Agriyieldz User Guide

The purpose of this document is to inform potential end-users about the functionality of the APSIM Wheat Yield Data Cube. Here, we describe the methodology, including the source data, APSIM crop model configurations and outputs available to end-users via an API.  See Agriyieldz API Guide for details on how to access the product via the API.

Background 

CSIRO have created a unique data product about wheat crop yields for Australia. This data product can provide insight into the wheat production potential, the variability of production and the riskiness of production potential. Data are available at the field scale, where fields, farms, or regions can be compared for productivity and variability in crop yield. Data have been created using the APSIM wheat crop model, a soil grid, and climate data from the SILO patched point data base. Data can be exported, via an API, to end-users for reprocessing and integrating into other data platforms. 

Methodology 

The APSIM Next Gen Crop model (Holzworth et al., 2018) was used to generate simulations of wheat crop yield across the grain growing belt of Australia in every year from 1990 to 2020. Simulations were created for each possible wheat crop field in Australia. CSIRO have mapped 1.7 million fields across the continent, and these simulations are designed to augment those field boundaries. These data allow for example, an individual or organisation to make field to field, farm to farm and region to region comparisons relating to production potential, and the variation in production.  

Locations

At each location (i.e. the centroid of a field), the functional soil was selected from a the ASRIS soil map. The soils were selected from 18 possible candidates, that represent the functional properties of soils across Australia (Searle et al., 2021). These soils are defined with sufficient complexity to provide insight into how crops and pastures will grow under different climates. They are not a literal translation of soil type, as defined by the Australian Soil Classification system. Rather, they are a subset of the soils from the APSOIL database. The complete list of soils used to create the simulations are provided in Table 2.  Simulations were only executed for the soil at the exact centroid location of the field, and simulations were only executed for sensible candidate soils for the location. For example, there are no peaty soils (51) in Western Australia, and simulations were not run for that soil in WA.

Table 1. Number of simulations performed across Australia to create the APSIM wheat data cube

Table 2. List of soils used to create the APSIM wheat data cube (Location of original soil profile in parentheses).

Management

To create the data cube, general crop management rules were created to grow a crop. Regardless of location, or season, sowing took place on the 15th May, with the crop emerging 5 days later. The broadly adapted wheat cultivar, Wyalkatchem, was selected. The objective was to grow a crop to its water limited yield potential, where yields were not limited by a lack of nutrients. To achieve yield potential, initial total soil nitrogen was 300 kg N/ha. Initial water, at January 1, was 10% of capacity, and initial residues were 1000 kg/ha wheat stubble. 

Data

For each location, in each year from 1990 to 2020, data are available. The primary output is wheat yield, but additional variables relating to soil water, growing season rainfall and occurrence of extreme events such as frost, and heat stress are available (Table 3). The outputs from the 30-year simulations of wheat are stored in a database. Geographic co-ordinates are stored, and it is possible to interrogate the data for a single year and for multiple years. Comparisons are possible between fields, regions and states. 

Table 3. Outputs, available from 1990 to 2020 for every field.  

Sense Testing and Data screening

Outputs from the model were compared to a simple French- Schultz model, that predicts yield in response to growing season rainfall and starting soil moisture.  Additional outputs about temperature extremes for heat and frost were extracted from the data. The goal was to create a simple evaluation that allowed the user to quickly gauge if the model is performing as expected for the desired location. APSIM is more sophisticated than the French-Schultz model, so perfect correlations are not expected. However, over a 30 year simulation, at one location, the mean error should be less than 20%.  Maps of the mean error, between the French-Schultz model and APSIM are created.  

Results

APSIM output, for WA, SA, Vic, NSW and Qld are presented in Figures one to five. The figures illustrate the spatial distribution of long term (30 year) potential yield provided by APSIM.  The mean deviation at a single location, from a French-Schultz equation for WA, SA and Vic are generally low, and average less than 20%. Within a year, the deviation can be higher, as the APSIM model is more complex, and accounts for a greater range of abiotic stresses than the French-Shultz equation. For example, In NSW and Qld, the mean deviation is higher, in part because the French-Shultz model does not summarise potential yield as well, where in-season rainfalls are often low.

 

Open

Figure 1. APSIM yields for Western Australia. Outputs are an average of the 30 year simulated yield for each point.

 

Open

Figure 2. Deviation from the French-Schultz model for APSIM output for WA. Outputs are the average difference between APSIM and French-Schultz over a 30 year period.

Additional Analyses

Output from the APSIM wheat data cube can be used to compare locations. In Figure 11,  the yields over 30 years, from two paddocks in Western Australia are compared. The paddock with low yield originates from the low rainfall zone (-31.5936 ,  118.2347). This particular field had a long term mean of 2.45 t/ha, with a standard deviation of 0.62 t/ha and a coefficient of variation of 26%. The other paddock originates from the high rainfall zone (-34.5714, 117.01). The long-term mean, standard deviation and coefficient of variation for this field was 3.52t/ha, .84 t/ha and 24%.  Data can be extracted from as many fields as needed, and where necessary aggregated to create regional assessments.

Figure 11. Crop yields, generated by the APSIM crop model from 1990 to 2020 for fields located in the high rainfall zone and low rainfall zone of Western Australia.

Conclusion

The APSIM wheat yield data cube is a first generation data output, that provides information about wheat yields across Australia. It can be accessed via API, and used to evaluate differences in wheat production across the continent.

References

Holzworth, Dean, N. I. Huth, J. Fainges, H. Brown, E. Zurcher, R. Cichota, S. Verrall, N. I. Herrmann, B. Zheng, and V. Snow. ‘APSIM Next Generation: Overcoming Challenges in Modernising a Farming Systems Model’. Environmental Modelling & Software 103 (1 May 2018): 43–51. https://doi.org/10.1016/j.envsoft.2018.02.002.

Searle, R, Hochman, Z, Horan H, Steinberg D (2021) A Method for Assessing the Spatial Drought Risk of Winter Cereal Cropping in Australia Using Digital Soil Mapping and Deterministic Crop Modelling. In Review.