Range value of raster data varies before and after mosaic implementation

Question

I am trying to raster::mosaic bunch of .tif files extracted from hdf4 MODIS files. My script is working fine but I am puzzled by the range of raster values (min, max) before and after data processing. The original hdf4 rasters are 8 bits integer raster ranged between 0-255 for every individual raster files before mosaicing. However, when I do a raster mosaic then the resulting raster is ranged between 0-25. My question is why?

I tried to reproduce this through a reproducible example but the problem doesn't apply to any example data such as below:

#------------------------
rx1 <- raster(nrows=2400, ncols=2400, vals = floor(runif(5760000, min=0, max=256)) )
extent(rx1) <- c(11119505, 12231456, -4447802, -3335852)
rx2 <-  raster(nrows=2400, ncols=2400, vals = floor(runif(5760000, min=0, max=256)) )
extent(rx2) <- c(12231456, 13343406, -4447802, -3335852)

lr.1 <- list()
lr.1[[1]] <- rx1
lr.1[[2]] <- rx2

lr.1$fun <- mean
rast.mosaic1 <- do.call(mosaic,lr.1)
rast.mosaic1

#class       : RasterLayer 
#dimensions  : 2400, 4800, 11520000  (nrow, ncol, ncell)
#resolution  : 463.3129, 463.3125  (x, y)
#extent      : 11119505, 13343407, -4447802, -3335852  (xmin, xmax, ymin, ymax)
#coord. ref. : +proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0 
#data source : in memory
#names       : layer 
#values      : 0, 255  (min, max)

If the same line of codes is applied to this data, then the result is:

lr.2 <- list()
lr.2[[1]] <- r[[1]]
lr.2[[2]] <- r[[3]]

lr.2$fun <- mean
rast.mosaic2 <- do.call(mosaic,lr.2)
rast.mosaic2

#class       : RasterLayer 
#dimensions  : 2400, 4800, 11520000  (nrow, ncol, ncell)
#resolution  : 463.3127, 463.3127  (x, y)
#extent      : 11119505, 13343406, -4447802, -3335852  (xmin, xmax, ymin, ymax)
#coord. ref. : +proj=sinu +lon_0=0 +x_0=0 +y_0=0 +a=6371007.181 +b=6371007.181 +units=m +no_defs 
#data source : in memory
#names       : layer 
#values      : 0, 25.4  (min, max)

Answer 1

When read into R the data has values from 0 to 25.4:

> r1 = raster("./Lai_500m1.tif")
> r2 = raster("./Lai_500m3.tif")
> range(r1[],na.rm=TRUE)
[1]  0.0 25.4
> range(r2[], na.rm=TRUE)
[1]  0.0 25.4

So this is happening before the mosaic. The mosaic is not relevant here.

255 is the "missing data" value. From the raster metadata:

Band 1 Block=2400x3 Type=Byte, ColorInterp=Gray
  Description = MOD15A2H MODIS/Terra Gridded 500M Leaf Area Index LAI (8-day composite)
  NoData Value=255
  Unit Type: m^2/m^2
  Offset: 0,   Scale:0.1

and there are:

> table(is.na(r1[]))

  FALSE    TRUE 
5759987      13 

13 out of 6 million in the first and 105 out of 6 million in the second file. You won't see them when plotting.

The metadata seems to define a lot of special values:

255 = _Fillvalue, assigned when:
    * the MOD09GA suf. reflectance for channel VIS, NIR was assigned its _Fillvalue, or
    * land cover pixel itself was assigned _Fillvalus 255 or 254.
254 = land cover assigned as perennial salt or inland fresh water.
253 = land cover assigned as barren, sparse vegetation (rock, tundra, desert.)
252 = land cover assigned as perennial snow, ice.
251 = land cover assigned as "permanent" wetlands/inundated marshlands.
250 = land cover assigned as urban/built-up.
249 = land cover assigned as "unclassified" or not able to determine.
248 = no standard deviation available, pixel produced using backup method.

and I think these are (apart from 255 which maps to NA in R) scaled by 0.1 because of that scaling factor in the previous metadata block. Hence in r1 there's a few pixels that are 25.4 (perennial salt or inland fresh water), 25.3 (barren, sparse vegetation) and 25.0 (urban/built-up). The rest are all between 0 and 7.

Anyway, that's not relevant to understanding why your mosaic was between 0 and 25.4 That was because your data was truly between 0 and 25.4, and 25.5 is the missing data value, and R reads it as NA.