# How to explain the spatial relation of forest loss hotspots with several others variables

I am doing a research in categorizing and ordering of the factor affecting/fostering the creation of the forest loss hotspots. I have following data.

• Point location of forest losses from which i generated polygon using kernel density.
• Euclidean distance raster of several variables like several types of roads, brick kilns(that uses trees as fuel),location of local market where fallen trees are sold, forest guard posts and others that may have relation to the creation of forest loss.

I see there is a relation like loss hotspots tend to be closer to roads and distant to the location of the forest guard post.

Now my question is:

How can I spatially/statistically explain above phenomenon?

Update: My model details:

• I have ~80000 loss and ~80000 non-loss(randomly sampled)
• I have ~70 predictors

There is a vast amount of literature about (spatial) modelling in GIS. I do not know the details of your data, but assuming that your points are actually indicating the presence of a given phenomenon (forest loss), and assuming that you have the possibility to locate with certainty locations featuring the absence of the same phenomenon (i.e., no forest loss), you could try Binary Logistic Regression, in which the Dependent Variable is forest loss presence/absence (coded as 1 and 0 respectively), and your distances would act as Predictors.

This is something that is used, for instance, in landslides studies in order to assess the impact of different predictors in the odds for the positive outcome of an event, in this case presence of a landslide.

As for GIS: -I would create a point SHAPEFILE representing all the locations under study, that is the ones featuring forest loss and the ones featuring no forest loss; -Create a field for your DV, which would contain 0 for the point with NO loss, and 1 for points WITH loss. -Then, extract the values of your Predictors (e.g., distance from road, location of kilns, etc) to your points.

At that point, you can export the table and perform Logistic Regression outside GIS, in R for example. First, you should check if the model is significant, and then, the estimated Betas (coefficients) should inform you about the strength and direction of the "influence" of your predictors on the chances (actually, odds) for forest loss.

You can plug the Betas (and intercept) in the Raster Calculator to produce a forest loss probability maps. If you are not familiar with R (which is free), you can perform Logistic Regression via a quite simple function which is provided by a website that I found just googling a little (http://cainarchaeology.weebly.com/r-function-for-binary-logistic-regression.html).

There are more things to check for and take into account in regression modelling with spatial data: for instance, the spatial autocorrelation among the model's residuals. If that is the case, you could try to subsample from your universe of points in order to increase the distance between your sampling location so alleviating for any existing spatial autocorrelation. Otherwise, you could resort to autoregressive modelling strategies (but I am not so familiar with these).

These are just my two cents. Hope they prove useful.

Zhang, Z. X., Zhang, H. Y., & Zhou, D. W. (2010). Using GIS spatial analysis and logistic regression to predict the probabilities of human-caused grassland fires. Journal of Arid Environments, 74(3), 386–393. https://doi.org/10.1016/j.jaridenv.2009.09.024

Arekhi, S. (2011). Modeling spatial pattern of deforestation using GIS and logistic regression: A case study of northern Ilam forests, Ilam province, Iran. African Journal of Biotechnology, 10(72), 16236–16249. https://doi.org/10.5897/AJB11.1122

Hu, Z., & Lo, C. P. (2007). Modeling urban growth in Atlanta using logistic regression. Computers, Environment and Urban Systems, 31, 667–688. https://doi.org/10.1016/j.compenvurbsys.2006.11.001

Wang, L., Sawada, K., & Moriguchi, S. (2011). Landslide Susceptibility Mapping by Using Logistic Regression Model with Neighborhood Analysis: A Case Study in Mizunami City. International Journal of Geomate, 1(2), 99–104. Retrieved from http://www.gi-j.com/serial 2/99-104-2c-wang.pdf

• Some other pieces of information that you should also provide are: how many point location do you have? How many predictors are you going to use? Also, bear in mind that (as many source in literature stress) the spatial auto-correlation you have to check for is among the model's residuals, not among the values of your predictors. BONUS advice: you should also be careful to not use a number of predictors too large relative to your sample size. – NewAtGis Mar 19 '17 at 17:58