Regional Smoothing of Insecticide Use Data
Depends
To run this analysis two files are required:
rat_ins.tifraster file produced in a similar way to the method reported in the repository R_SpatialAnalysis_Mapping. This is a raster file of insecticide use across the Prairie Pothole Region of Canada- Folder
catchmentwith a vector file containing deliniations of 5000+ river basins.
R version 4.0.3
Packages: mgcv, tidyverse, rgdal, raster, proj4, spdep, mgcv, viridis, gridExtra
Code and analysis was largely based on this blog created by Gavin Simpson.
Description
Here by smoothing the insecticide distribution, it would be possible to identify the “hotspot” areas with regard to insecticide use across the Prairies.
Initially, I produced the raster rat_ins.tif, which represents insecticide use (kg/ha) across the Prairie Pothole Region of Canada at 1-km resolution. This raster file was overlaid to the river basin vector file catchment.shp. I calculated mean insecticide distribution for each river basin. I fitted generalized additive models (GAM) to the data found in the packages mgcv. The level of smoothness is defined by the Markov Random Field (MRF), which allows modeling of spatial data with an intrinsic Gaussian Markov random field (GMRF). As explained by G.Simpson: 'MRFs are quite flexible as you can think about them as representing an undirected graph whose nodes are your samples and the connections between the nodes are specified via a neighbourhood structure'.
Below I'll show two sets of figures produced by this analysis. Figure 1 shows the distribution when mean insecticide use is used (basically a choropleth map) and and with full rank MRF, which is pretty close to the mean map.
Figure 1: Mean application per catchement and GAM with full MRF ranks for insecticide
Figure 2 shows three sets of GAM models with 30-1000 MRF ranks. You can visually see the smoothing happening more in 30 ranks than in 1000 MRF. In comparison to Figure 1, in Figure 2 it is easier to visually identify areas which are different from neighboring locations, and the regional hotspots are more evident as a result of smoothing.
Figure 2: Different levels of smoothness for insecticide use
References:
https://www.fromthebottomoftheheap.net/2017/10/19/first-steps-with-mrf-smooths/