1 of 32

Agrivate

Agricultural Forecasting for Farmers

2 of 32

Needfinding

3 of 32

Value

97% of all farmable land in the United States grow grains (wheat, soybeans, corn).
75% reductions in yields when optimal moisture is not maintained throughout the entire hop growing season.
The Midwest United States produces 78% of soybean and 82% of corn in the U.S., so we focused on that area of the country.
Optimizing the water usage for grain production would conserve resources and reduce the amount of runoff into major rivers in the region.

4 of 32

Corn Production in the United States

5 of 32

Soybean Production in the United States

6 of 32

Soil Monitoring

Poor soil monitoring and overuse of fertilizers can lead to water runoff that pollutes the waters and creates algae blooms that lead to dead zones in the oceans.
Soil monitors exist, but how could we predict and provide corrective measures days in advance for farmers?

How can we use this information to conserve water and prevent the pollutants caused by water runoff?

7 of 32

Problem Statement

How can we integrate machine learning into an application farmers can use to monitor and correct soil moisture in their fields?

8 of 32

Forecasting Model

9 of 32

Forecasting / Recommendation Model Overview

Precipitation

Air Temp

Humidity

U / V Wind Components

U-Net Soil Moisture Model

Soil Moisture

Evapotranspiration

Calculations

Irrigation Recommendation Calculations

Crop Parameters

Farmer Recommendation

10 of 32

Forecasting / Recommendation Model Overview

Data Used (22 Gigabytes of Data Total for 2019)

CPC Precipitation Data

Daily, Realtime
0.25 - degree latitude x 0.25 - degree longitude US grid (300x120)

NCEP/NCAR Reanalysis Weather Data

Daily, Realtime
2.5 degree latitude x 2.5 degree longitude global grid

NOAA Soil Moisture Products System (SMOPS) Daily Blended Products

Daily, Realtime
Global 0.25 x 0.25 degree grid

Data Processing

Crop to Bounded Area of Iowa

Lat: 40 to 46.31 degrees
Lon: -97 to -90.69 degrees

0.1 x 0.1 Degree Grid (~5.6 mile resolution)

11 of 32

U-Net Soil Moisture Model

Based on classic “U-Net: Convolutional Networks for Biomedical Image Segmentation” (https://arxiv.org/abs/1505.04597)
Modified to create new novel U-Net Regression Model

Output Activation: Linear
Dropout: 0.2
Batch Normalization is Used
Input Shape: (128,128,1)
Output Shape: (128,128,1)

Input

12 Hour Accumulated Precipitation for Bounded Area of Iowa
0.1x0.1 Degree Grid
Lat: 40 to 46.31 degrees
Lon: -97 to -90.69 degrees

Output

Soil Moisture over same gridded area

12 of 32

U-Net Soil Moisture Model

Precipitation

128x128x1

Soil Moisture

128x128x1

13 of 32

U-Net Soil Moisture Model Results

Final Mean Absolute Error:

±5.28 % Soil Moisture

Can be improved:

Deeper network
More filters / weights
More input Variables
More lagged input

14 of 32

U-Net Soil Moisture Model Results

Input

Precipitation

Predicted Output Soil Moisture

Actual Output Soil Moisture

15 of 32

Crop Resource Modeling

16 of 32

Evapotranspiration

Evapotranspiration is the sum evaporation of water from soil and transpiration of water from crops calculated from daily temperature normals, surface wind speeds, relative humidity, and sunshine
Evapotranspiration calculation give the millimeters per day lost based on all these daily weather normals.

17 of 32

Penman-Monteith Evapotranspiration Calculation

18 of 32

Corn Water Requirements per Maturity Stage

From the University of Missouri extension irrigation website

19 of 32

Soybean Water Requirements per Maturity Stage

From the University of Missouri extension irrigation website

20 of 32

Irrigation Recommendation Equation

Water Needed = Water In - Water Out

Water Needed = Water Requirements of Crop
Water In = (Soil Moisture Prediction based on precipitation) + (Irrigation by Farmer)
Water Out = Evapotranspiration

Water Requirements of Crop = Soil Moisture Prediction + Irrigation by Farmer - Evapotranspiration

Algebra

Recommended Irrigation by Farmer = Water Requirements of Crop - Soil Moisture Prediction + Evapotranspiration

21 of 32

Farmer Interface

22 of 32

Info / Tools

23 of 32

Impact

24 of 32

Relevance

According to the U.S. Department of Agriculture, the agricultural sector uses about 80% of the consumptive water use in the nation and about 90% of the consumptive water use in just the midwestern states.
Our model creates recommendations for precise irrigation levels based on a soil moisture prediction, but does not currently take into account evapotranspiration due to technical difficulties.
With precise recommendations of irrigation needed per field, farmers would only need to spray the fields that need the water and the amount that they need without having to generalize the entire field.

25 of 32

Looking Ahead

The conservation of water in the agricultural sector by using more precise irrigation methods would have a big impact on overall water conservation in the nation
Benefits

saving money and energy
protecting drinking water supply
reduced pollution and runoff
reduced need for water treatment plants
preservation of aquatic life

26 of 32

U.S. Waterways

27 of 32

Corn Production in the United States

28 of 32

Soybean Production in the United States

29 of 32

The Louisiana Dead Zone

30 of 32

Thank You!

Q & A

31 of 32

References

32 of 32

References