This is a long one, forgive me...
I am hoping to see if my workflow looks solid to anyone with experience in plot level analysis.
The problem is as follows: I extract height metrics at plot level (see workflow below), combine height metrics into csv containing plot level dependent variable (Basal Area), run my regression in R using independent variables suggested by prior research with good data, but I get an r^2 of about .24...
My suspicion is that something I am doing during my data processing is wrong and was hoping someone who has done this could look it over. The reason for my suspicion is that 1) I am new to lidar and data processing and 2) my canopy cover data looks pretty messy...
Unlicensed LAStools (I'm a grad student)
1) index, tile and buffer (lasindex, lastile, 200m tiles, 20m buffers)
2) ground classify (lasground, small buildings so select towns, ultrafine)
3) replace z (lasheight)
4) remove noise (the default settings worked well, I didn't have much noise to begin)
5) clip 1/10acre (.04ha) plots
6) extract cloud metrics (tried both FUSION and LASCanopy)
7) BA is positively skewed, take natural log (RStudio: datum$logBA=log(datum$BA)
8) run regression in RStudio (results=lm(BA~Cov+P99, data=datum)
9) results: r^2=0.24
Now, I've checked a raw clipped plot against the normalized one, and it looked fine, no loss of data or anything, so I'm pretty confused... I'll paste a link to some samples.. maybe 1/10acre .04ha plots are too small? point density is too small? Thanks.