compilation_of_land_use_datasets - PIK-LPJmL/LPJmL GitHub Wiki

Compilation of land use datasets

[[TOC]]

structure of binary land-use data sets

All LPJmL input data sets in clm format have a header and contain raw binary data in float (4-byte) or short (2-byte) format. An example of writing the header for a new clm2 file is given at Adding new input files.
With that function, a land use data set can be generated with R using code like this:

# create an array to store data (here for 32 bands and 401 years (1700-2100))
data.out <- array(0,dim=c(32,NPIX,401)) # use this order to allow for as.vector()
fill data.out accordingly, make sure all sums over the 32 bands are between 0 and 1
...
# open file for binary writing
f.out <- file(myfilename,"wb")
# write header
fwriteheader(f.out,"LPJLUSE",32, 0.001,fy=1700,ny=401)
# write data
writeBin(as.integer(as.vector(data.out)*1000),f.out,size=2)
close(f.out)

Dataset 1: 29 bands dataset, documented in Fader et al. (2010)

The data from Portmann et al. (2010) include maximal monthly harvested areas in a 0.5° resolution, we assumed for each pixel and CFT the month with the highest harvested area to be representative for the annual harvested area. In a first step, we combined the total fractional coverage of a grid cell with cropland and pasture (Ramankutty et al., 2008) and the fractional distribution of each of the irrigated and rainfed crop types (Portmann et al., 2010), which both represent the situation around the year 2000. While the original pasture fractions from Ramankutty et al. (2008) were taken without modifications, we made sure that the sum of fractions of the individual CFTs in a cell (from Portmann et al. (2010)) did not exceed the total agricultural area in a cell from Ramankutty et al. (2008), which could occur e.g. if there is multi-cropping (more than one crop cycle within a year in the same grid cell). We achieved this by proportionally reducing the fractions of each CFT to fit the total cropland fraction. If, in turn, regions classified as cropland by Ramankutty et al. (2008) are not considered by Portmann et al. (2010), we added these additional fractions to our model’s “rainfed others” CFT category. In a second step, the thus created land-use dataset for around the year 2000 was extrapolated backward to the year 1700—following the relative changes (aggregated from 50 to 300 resolution) given by the HYDE 3 dataset of the decadal cropland and pasture extent (Klein Goldewijk and van Drecht, 2006)—which is required for a correct model spin-up. Analogous to an earlier interpolation procedure (Bondeau et al., 2007), we assumed that a CFT’s relative fraction of the dynamic total cropland area within a grid cell was constant over time. For the period 2001–2005 (not covered by HYDE), the CFT fractions were assumed to follow the trends over the period 1990–2000. Since the HYDE data do not provide information about the historical extent of irrigated areas, we assumed a global irrigated area in 1700 of 3 Mha (Stefan Siebert, personal communication). To estimate the temporal evolution of this area, we calculated the global trend from Hoekstra’s (1998) decadal data for 1900–1990 and used this trend for the periods 1700–1899 and 1991-2005 as well. Subsequently, the historical irrigated area in each grid cell was calculated extrapolating the proportion of every grid-cell to the total irrigated area (globally) back in time.

Dataset 2: 32 bands dataset (not documented) was done by Sibyll by

subtracting sugar cane areas from the “others” of dataset 1.

Dataset 3: 32 bands dataset documented in Porkka et al. (in prep) and done by Matti Kummu.

Takes dataset 1 and merge it with the HID dataset by Siebert et al (2015) for 1900-2005 – AEI (HID)

  1. aggregation of 5 arc-min HID data to 30 arc-min resolution
  2. interpolated for each year (HID dataset is with 10-year intervals for 1900-1980, and with 5-year intervals for 1980-2005)
  3. allocation of AEI to LPJ landuse, repeated for each year
    • identification of grid cells where AEI (HID) > cultivated_area (dataset 1)
      • entire cultivated area is irrigated
    • identification of grid cells where AEI (HID) > 0 && AEI (dataset 1) = 0 =>AEI (HID) allocated based on year 2000 AEI (dataset 1) share of irrigation between CFTs (i.e. if CFT2 and CFT4 had shares of 1/3 and 2/3 of year 2000 total AEI (dataset 1), the AEI (HID) was allocated in a similar way). This wad done in an iterative way, i.e. if a CFT got 100% irrigated, the remaining AEI (HID) was shared between the other CFTs irrigated in year 2000 in dataset 1
      • if still remaining AEI (HID) after CFTs irrigated in year 2000 were fully irrigated, the remaining AEI (HID) was allocated to dominant not-yet-irrigated crops
    • identification of cells where AEI (HID) = 0 && cultivated_area(dataset 1) > 0
    • AEI (LPJ_new) set to zero for all CFTs
    • identification of cells where AEI (HID) > 0 && AEI (dataset 1) > 0 && AEI (HID) < cultivated_area(dataset 1)
      • as in step b, AEI (HID) is allocated based on year 2000 AEI (dataset 1) share of irrigation between CFTs (i.e. if CFT2 and CFT4 had shares of 1/3 and 2/3 of year 2000 total AEI (dataset 1), the AEI (HID) was allocated in a similar way). This wad done in an iterative way, i.e. if a CFT got 100% irrigated, the remaining AEI (HID) was shared between the other CFTs irrigated in year 2000 in dataset 1
      • if still remaining AEI (HID) after CFTs irrigated in year 2000 were fully irrigated, the remaining AEI (HID) was allocated to dominant not-yet-irrigated crops
    • identification of cells where remaining AEI (HID) and pasture
    • remaining AEI (HID) allocated to pasture
  4. Compiling a new LPJ historical landuse dataset

Dataset 4: 64 bands, done by Jonas.

Takes dataset 3 and separates irrigated areas in surface, sprinkler and drip irrigated.
More for this is actually found in the parent : Input

Dataset 5: 52 bands dataset documented in Fader et al.(2015)

Merging 2014 versions of HYDE, MIRCA, Portmann and Ramankutty in a similar (but not identical) approach to dataset 1.

rainfed band crop irrigated band
1 Temperate Cereals 27
2 Rice 28
3 Maize 29
4 Tropical Cereals 30
5 Pulses 31
6 Temperate Roots 32
7 Potatoes 33 n
8 Tropical Roots 34
9 Sunflower 35
10 Soybeans 36
11 Groundnuts 37
12 Rapeseed 38
13 Sugar Cane 39
14 Citrus 40n
15 Non Citrus Orchards 41n
16 Date Palm 42n
17 Olives 43n
18 Nuts Trees 44n
19 Grapes 45n
20 Vegetables 46n
21 Cotton 47n
22 Fodder grass 48n
23 Others 49
24 Manage Grasslands 50
25 Bioenergy Grass 51
26 Bioenergy Tree 52

Dataset 6: 76 bands dataset, takes dataset 3 and merges with dataset 5 by:

  • adding potatoes areas from dataset 5 to temperate roots
  • separating from others the three new bands (perennials). When others did not have enough areas, it was looked if managed grass did. If there were still missing areas the perennial areas of dataset 5 were still adopted for the new bands, when setting others and managed grass to zero (this is why the areas of dataset 6 are slightly higher than the ones in dataset 3).

Then the procedure for separating the surface, sprinkler and drip irrigation is applied, delivering 4 times following bands:

band crop
1 Temperate Cereals
2 Rice
3 Maize
4 Tropical Cereals
5 Pulses
6 Temperate Roots
7 Tropical Roots
8 Sunflower
9 Soybeans
10 Groundnuts
11 Rapeseed
12 Sugar Cane
13 Evergreen broadleaved agricultural trees
14 Big deciduous broadleaved agricultural trees
15 small deciduous broadleaved agricultural trees
16 Others
17 Manage Grasslands
18 Bioenergy Grass
19 Bioenergy Tree
⚠️ **GitHub.com Fallback** ⚠️