
This repository contains the data from Saunders et al. "Continuity in Deep Learning Satellite Inundation Mapping from MODIS to VIIRS", submitted to the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.
These data are three-dimensional (lat, lon, time), gridded, historical estimates of inundation created using optical surface reflectance measurements from the MODIS or VIIRS satellites, and a combined Convolutional Neural Network and Long Short-Term Memory Network (CNN+LSTM) deep learning model.
These data are an updated version of the data used in Giezendanner et al. (2023a), available from: https://datacommons.cyverse….
The corresponding code to train and infer the CNN+LSTM model is available from: https://github.com/GieziJo/….
There are three zip folders, each containing fraction of inundated area estimates for the entire area of Bangladesh: (i) MODIS.zip (2001–2024), (ii) VIIRS.zip (2011–2024), and (iii) MODIS_7Bands.zip (using seven surface reflectance bands, 2001–2024).
| Values: | Fraction of inundated area estimates, continuous values between 0 and 1 (float32). |
| File format: | Zarr: (latitude; longitude; time). See Instructions for guidance on file and data access. |
| Spatial coverage: | Bangladesh. |
| Temporal coverage: | MODIS: 2001–2024. VIIRS: 2011–2024. |
| Spatial resolution: | 500 m nominal pixel resolution. |
| Temporal resolution: | 8-days. |
Each .zip folder contains data stored in Zarr format. First, download and unzip the data. Then, the Zarr data (latitude; longitude; time) can be loaded (and if desire converted to other common geospatial data formats) using the xarray Python package. Having installed xarray, use xarray.open_zarr(dataset) to load the data. Note that the Zarr data are stored in 'chunks'. For more information, see the xarray and Zarr documentation: https://docs.xarray.dev/en/…; https://zarr.readthedocs.io….
For the most accurate historical inundation estimates, users are recommended to use the MODIS_7Bands data, which was created using seven MODIS surface reflectance bands (the other datasets used only three bands). As shown in Figure S4 of Saunders et al. "Continuity in Deep Learning Satellite Inundation Mapping from MODIS to VIIRS", inundation estimates created using seven bands slightly outperform estimates from three bands of MODIS/VIIRS data.
For applications that require consistent inundation estimates in the future (i.e., for years after 2024), users are recommended to use the VIIRS (three-band) data. We intend to update the VIIRS data at semi-regular intervals in the future. The MODIS data shows some indication of sensor measurement degradation from 2022 onwards, likely associated with the orbital drift of MODIS Terra nearing it's end-of-life. Please contact the authors if you have a specific need that we may be able to assist with.
Giezendanner, J., Mukherjee, R., Purri, M., Thomas, M., Mauerman, M., Islam, A.K.M.S., Tellman, B., 2023a. Inferring the past: a combined CNN–LSTM deep learning framework to fuse satellites for historical inundation mapping, in: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Presented at the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), IEEE, Vancouver, BC, Canada, pp. 2155–2165. https://doi.org/10.1109/CVP…
Giezendanner, J., 2023b. Bangladesh Inundation History. https://doi.org/10.25739/2E…
Giezendanner, J., 2023c. GieziJo/cvpr23-earthvision-CNN-LSTM-Inundation. https://github.com/GieziJo/…