Masters thesis help with soil variability

Question

I am currently a masters student working on analyzing my first year of data that I've collected from both plant and soil data collection. I work as program technician in cotton research and analyzing the effect different soil textures have on plant maturity and development. In the 3 fields we have collected plant data using Cotman (if you know what that is great) we have collected bi-annual (fall after harvest and mid to early spring) soil sample collection with a giddings probe to a depth of 36". These were then sent to a lab for a complete nutrient analysis (N,P,K, and all the various micro nutrients).

My question being a newcomer to the geostatistcal analysis is what tools should I use to show the relationship of texture to its nutrient holding capacity of the soil as well as its affect on plant maturity and structure. I've talked to my professors who have basically given me little to no help in where to start with my massive load of data. Any advice?

I forgot to mention soil texture was collected using Veris 3150 soil EC mapping unit.

MarcoBoeringa · Answer

Hi Erin,

I have a few remarks based on your comments:

- First and foremost, I really think you should try and contact a statistician at your university to help you out on ways to analyze this data. I am not a statistician myself, nor researcher at the moment (although I have an MSc in Biology), but I have never understood the great reluctance of some researchers to call in external help on this. They spend years and tons of money collecting data, and than "forget" the need for proper analysis. Even if it is "just" a master thesis, you have collected data that may be valuable to others and your guiding professors / PhD students. Yes, the statisticians may be busy and difficult to get hold of, but may be of invaluable help in showing you the ways to handle your data.

- You suggest having multiple factors influencing your growth / yield data (soil parameters, treatment types). Many of these may be correlated to each other. You may not get significant results putting one parameter (e.g. soil nitrogen content) against the yield data. Often, it may be necessary to analyse all data / multiple parameters at once, in one statistical analysis. You may either need or want to bundle multiple dependent or independent variables to get meaningful results. There are methods for that in statistical packages, e.g. like the MANOVA I wrote, but there are others as well. Have a clear look at all the options in the statistical package, or read Wikipedia pages etc., so as to understand the uses of different methods of analysing your data.

- You wrote: The issue for my data is that well its kinda "parabola shaped"

Often, it is necessary to normalize your data by applying a transformation, e.g. taking the log of measurements before analysing the data. Most statistical methods assume normal distributions of the datasets, if your data isn't, you should transform it before analysis. Most statistical packages, and also ESRI's Geostatistical Analyst, offer options for testing for normality, and allow you to choose a proper transformation that best fits your data.

- I also think you should have clear Hypothesis in order to test your data statistically. I get the feeling you are still in the process of trying to find out what questions you exactly want to answer... while already having collected all the data!

- Based on your additional information in your last post, this re-affirms my believe you should first analyse the data using normal statistical methods before venturing into the world of "geostatistics". I don't think this kind of data is well suited to geostatistical analysis.

ErinKelly · Answer

I've done some initial analysis with my soils in SAS 9.3 running some correlations. The issue for my data is that well its kinda "parabola shaped". I have high variability of different soil types within an individual field. I have high clay content ( a routon dundee crevasse complex is what its classified as) and a Mhoon fine sandy loam in the same field (nearly a beach in some areas). These are dramatically different soils but their effect on the plant is similar (diminished structure, reduced water uptake, etc.) these areas have greatly diminished yield but for very different reasons ( soil moisture capacity or compaction). I'm just a little confused on how to account for this variability and show their impact or does that even matter as far as variability with the soil texture. I've also looked at the different soil nutrients in a correlation and found it to explain any of the plant data. I'm just a wee bit frustrated with my data and resources

My intent with my project is to show that fields associated with high soil textural variability will have a sufficient enough difference in  yield and development of the crop (herbcide and insecticide application) associated with soil texture that they can benefit from zone based management strategies with the new tools used in production like variable rate application for herbicides, fertilizer, seeding rates and so on. Ultimately trying to set some "floors" and "ceilings" hopefully on yield expectations on the different soil types.

I don't know if that helped in understanding what I need from my analysis or just made things more confusing :confused:

MarcoBoeringa · Answer

My question being a newcomer to the geostatistcal analysis is what tools should I use to show the relationship of texture to its nutrient holding capacity of the soil as well as its affect on plant maturity and structure. I've talked to my professors who have basically given me little to no help in where to start with my massive load of data. Any advice?

To be honest, I would be very weary of using Geostatistical Analyst for something like soil data. Soils and rock types often vary very erratically, meaning the data doesn't conform very well to basic assumptions / prerequisites for interpolations. It often violates continuity prerequisites. E.g. you can have two points close together having completely different soil types (e.g. peat / sand) over a very short distance, leading to entirely different measurements as well.

I think your best bet for analyzing this data, is to start out with classical statistical analysis. Forget about Geostatistical Analyst, and use a program like S-PLUS, SPSS or any of the other multitude of commercial or Open Source statistical packages to do an Analysis of Variance (ANOVA) or Multivariate Analysis of Variance (MANOVA). There may be other tests suited to your type of data as well, have a careful look at the prerequisites for each test to understand its uses, and maybe contact a statistician at your university for some good advice.

It should give you a pretty good idea of whether there are any statistical relationships / correlations between your soil textures and plant maturity. I did something similar during my Biology study based on growth data of transplanted species moved to different succesional zones on a beach plane.

Once statistical correlations have been determined, you might consider creating maps by simply classifying a soil type / geological map or any other base map related to your data, based on the results of the statistical analysis.

I know it isn't as "attractive" and "fancy/in vogue" as a colorful Geostatistical Analyst map, but do you want sound statistical analysis, or just wall paper? 😉 (no pun intended, I just want to make the point).

By the way, I am not suggesting Geostatistical Analyst and its interpolation methods do not have its place in your type of research. There may be cases where it is entirely valid. I am just saying that you should be careful applying it and mind the prerequisites, and never forget that there is the option of "non-spatial" statistical analysis, that you may even need to consider first, before venturing into the world of geostatistics.