Source: USDA NASS CDL Layer
Climate is arguably the most significant force in environmental change. Cropland is directly affected by climatic changes through loss in soil moisture, increases in CO2 levels, changes in temperature, and extreme weather conditions. South-Eastern North Dakota is a widespread producer of spring wheat, corn, soybeans, and barley. These crops are largely affected by weather patterns and changes in climate. Creating a state transition model helps predict hypothetical outcomes for crop production and informs management decisions farmers make, yielding an improved crop output.
What is a State Transition Model?
A state transition model shows transitions from one state to another due to a certain external variable. For example, a coniferous forest would transition to a deciduous forest due to a wildfire. Through a diagram, you can see a certain stratum undergo state transitions. A state to state transition has a probability attached. For example the probability of “Grassland” changing state to “recently burnt.” The diagram below shows different transitions between states in the Savannah.
Source: Wiki Resilience Assessment
Figure 1: Each box indicates the type of state in the ecological system. The arrows indicate the transition to each state which is represented by ‘T1.’ ‘T1’ stands for transition one, ‘T2’ stands for transition two and so on.
SyncroSim is a state transition software based off of Monte Carlo iterations. Monte Carlo iterations use a sample mean from each iteration to factor out uncertainty. It is a framework for running scenario-based simulations. SyncroSim uses computer languages such as R Scripts to run models based on probabilities. These probabilities help output hypothetical scenarios. It provides better data visualization capabilities and decreases uncertainty compared to other state transition or agent based modeling softwares. You can integrate TIFF files into the program to run simulations across the cells in the image.
Cropland State Transition Model
When creating a state transition model, start by identifying which state classes are being worked with. If a farmer were to decide to change the type of crop planted in a certain farming grid, you would have to know all the possible outcomes and variables that will create the maximum yield. The stratum, or initial layer of land that we are running a model over, can be called cropland. Next, one should identify the type of class-to-class transitions desired. In my cropland state transition I have four states, spring wheat, corn, soybeans, and barley. From each state, I have classified five transition types: temperature, precipitation, harvest, erosion, and crop rotation. The model gives probability for each transition type.
Figure 2: Creating probabilistic transitions, on how states transition based on erosion, temperature or harvest. A hypothetical probability is given on the likely chance the transition will occur. This creates a model like Figure 3 below.
Here is an example of utilizing the software using made of values as represented in the image above spring wheat has three transitions, or possible outcomes: spring wheat to barley due to erosion, which has a probability of 2.5%, spring wheat to soybeans due to temperature, which has a probability of 2.5%, and spring wheat back to spring wheat due to harvest, which has a 95% probability. These probabilities can be projected onto a reclassified TIFF image, which then predicts which crops stay as spring wheat versus switch to a different crop due to climate variables. Below is a possible cropland state transition model.
Figure 3: State and transition model in the Syncrosim software. Using the transition created in Figure 2.
Adding Spatial Dynamics to a Model with a Learning Curve
The model created is a made up model with made up variables. My probabilities are not based on any external variables such as related literature, statistical analysis, or professional expertise. That being said, below is an example of how the cells from a TIFF are given new states, calculated through Monte Carlo iterations to produce a new map based off of the probabilities given. The spatial dynamics that I ran are randomized to each cell because I did not set a minimum TST value (time-since-transition). This will allow for pixel values that are equal to change collectively instead of changing randomly with each individual cell. These images demonstrate how cropland of corn, (yellow) and soybean (orange) pixels are randomized for each probability, and the cell is changed to a different state.
State Transitions models have a multitude of benefits such as utilizing a finite number of input values, being able to look at a sequence of events or transition of stratum over a period of time, its use of tabular representation of the systems behavior, and the benefit of representing results in a less obvious way. The next steps for the state transition models is to run the cells dependent of each other, this can be done by classifying more advanced variables and giving each cell a range on a maximum and minimum direction it can run in. Working with a climate scientist on weather patterns and creating scientifically accurate probabilities is also a next step to better predict a hypothetical transition model. Lastly, reclassifying geoTIFF images to create new images that use a pixel value of one to represent a certain crop such as, corn, barley, or spring wheat, and all other pixel values as 0. This will help cells move more dependent of each other and will get rid of the randomization of cells.