The objective of this study is to predict future tropical forest cover presence and types using multitemporal imaging spectroscopy data. Accurately predicting land cover changes with image time series is vital for assessing the effects of climate change and land management on forest resources. Data were obtained from the DLR Earth Sensing Imaging Spectrometer (DESIS) covering a region in the West Godavari district of Andhra Pradesh, India. DESIS, mounted on the International Space Station, records Earth observation data in 235 channels over a 400-1000 nm spectral range. Five overlapping cloud-free images were selected, capturing the seasonal variability among land covers to form the multitemporal image stack. 1,070 randomly generated training points spanning the five dates were visually classified into four land cover classes: forest plantation, palm plantation, natural forest, and non-forest. Future land cover was predicted using the following steps: (1) A recurrent neural network with long short-term memory (LSTM) was used to predict future reflectance values of the 235 bands for each point. This model had an R^2 coefficient of 83.0%. (2) A multi-layer neural network using Keras was trained on the classified points from each image with 5-fold cross-validation, achieving an accuracy of 73.0%. (3) The classification model from step 2 was then applied to the reflectance data generated from the LSTM (step 1) to predict the future land type at each point. The combined land cover prediction framework, titled Forestry and Other Land Use Neural Network (FOLU-Net), enables predictions of land use change without the need for potentially error-prone land use classifications at each prior time step necessitated by approaches such as Markov chain analysis. Our findings demonstrate a robust framework for characterizing the evolution of land cover using multitemporal imaging spectroscopy.
This study is presented in a manuscript under review at Ecological Informatics.
Publisher
University Libraries, Virginia TechCorresponding Author Name
Paige T. WilliamsFiles/Folders in Dataset and Description
FOLDER: GeoTIFF Files - This folder contains data for five overlapping 16-bit GeoTIFF images collected from the DLR Earth Sensing Image Spectrometer (DESIS), a hyperspectral imaging spectrometer launched in 2018 and installed on the International Space Station. DESIS has high spectral resolution of 235 spectral bands across 400-1000 nm and spatial resolution of 30 m. The spectrometer captures images in strips of tiles, allowing target-specific image data. The product level of the data is L2A. These images provide hyperspectral imagery of land cover in the West Godavari district near the Godavari River in Andhra Pradesh, India. The data was collected on the following five dates: May 10, 2019, March 25, 2020, April 02, 2020, December 09, 2020, February 10, 2021. Along with the original images, the clipped images, which have been cropped to the intersection of the five original images, are also provided. The bounding box of the intersection is characterized as the following using the WGS84 latitude/longitude coordinate system: lower left corner at (81.146, 16.941) and upper right corner at (81.218, 17.010). This folder contains the following ten files:
– 2019_05_sq_clipped.tif - The GeoTIFF file for the region clipped to the aforementioned bounding box taken on May 10, 2019 by DESIS.
– 2020_03_sq_clipped.tif - The GeoTIFF file for the region clipped to the aforementioned bounding box taken on March 25, 2020 by DESIS.
– 2020_04_sq_clipped.tif - The GeoTIFF file for the region clipped to the aforementioned bounding box taken on April 02, 2020 by DESIS.
– 2020_12_sq_clipped.tif - The GeoTIFF file for the region clipped to the aforementioned bounding box taken on December 09, 2020 by DESIS.
– 2021_02_sq_clipped.tif - The GeoTIFF file for the region clipped to the aforementioned bounding box taken on February 10, 2021 by DESIS.
– DESIS-HSI-L2A-DT0316537084_003-20190510T031347-V0210-SPECTRAL_IMAGE.tif - The original GeoTIFF file covering part of the West Godavari district of Andhra Pradesh, India taken on May 10, 2019 by DESIS.
– DESIS-HSI-L2A-DT0434501528_004-20200325T100710-V0210-SPECTRAL_IMAGE.tif - The original GeoTIFF file covering part of the West Godavari district of Andhra Pradesh, India taken on March 25, 2020 by DESIS.
– DESIS-HSI-L2A-DT0437483472_005-20200402T070144-V0210-SPECTRAL_IMAGE.tif - The original GeoTIFF file covering part of the West Godavari district of Andhra Pradesh, India taken on April 02, 2020 by DESIS.
– DESIS-HSI-L2A-DT0530004280_003-20201209T034704-V0213-SPECTRAL_IMAGE.tif - The original GeoTIFF file covering part of the West Godavari district of Andhra Pradesh, India taken on December 09, 2020 by DESIS.
– DESIS-HSI-L2A-DT0553210416_004-20210210T025056-V0213-SPECTRAL_IMAGE.tif - The original GeoTIFF file covering part of the West Godavari district of Andhra Pradesh, India taken on February 10, 2021 by DESIS.
FOLDER: LSTM Data - This folder contains CSV files providing the spectral data at 5,000 randomly sampled coordinates in the West Godavari district near the Godavari River in Andhra Pradesh, India. These coordinates were selected from the overlapping region of five hyperspectral images with a bounding box characterized as the following using the WGS84 latitude/longitude coordinate system: lower left corner at (81.146, 16.941) and upper right corner at (81.218, 17.010). The images were collected from the DLR Earth Sensing Image Spectrometer (DESIS), a hyperspectral imaging spectrometer launched in 2018 and installed on the International Space Station. DESIS has high spectral resolution of 235 spectral bands across 400-1000 nm and spatial resolution of 30 m. The CSV provides the spectral data of the 235 bands for each coordinate for each of the five dates at which the DESIS images were collected (May 10, 2019, March 25, 2020, April 02, 2020, December 09, 2020, February 10, 2021). The spectral data was extracted from each of the five images and compiled into an aggregate CSV file with every group of five rows representing the spectral data in chronological order for each date for a particular coordinate. This folder contains the following two files:
– lstm_train.csv - The training data for the LSTM which consists of 80% of the original data, providing the spectral data for 4,000 coordinates.
– lstm_test.csv - The testing data for the LSTM which consists of 20% of the original data, providing the spectral data for 1,000 coordinates.
FOLDER: DNN Data - This folder contains a CSV providing the spectral data and land cover classification at 1,070 locations in the West Godavari district near the Godavari River in Andhra Pradesh, India. These coordinates lie in the the overlapping region of five hyperspectral images with a bounding box characterized as the following using the WGS84 latitude/longitude coordinate system: lower left corner at (81.146, 16.941) and upper right corner at (81.218, 17.010). The images were collected from the DLR Earth Sensing Image Spectrometer (DESIS), a hyperspectral imaging spectrometer launched in 2018 and installed on the International Space Station. DESIS has high spectral resolution of 235 spectral bands across 400-1000 nm and spatial resolution of 30 m. The CSV classifies each point into one of four land cover classes with the following class distribution: 11.1% forest plantation, 22.6% palm plantation, 17.1% natural forest, and 49.2% non-forest. The labels for the data entries were obtained through a variety of sources including in situ classification of coordinates and visual classification via Google Earth imagery for the associated location and timeframe. This folder contains the following one file:
– dnn_data.csv - The full training and testing data for the DNN which consists of the spectral data and land cover classification at 1,070 locations in the West Godavari district.
FOLDER: Code Files - This folder contains all of the code files used to build the aforementioned machine learning models and visualize the results of the land cover forecast. It is necessary to ensure that all libraries used in these programs are installed beforehand. This folder contains the following four files:
– LSTM_code_final.ipynb - The code used to train the LSTM, generate the spectral plots for each coordinate in the testing dataset, and save the LSTM model.
– DNN_code_final.ipynb - The code used to train the DNN, conduct five-fold cross validation, compare the performance of the DNN with the random forest, and save the DNN model.
– map_visualization_final.ipynb - The code used to synthesize the predicted image for February 2021 using the LSTM for pixel-by-pixel prediction of the spectral values.
– classification_map_final.ipynb - The code used to apply the land cover classifications to the predicted image for February 2021 using the DNN to classify each pixel by land class.
– MIT_license.txt - The text file containing the MIT license for the aforementioned code files used in this study.
FILE: abbreviations.txt - This file contains a list of all abbreviations and their expansions that are relevant for the data and code files used in this study.
