Hi Sadeepa,

Did you attempt to develop one final regression equation with all the plots? If that's the case than you can simple downscale your product (I guess it's a raster or can be converted into a .TIF raster file) into 10X10 size grids. In ENVI you can do the r sampling of rasters. If you don't want to use ENVI then you can simply try writing a script (may be Matlab) with the following steps,

1. Create a NAN array of your desired domain size. estimate the number of rows and columns by divising your domain dimensions by 10, as 10 is your new grid size. This will be the size of your NAN array. The idea is to apply the regression equation for each cell of the array (row,column).

2. Run a for loop to acess each element of your NAN array and then Apply the equation and update the NAN array with actual values.

dorry for the long descriptions if you're already an expert in scripting than the steps might be too basic for you.

You can use Spatial Association Regression techniques. You need data as features (polygon/point) to carry out these methods within a network like 10 * 10 meters! After obtaining coefficients and constant values of the regressions, you can apply the result for all cells of the network, and then convert it to raster!

Follow the methodology of the bellow link:

Article Analysing Spatial and Statistical Dependencies of Deforestat...

Is there a statistically significant difference between the 50-by-50 and the 25-by-25 based models?

@Jack Kornfield

First two models were developed separately. No statistically significant difference was found. Thus one model was developed using all the plots.

Dear Sadeepa,

In my opinion, the practical approach is to export your processed remote sensing outcome in geotif format at pixel size of 50m, and then import it into GIS package such as ArcGIS, while your field data are added as point shapefile. Using Extraction to Points function, you will get the remote sensing output value(s) at pixel level corresponding to each field plot.

You can then use either SPSS or SYSTAT or SAS etc to conduct different kinds of regression analysis (linear, nonlinear, single or multiple) to obtain your regression model(s).

Wish this can help you.

All the best,

Weicheng

Dear Sadeepa,

The models are the same ----Great.

All measurements have error. Do you know the measurement error of your instrument. If not find it out. Do you know how well you would like to apply your model?

To obtain information that is good at 10m-by-10m, you might just think of sampling at intervals smaller than 10 m and applying a two-dimensional Gaussian filter; choose the sampling rate and width of the filter to minimizing measurement error.

Of course, the data format will also enter into determining what is best for you to perform the filtering. Good Luck

Regards,

Jack

https://www.quora.com/How-do-satellites-use-IR-cameras-to-do-thermal-images-from-hundreds-of-kilometres

https://www.quora.com/How-do-satellites-use-IR-cameras-to-do-thermal-images-from-hundreds-of-kilometres

Hi Sadeepa,

The easiest and simple option will be to covert you all data to 25x25 grid and use it to generate the regression model. In second step convert the image to 25 x 25 m grid and apply the regression model using raster calculator in ArcGIS.

The other thing which will be important to know is the type of regression model. Are you using linear regression? Multilinear regression or polynomial or what?

I would prefer to use random forest regression as it is more accurate and handle the nonlinear relationship.

Article Poppy Crop Height and Capsule Volume Estimation from a Singl...

Article Poppy Crop Capsule Volume Estimation using UAS Remote Sensin...