure may be implemented in hardware, firmware, software, or any combination thereof. For a hardware implementation, the processing units, or processors(s) or controller(s) may be implemented within one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field- programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, electronic devices, other electronic units designed to perform the functions described herein, or a combination thereof. For a firmware and/or software implementation, software codes may be stored in a memory and executed by a processor. Memory may be implemented within the processor unit or external to the processor unit. As used herein the term “memory” refers to any type of volatile memory or non-volatile memory.

In the specification, the terms "comprise, comprises, comprised and comprising" or any variation thereof and the terms include, includes, included and including" or any variation thereof are considered to be totally interchangeable and they should all be afforded the widest possible interpretation and vice versa.

The invention is not limited to the embodiments hereinbefore described but may be varied in both construction and detail.