Forecasting field level crop yield during a growing season

Abstract

A method for predicting field specific crop yield recommendations for a field may be accomplished using a server computer system that is configured and programmed to receive over a digital communication network, electronic digital data representing agricultural data records, including remotely sensed spectral property of plant records and soil moisture records. Using digitally programmed data record aggregation instructions, the computer system is programmed to receive digital data representing including remotely sensed spectral property of plant records and soil moisture records. Using the digitally programmed data record aggregation instructions, the computer system is programmed to aggregate the one or more digital agricultural records to create and store, in computer memory, one or more geo-specific time series over a specified time. Using the digitally programmed data record aggregation instructions, the computer system is programmed to select one or more representative features from the one or more geo-specific time series and create, for each specific geographic area, a covariate matrix in computer memory comprising the representative features selected from the one or more geo-specific time series. Using mixture linear regression instructions, the computer system is programmed to assign a probability value to a component group in a set of parameter component groups, where each component group within the set of parameter component groups includes one or more regression coefficients calculated from a probability distribution and an error term calculated from a probability distribution. Using distribution generation instructions, the computer system is programmed to generate the probability distributions used to determine the one or more regression coefficients and the error term, the probability distribution used to generate the error term is defined with a mean parameter set at zero and a variance parameter set to a field specific bias coefficient.

1 .- 24 . (canceled) 
     
     
         25 . A computer-implemented method of forecasting crop yield, comprising:
 obtaining, by a processor, one or more agricultural data records that represent one or more types of data for plants at one or more fields during a specific time period of one or more years, wherein the one more types of data include at least a remotely sensed spectral property of plant records;   creating, by the processor, one or more geo-specific time series over the specific time period from the one or more agricultural data records;   selecting one or more representative features from the one or more geo-specific time series, including an aggregate value of a remotely sensed spectral property over multiple observations computed from the geo-specific time series;   creating, for a specific geo-location, a covariate matrix comprising the one or more representative features; and   determining a field specific crop yield for a specific date from the covariate matrix for the specific geo-location, one or more regression coefficients calculated from a first probability distribution and an error term calculated from a second probability distribution based on historical data.   
     
     
         26 . The computer-implemented method of  claim 25 , the obtaining comprising:
 receiving soil moisture data for the one or more fields;   predicting remotely sensed signals based on the soil moisture data.   
     
     
         27 . The computer-implemented method of  claim 25 , the creating comprising applying locally weighted scatterplot smoothing to a subset of the one or more geo-specific time series. 
     
     
         28 . The computer-implemented method of  claim 25 , the creating comprising applying a scaled Gaussian density to the one or more geo-specific time series. 
     
     
         29 . The computer-implemented method of  claim 25 , the remotely sensed spectral property being a vegetation index computed over one or more wavelength ranges. 
     
     
         30 . The computer-implemented method of  claim 25 , the one or more representative features including a maximum value of the remotely sensed spectral property for a specific period within a year across the one or more fields when the specific date is before a threshold point in a crop season. 
     
     
         31 . The computer-implemented method of  claim 25 ,
 the one or more types of data including soil moisture data for the one or more fields,   the one or more representative features including a soil wetness value or a soil dryness value of a field over a certain number of days, wherein the soil wetness value is above a first moisture threshold and the soil dryness value is below a second moisture threshold.   
     
     
         32 . The computer-implemented method of  claim 25 , the determining further comprising
 computing a linear regression model for the specific geo-location that describes a relationship between a dependent variable of crop yield for the specific geo-location and independent variables corresponding to the covariate matrix,   the linear regression model comprising the one or more regression coefficients.   
     
     
         33 . The computer-implemented method of  claim 25 , each of the first probability distribution and the second probability distribution being a normal distribution. 
     
     
         34 . The computer-implemented method of  claim 25 , the determining further comprising determining a prediction interval associated with the field specific crop yield, wherein the prediction interval is range of values that measures a level of certainty associated with the field specific crop yield. 
     
     
         35 . One or more non-transitory storage media storing instructions which, when executed by one or more computing devices, cause performance of a method of forecasting crop yield, the method comprising:
 obtaining one or more agricultural data records that represent one or more types of data for plants at one or more fields during a specific time period of one or more years, wherein the one more types of data include at least a remotely sensed spectral property of plant records;   creating one or more geo-specific time series over the specific time period from the one or more agricultural data records;   selecting one or more representative features from the one or more geo-specific time series, including an aggregate value of a remotely sensed spectral property over multiple observations computed from the geo-specific time series;   creating, for a specific geo-location, a covariate matrix comprising the one or more representative features; and   determining a field specific crop yield for a specific date from the covariate matrix for the specific geo-location, one or more regression coefficients calculated from a first probability distribution and an error term calculated from a second probability distribution based on historical data.   
     
     
         36 . The one or more non-transitory storage media storing of  claim 35 , the obtaining comprising:
 receiving soil moisture data for the one or more fields;   predicting remotely sensed signals based on the soil moisture data.   
     
     
         37 . The one or more non-transitory storage media storing of  claim 35 , the creating comprising applying locally weighted scatterplot smoothing to a subset of the one or more geo-specific time series. 
     
     
         38 . The one or more non-transitory storage media storing of  claim 35 , the creating comprising applying a scaled Gaussian density to the one or more geo-specific time series. 
     
     
         39 . The one or more non-transitory storage media storing of  claim 35 , the remotely sensed spectral property being a vegetation index computed over one or more wavelength ranges. 
     
     
         40 . The one or more non-transitory storage media storing of  claim 35 , the one or more representative features including a maximum value of the remotely sensed spectral property for a specific period within a year across the one or more fields when the specific date is before a threshold point in a crop season. 
     
     
         41 . The one or more non-transitory storage media storing of  claim 35 ,
 the one or more types of data including soil moisture data for the one or more fields,   the one or more representative features including a soil wetness value or a soil dryness value of a field over a certain number of days, wherein the soil wetness value is above a first moisture threshold and the soil dryness value is below a second moisture threshold.   
     
     
         42 . The one or more non-transitory storage media storing of  claim 35 , the determining further comprising
 computing a linear regression model for the specific geo-location that describes a relationship between a dependent variable of crop yield for the specific geo-location and independent variables corresponding to the covariate matrix,   the linear regression model comprising the one or more regression coefficients.   
     
     
         43 . The one or more non-transitory storage media storing of  claim 35 , each of the first probability distribution and the second probability distribution being a normal distribution. 
     
     
         44 . The one or more non-transitory storage media storing of  claim 35 , the determining further comprising determining a prediction interval associated with the field specific crop yield, wherein the prediction interval is range of values that measures a level of certainty associated with the field specific crop yield.