Input type
Input for HLPS must be in datacube format!
See also
Check out the datacube tutorial, which explains how we define a datacube.
The different Submodules either process ARD or feature datasets:
ARD are Level 2 Analysis Ready Data.
Alternatively, Level 3 Best Available Pixel composites can be input, too. They consist of a reflectance product (mostly BOA, but TOA, IMP, BAP are supported, too), and pixel-based quality information (mostly QAI, but INF is supported, too). These input data need to follow a strict data format, including number of bands, naming convention with time stamp, sensor etc.
See also
Check out the ARD tutorial, which explains what Analysis Ready Data are, and how to use the FORCE Level 2 Processing System to generate them..
If multiple sensors are used, analyses are restricted to the overlapping bands:
SENSOR
BLUE
GREEN
RED
RE1
RE2
RE3
BNIR
NIR
SWIR1
SWIR2
VV
VH
LND04
Landsat 4 TM
1
2
3
4
5
6
LND05
Landsat 5 TM
1
2
3
4
5
6
LND07
Landsat 7 ETM+
1
2
3
4
5
6
LND08
Landsat 8 OLI
1
2
3
4
5
6
SEN2A
Sentinel-2A
1
2
3
4
5
6
7
8
9
10
SEN2B
Sentinel-2B
1
2
3
4
5
6
7
8
9
10
sen2a
Sentinel-2A
1
2
3
7
sen2b
Sentinel-2B
1
2
3
7
S1AIA
Sentinel-1A IW asc.
1
2
S1BIA
Sentinel-1B IW asc.
1
2
S1AID
Sentinel-1A IW desc.
1
2
S1BID
Sentinel-1B IW desc.
1
2
LNDLG
Landsat legacy bands
1
2
3
4
5
6
SEN2L
Sentinel-2 land bands
1
2
3
4
5
6
7
8
9
10
SEN2H
Sentinel-2 high-res
1
2
3
7
R-G-B
Visible bands
1
2
3
VVVHP
VV/VH Dual Polarized
1
2
Feature datasets can be anything from individual ARD datasets to external datasets like precipitation or DEM.
Most often, features are generated by one HLPS submodule, and then used by another one, e.g. generate Spectral Temporal Metrics with Time Series Analysis, then use these outputs as features in Machine Learning. The most important constraint is: HLPS only knows 16bit signed input, thus if you import external data, you need to scale accordingly.