Water clarity (SDD) modeling and processing codes based on Deep Gated Recurrent Netowrk (DGRN) model:

"Water clarity mapping of global lakes using a novel hybrid deep-learning-based recurrent model with Landsat OLI images"

Citation:

Y. He, Z. Lu, W. J. Wang, D. Zhang, Y. L. Zhang, B. Q. Qin, K. Shi*, & X. F. Yang. 2022. Water clarity mapping of global lakes using a novel hybrid deep-learning-based recurrent model with Landsat OLI images. Water Research, 215, 118241. https://doi.org/10.1016/j.watres.2022.118241

Notes: This repo only uploaded the codes which are used in this ciation. The in-situ measurements may share in the future.

Data pre-processing (almost working in GEE platform)

1. The global water bodies extraction:

This code extracted the pixels with water occurrence ≥ 25% based on the JRC Global Surface Water Mapping Layers, v1.3, which is provided by GEE. Detailed information was described in the above-mentioned citation.

The online accessed link is shown below:

https://code.earthengine.goo-gle.com/?scriptPath=users%2Fheyuan9874%2Fextract_inland_water%3Aextract_inland_water1

2. The surface reflectance of Landsat 8 OLI extraction:

This code extracted the lake-average of surface reflectance for B1, B2, B3, B4, B5, and B7 of Landsat 8 OLI based on global water bodies and lake boundary (provided by HydroLAKES [https://www.hydrosheds.org/products/hydrolakes]). Detailed information was described in the above-mentioned citation.

The online accessed link is shown below:

https://code.earthengine.google.com/529cda3218497357af0799cda42e4d39

Deep-learning-based models (working in Python platform)

1. DRGN_SDD_Landsat8 (also upload to releases)

This folder save the DGRN model, which input with normalized surface reflectance and output the log-transformed SDD. DGRN model running with Keras 2.6.0. The model architecture and parameterization are shown in the citation.

Data post-processing

1. draw_temporal_trend.py

The draw_temporal_trend.py is used to calculate the lake-average of SDD from surface reflectance based on DGRN model and extract the global lakes without ice-covered regions.

2. Fig1.drawio & Graphical_Abstract.drawio & FigS2.drawio

The Fig1.drawio, Graphical_Abstract.drawio, and FigS2.mxd are running in Drawio software (also work in VScode), which are used to draw the Fig. 1 (model architecture), the graphical abstract, and the working flow of Landsat imagery in the citation.

3. optical_properties.py

The optical_properties.py is used to describe the optical properties distribution of surface reflectance, as shown as Fig. 2.

4. sdd_DGRN.py

The sdd_DGRN.py is used to describe the model accuracy of the DGRN model in training and testing sets, as shown as Fig. 3.

5. Fig4.ipynb

The Fig4.ipynb draws the heatmap to depict the comparison among multiple empirical models, as shown as Fig. 4.

6. Fig5.py

The Fig5.py draws the Taylor diagram to depict the comparison among multiple empirical models, as shown as Fig. 5.

7. basemap_mapping_backups.ipynb

The basemap_mapping_backups.ipynb draws the spatio-temporal results of lake-average of SDD, as shown as Fig. 6, 7, 8, 9, 11, 12, S6, S7, S8. Detailed information can be accessed in the citation.

8. Fig10.ipynb

The Fig10.ipynb draws the histogram to depict the contributions of lake-specific characteristics to SDD, which is shown in Fig. 10.

9. normalization_sdd_l8.py

The normalization_sdd_l8.py draws the scatter plots to depict the model accuracy of multiple empirical models, which is shown in Fig. S3.

10. case_lakes.mxd & FigS1.mxd

The case_lakes.mxd, and FigS1.mxd are working in ArcGIS 10.3, which were used to describle the spatial variability of four lake cases and the distribution of in-situ measurements.

11. FigS5.ipynb

The FigS5.ipynb draws the probability distribution density of SDD in four lake cases, which is shown in Fig. S5.