REN R 690
Discussion
The importance data from the RandomForest in combination with the CART variance explained was used to determine the essential variables needed from the DEM. From the CART analysis, elevation, SDA and gradient explained the most variance in the data set and either could have been used as a first node. From RandomForest, for both the EC and satellite zone predictions, elevation was the most important variable followed be either SCA, SDA or gradient. Elevation is easy and accurate to determine from the differential GPS. The remaining variables are DEM outputs. Recall that SCA is the area upslope of a contour segment that recieves to flow from that segment while SDA is teh area downslope of a contour segment that contributes to flow from that segment. SCA is correlated with SDA so effectively using one you can easily determine the other. In both cases, gradient was either second or third in order of importance. The placement of gradient in the order of importance is not necessary, just that it is one of the top ones is. From the multi variate analysis performed on the data set, the most important information is elevation and SCA, SDA and gradient from the DEM output.
The model created in RandomForest for predicting EC visually appeared to predict the EC levels. The RandomForest model made using the zones obtained from satellite imagery also had close similarities between actual and predicted values. Either of these models can be used for applying variable rates of fertilizers on fields in the area with similar soil types and topography. The models are good for getting a general idea of the varying fertility and moisture contents in a the field based on the topography. The model does not predict the actual EC of the field. The EC values in the model were from the Lacombe data set and represent the moisture content and fertility at that specific time and area. To gain an understanding of this imagine applying the model to a field near Whitecourt. Whitecourt has higher precipitation than Lacombe so the fields should have a higher moisture content (EC). When you run the model the EC levels for the Whitecourt field will be low when they should be larger. This is because the EC values are for Lacombe conditons. When using the model, this should be known.
The model created in RandomForest for predicting EC visually appeared to predict the EC levels. The RandomForest model made using the zones obtained from satellite imagery also had close similarities between actual and predicted values. Either of these models can be used for applying variable rates of fertilizers on fields in the area with similar soil types and topography. The models are good for getting a general idea of the varying fertility and moisture contents in a the field based on the topography. The model does not predict the actual EC of the field. The EC values in the model were from the Lacombe data set and represent the moisture content and fertility at that specific time and area. To gain an understanding of this imagine applying the model to a field near Whitecourt. Whitecourt has higher precipitation than Lacombe so the fields should have a higher moisture content (EC). When you run the model the EC levels for the Whitecourt field will be low when they should be larger. This is because the EC values are for Lacombe conditons. When using the model, this should be known.
Conclusions
There are various practical applications from these findings. Determining the topographic variables that influence the soil characteristics related to crop growth allows for a more concise output from the DEM. Rather than using seven output variables from the DEM in the model, realistically only three are needed: SDA, SCA and gradient. The model created in RandomForest visually appears to be accurate but still needs to be tested on a seperate data set before any further conclusions can be made. The proposed model is simple and less costly than other methods used for determining the varying fertility of a field but still precise. This will benefit the producer and ultimately the consumer
