Spatial modelling of soil nutrients using random forest algorithm

Machine learning techniques for predictions are very common nowadays.
But spatial predictions are not as common as normal prediction. Spatial
predictions can be done by using different Machine learning algorithms.
The study was proposed to predict spatial distribution of five soil nutrients
based on the Random Forest algorithm.Random forest is a machine
learning algorithm used here for spatial prediction. Spatial autocorrelation
techniques are sometimes biased and this is suboptimal. This
study presents a random forest for spatial predictions where buffer
distances from observed spatial points that are used as explanatory
variables. It incorporates geographical proximity effects into the prediction
process. The result shows that random forest can obtain equally accurate
and unbiased predictions as different versions of kriging. Advantages of
using random forest over kriging are that it needs no rigid statistical
assumptions about the distribution of the target variable. It gives more
informative maps. It is less biased than kriging. Random forest framework
for spatial prediction is more accurate because it trains an amount of data
and prediction is done by using that trained data. There are some
disadvantages also for random forest,its exponentially growing
computational intensity with increase of calibration data and covariates.
The high sensitivity of predictions to input data quality is also a
disadvantage. Random forest’s success is because of its training data
quality, especially its spatial data sampling quality.It minimizes
extrapolation problems and any type of bias in data.

MSC GIS 2017-2019 INT Radhakrishnan T