hpscil / custnlffm

cuSTNLFFM is a GPU-enabled Spatial and Temporal Non-Local Filter-based Fusion Model (STNLFFM)

cuSTNLFFM

Version 1.0

Overview

MODIS and Landsat surface reflectance products have complementary characteristics in terms of spatial and temporal resolutions. To fully exploit these datasets, the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) was developed by Gao et al. (2006). The STARFM approach blends the high-frequency temporal information from MODIS and the high-resolution spatial information from Landsat to generate synthetic surface reflectance products at 30m spatial resolution and daily temporal resolution. STARFM uses one or more pairs of Landsat-MODIS images collected on the same dates to predict surface reflectance at Landsat resolution on other MODIS observation dates. Many image fusion models have been deveoiped since then. Cheng et al. (2017) deveoped a Spatial and Temporal Non-Local Filter-based Fusion Model (STNLFFM), which derives a new transformation relationship between the multi-temporal fine-resolution reflectance images with the help of multi-temporal coarse-resolution reflectance images, and makes full use of the high degree of spatiotemporal redundancy in the image sequence to produce the final prediction. However, the computational performance of STNLFFM has been a bottleneck for mass production.

To overcome the computational barrier and support mass production of large-size images, we designed and implemented a GPU-enabled STNLFFM program based on the Compute Unified Device Architecture (CUDA), called cuSTNLFFM. By taking advantages of the large amount of concurrent computing threads of a GPU, cuSTNLFFM can greatly reduce the computing time and improve the computational performance. Experiments showed that cuSTNLFFM achieved a speedup of 100 using a Nvidia Tesla K40 GPU, compared with a sequential STNLFFM program running on an Intel Xeon E3-1226 CPU.

Key features of cuSTNLFFM:

• Supports a wide range of CUDA-enabled GPUs (https://developer.nvidia.com/cuda-gpus)
  • Automatic setting of the numbers of threads and thread blocks according to the GPU’s available computing resources (e.g., memory, streaming multiprocessors, and warp)
  • Adaptive cyclic task assignment to achieve better load balance
  • Optimized use of registers to improve the computational performance
  • Adaptive data decomposition when the size of images exceeds the GPU’s memory
  • All above are completely transparent to users
• Outputs any number of prediction images
• Supports a wide range of image formats (see http://gdal.org/formats_list.html)
• Supports both Windows and Linux/Unix operating systems

References

• Gao, F.; Masek, J.; Schwaller, M. and Hall, F. On the Blending of the Landsat and MODIS Surface Reflectance: Predict Daily Landsat Surface Reflectance, IEEE Transactions on Geoscience and Remote Sensing. 2006, 44(8):2207-2218.
• Cheng, Q.; Liu H.; Shen H.; Wu, P. and Zhang, L. A Spatial and Temporal Nonlocal Filter-Based Data Fusion Method. IEEE Transactions on Geoscience & Remote Sensing, 2017, 55(8):4476-4488

To Cite cuSTNLFFM in Publications

• Please cite the following reference:
  Gao, H., Zhu, X., Guan, Q., Yang, X., Yao, Y., Zeng, W., Peng, X., 2021. cuFSDAF: An Enhanced Flexible Spatiotemporal Data Fusion Algorithm Parallelized Using Graphics Processing Units. IEEE Transactions on Geoscience and Remote Sensing. https://doi.org/10.1109/TGRS.2021.3080384

Compilation

• Requirements:
  • A computer with a CUDA-enabled GPU (https://developer.nvidia.com/cuda-gpus)
  • A C/C++ compiler (e.g., Microsoft Visual Studio for Windows, and gcc/g++ for Linux/Unix) installed and tested
  • Nvidia CUDA Toolkit (https://developer.nvidia.com/cuda-downloads) installed and tested
  • Geospatial Data Abstraction Library (GDAL, http://gdal.org) installed and tested
• For the Windows operating system (using MS Visual Studio as an example)
  1. Open all the source codes in Visual Studio
  2. Click menu Project -> Properties -> VC++ Directories -> Include Directories, and add the “include” directory of GDAL (e.g., C:\GDAL\include)
  3. Click menu Project -> Properties -> VC++ Directories -> Lib Directories, and add the “lib” directory of GDAL (e.g., C:\GDAL\lib)
  4. Click menu Build -> Build Solution
     Once successfully compiled, an executable file, cuSTNLFFM.exe, is created.
• For the Linux/Unix operating system (using the CUDA compiler --- nvcc)
  In a Linux/Unix terminal, type in:
  • $ cd /the-directory-of-source-codes/
  • $ nvcc -o cuSTNLFFM kernel.cu cuLayer.cpp ReadParameters.cpp fusion.cpp -lgdal
    Once successfully compiled, an executable file, cuSTNLFFM, is created.

Usage

• Before running the program, make sure that all Landsat and MODIS images have been pre-processed and co-registered. They must have:
  • the same spatial resolution (i.e., Landsat resolution --- 30m)
  • the same image size (i.e., numbers of rows and columns)
  • the same map projection
• A text file must be manually created to specify the input and output images, and other parameters for the STNFLLM model.
  Example (# for comments):

STNLFFM_PARAMETER_START

#The number of input pairs of Landsat-MODIS images (>=1)
NUM_IN_PAIRS = 1

#The input MODIS images
#File names are separated by space
IN_PAIR_MODIS_FNAME = D:\cuda\shikong\testdata\M_2002_01_04.tif

#The input Landsat images
#File names are separated by space

IN_PAIR_LANDSAT_FNAME = D:\cuda\shikong\testdata\L_2002_01_04.tif

#The MODIS images for the prediction dates
#Multiple images can be given
#File names are separated by space
IN_PDAY_MODIS_FNAME = D:\cuda\shikong\testdata\M_2002_02_12.tif

#The output synthetic prediction images
#Multiple images can be given
#File names are separated by space
OUT_PDAY_LANDSAT_FNAME = D:\cuda\shikong\testdata\estr.tif

#The_width of searching_window
The_width_of_searching_window = 51

#is_limit_a_CalcuRela
is_limit_a_CalcuRela = 1

#Filter_parameters
Filter_parameters = 0.15

#L_ERR,High resolution data observation error
L_ERR = 0.005

#M_ERR,Low resolution data observation error
M_ERR = 0.005

#Spectral empirical parameters
d = 0.01

#Reflectivity products, exponential products and surface temperature products are marked as 1 / 2 / 3
p = 1

#Block side length
patchSize= 1

#Output image format (optional)
#Will be used when the extension of the output files
#is not given
G_Type = GTIff

STNLFFM_PARAMETER_END

• The program runs as a command line. You may use the Command (i.e., cmd) in Windows, or a terminal in Linux/Unix.

  • For the Windows version:
    $ cuSTNLFFM.exe parameters.txt
  • For the Linux/Unix version:
    $ ./cuSTNLFFM parameters.txt

• Note: The computational performance of cuSTNLFFM largely depends on the GPU. The more powerful is the GPU, the better performance.