Long-range temperature forecasting

1. A method for providing an improvement in long-range temperature forecasting using agricultural applications, the method comprising:

an agricultural intelligence computer receiving an observation dataset that identifies at least a forecast climatology model;

the agricultural intelligence computer training a climatology model on the observation dataset to generate an observed climatology model;

wherein the climatology model includes as covariates a combination of a linear trend and a series of harmonics;

the agricultural intelligence computer identifying observed anomalies based on the climatology model and the observation dataset;

the agricultural intelligence computer identifying forecast anomalies based on at least he forecast climatology model;

the agricultural intelligence computer modeling the observed anomalies as a function of the forecast anomalies to generate a calibration function that is used to correct the forecast climatology model based on the observed anomalies;

the agricultural intelligence computer, upon receiving a new forecast produced by the forecast climatology model, calibrating the new forecast using the calibration function.

2. The method of claim 1 , wherein the forecast anomalies are further identified based on a reforecast dataset that represents predictions produced by a forecast model for one or more past times and the new forecast is a prediction produced by the forecast model for a future time.

3. The method of claim 2 , wherein the observation dataset further identifies values for an environmental variable that relates to temperature.

4. The method of claim 3 , wherein the environmental variable is average temperature and a climatology model climatological average temperature.

5. The method of claim 1 , wherein modeling the observed anomalies as a function of the forecast anomalies is performed using Nonhomogeneous Gaussian Regression.

6. The method of claim 5 , wherein the Nonhomogeneous Gaussian Regression uses minimizing continuous ranked probability score or maximizing log-likelihood as an objective function.

7. The method of claim 5 , wherein the Nonhomogeneous Gaussian Regression uses one or more covariants representing one or more of: seasonal bias, ensemble mean, ensemble standard deviation, recent anomalies, forecasts from previous run times, or forecasts at other lead times.

8. The method of claim 3 , wherein the observation dataset includes a value and a variance for the environmental variable at one or more times and the reforecast dataset includes a value, a lead time, and a variance for the environmental variable at the one or more times.

9. The method of claim 1 , wherein calibrating the new forecast using the calibration function is performed by:

using the new forecast as input to the forecast climatology model to obtain a forecast climatological value;

computing a forecast anomaly based on the forecast climatological value and the new forecast;

using the forecast anomaly as input to the calibration function to generate a calibrated forecast anomaly;

generating an observational climatological value based on a time of prediction of the new forecast using the observed climatology model; and

adding the forecast anomaly to the observational climatological value to generate the calibrated forecast.

10. The method of claim 9 , wherein the calibration function takes as input the value for the forecast anomaly, a variance for the forecast anomaly, and a lead time for the new forecast and produces a calibrated value for the forecast anomaly and a calibrated variance for the forecast anomaly based on the lead time.

11. A system for providing an improvement in long-range temperature forecasting using agricultural applications, the system comprising:

one or more processors;

one or more non-transitory computer-readable storage mediums storing one or more instructions which, when executed by the one or more processors, cause the one or more processors to perform:

receiving an observation dataset that identifies at least a forecast climatology model;

training a climatology model on the observation dataset to generate an observed climatology model;

wherein the climatology model includes as covariates a combination of a linear trend and a series of harmonics;

identifying observed anomalies based on the climatology model and the observation dataset;

identifying forecast anomalies based on at least the forecast climatology model;

modeling the observed anomalies as a function of the forecast anomalies to generate a calibration function that is used to correct the forecast climatology model based on the observed anomalies;

upon receiving a new forecast produced by the forecast climatology model, calibrating the new forecast using the calibration function.

12. The system of claim 11 , wherein the forecast anomalies are further identified based on a reforecast dataset that represents predictions produced by a forecast model for one or more past times and the new forecast is a prediction produced by the forecast model for a future time.

13. The system of claim 12 , wherein the observation dataset further identifies values for an environmental variable that relates to temperature.

14. The system of claim 13 , wherein the environmental variable is average temperature and a climatology model climatological average temperature.

15. The system of claim 11 , wherein modeling the observed anomalies as a function of the forecast anomalies is performed using Nonhomogeneous Gaussian Regression.

16. The system of claim 15 , wherein the Nonhomogeneous Gaussian Regression uses minimizing continuous ranked probability score or maximizing log-likelihood as an objective function.

17. The system of claim 15 , wherein the Nonhomogeneous Gaussian Regression uses one or more covariants representing one or more of: seasonal bias, ensemble mean, ensemble standard deviation, recent anomalies, forecasts from previous run times, or forecasts at other lead times.

18. The system of claim 13 , wherein the observed dataset includes a value and a variance for the environmental variable at one or more times and the reforecast dataset includes a value, a lead time, and a variance for the environmental variable at the one or more times.

19. The system of claim 11 , wherein calibrating the new forecast using the calibration function is performed by:

using the new forecast as input to the forecast climatology model to obtain a forecast climatological value;

computing a forecast anomaly based on the forecast climatological value and the new forecast;

using the forecast anomaly as input to the calibration function to generate a calibrated forecast anomaly;

generating an observational climatological value based on a time of prediction of the new forecast using the observed climatology model; and

adding the forecast anomaly to the observational climatological value to generate the calibrated forecast.

20. The system of claim 19 , wherein the calibration function takes as input the value for the forecast anomaly, a variance for the forecast anomaly, and a lead time for the new forecast and produces a calibrated value for the forecast anomaly and a calibrated variance for the forecast anomaly based on the lead time.