# Kriging method vs. IDW in soil moisture interpolation

I have XY coordinates and soil moisture values for a grid of 10m and I used add XY data to import my excel data to ArcMap 10.5. I applied both IDW and kriging methods for interpolation. The results are very different and after studying a bit about the differences, I still cannot say which one is more accurate? My second problem is that kriging gives the min and max value of 5-44, while IDW gives the range between 3-62 and my actual soil moisture values are closer to the values of IDW. I did not change any default while processing these two methods and I do not know why kriging just deletes some of the most important soil moisture values after 44 m3.m-3.

I attach the two images for comparison if anybody can help.

]1

Kriging is an advanced method based on the construction of a semi-variogram, which require al large number of points (or some knowledge about the phenomenom). In your case their seem to be only little points, so it might be difficult to fit the best model. First thing that you could do is removing the nugget effect of the kriging so that it always use point values. You could also try to manage the anisotropy of your data (more spatial correlation along the NE direction than along the NW direction, so it is better to select neighbours in the NE direction)

In your case, the best tool to get a quick result would be a TIN.

• Thanks for your reply. I searched for TIN tool and created this TIN. I also applied Spline method which gives somehow good results but adds a new class value more than the actual existing values and it also produces some negative values! I read that kriging is suitable for normally distributed variables and in my case, the soil water content is about 10 in all of the field but changes suddenly in the middle up to 50. Commented Nov 6, 2018 at 10:13
• IDW, simple kriging and TIN are exact interpolator and never go out of range. Spline is not always an exact interpolator and it goes out of range. I wouldn't discard kriging because of the sudden change, but when you have only a few pairs of distances and no pairs of points lying close to each other, it is difficult to find the most appropriate kriging model. i think that kriging makes the hypothesis of normally distributed residual errors, not of normally distributed values. Commented Nov 6, 2018 at 10:41

An IDW would likely work fairly well here given your distribution of points, but there are a couple of things to consider:

1. Parameters: Your IDW points are weighted far too heavily. You can see this because each point has a defined circle around it rather than a nice smooth gradient. In this regard think of your surface as a bed sheet and your points as bowling balls. If the points are too heavy, they will cause local area sagging where the points are.

2. Resolution: The coarser your resolution the less pixelated your result may be. It looks like you're attempting a fairly high resolution IDW given the spaces between your points

There are two main variables here - power (weight) and smoothing. You need to change these until you get something that meets your ideal needs. Really, you should keep some points out of the dataset and then use them as independent validation metrics afterwards. First I would suggest making the power very small and the smoothing very high - you can compare it with your output above and get an idea of what effect each of the parameters has on your surface.

Some more info on the IDW (and Other) algorithm is given here: https://gdal.org/programs/gdal_grid.html#interpolation-algorithms