# Land2Sent

A benchmark for Landsat to Sentinel imageries via deep learning based super-resolution methods.

Dataset Details

Image pairs of this dataset queried inside Google Earth Engine using the following criteria's: Cloudless images, the year 2023 images, acquisition time less than 1 hour between image pairs, and common area of pairs must be at least 100 km x 100 km. 30 pairs manually selected from these pairs using Land2Sent GEE application. Sentinel images are tiled 480 x 480 pixels and Landsat images are tiled 160 x 160 pixels for super resolution model training. Total of 15066 tiles produced. The dataset splitted %70 training, %20 validation, and %10 testing.

Benchmark Results

Metrics

Metric values on normalized 4-band images:

Method	PSNR↑	SSIM↑	AG↑	NIQE↓	PI↓
CAFRN	35.212	0.964	0.00545	20.327	13.764
DCM	35.468	0.967	0.00533	20.283	13.882
FENet	35.061	0.963	0.00531	20.214	13.829
HAUNet	35.932	0.965	0.00567	20.317	13.736
HSENet	35.588	0.963	0.00571	20.338	13.670
MHAN	35.801	0.970	0.00582	20.295	13.627
Omnisr	35.530	0.965	0.00559	20.366	13.696
RCAN	36.171	0.965	0.00614	20.305	13.487
SAN	35.653	0.966	0.00558	20.260	13.655
RDN	36.636	0.972	0.00594	20.364	13.555

Metric values on original 16-bit images

Method	PSNR↑	SSIM↑	AG↑	NIQE↓	PI↓
CAFRN	50.262	0.986	0.00082	20.605	14.155
DCM	50.388	0.987	0.00086	20.615	14.221
FENet	49.977	0.987	0.00082	20.529	14.112
HAUNet	50.879	0.989	0.00085	20.496	14.029
HSENet	51.748	0.991	0.00081	20.702	14.171
MHAN	51.362	0.991	0.00085	20.765	14.195
Omnisr	50.470	0.984	0.00083	20.551	14.171
RCAN	51.684	0.991	0.00085	20.765	14.180
SAN	50.977	0.989	0.00084	20.705	14.183
RDN	51.877	0.991	0.00086	20.880	14.213

Correlation of coefficient (R²) values of NDVI

Data	Image	HR-LR	CAFRN	DCM	FENet	HAUNet	HSENet	MHAN	Omnisr	RCAN	RDM	SAN
Normalized	1	0.385	0.700	0.695	0.695	0.711	0.703	0.686	0.709	0.718	0.725	0.689
Normalized	2	0.902	0.958	0.958	0.952	0.963	0.962	0.965	0.960	0.968	0.969	0.962
Normalized	3	0.799	0.908	0.913	0.907	0.927	0.906	0.910	0.921	0.932	0.928	0.915
Normalized	4	0.824	0.945	0.941	0.941	0.950	0.948	0.950	0.949	0.957	0.958	0.946
Normalized	5	0.744	0.859	0.863	0.860	0.865	0.866	0.870	0.866	0.844	0.881	0.869
16-bit	1	0.385	0.684	0.672	0.679	0.696	0.695	0.680	0.696	0.692	0.696	0.687
16-bit	2	0.902	0.953	0.955	0.952	0.953	0.956	0.956	0.948	0.963	0.963	0.957
16-bit	3	0.799	0.907	0.902	0.897	0.912	0.908	0.909	0.906	0.914	0.923	0.916
16-bit	4	0.824	0.934	0.935	0.936	0.937	0.944	0.948	0.936	0.946	0.951	0.942
16-bit	5	0.744	0.852	0.854	0.851	0.843	0.861	0.858	0.856	0.871	0.865	0.846

RMSE values of NDVI

Data	Image	HR-LR	CAFRN	DCM	FENet	HAUNet	HSENet	MHAN	Omnisr	RCAN	RDM	SAN
Normalized	1	0.083	0.028	0.028	0.028	0.027	0.028	0.029	0.028	0.027	0.027	0.031
Normalized	2	0.350	0.067	0.066	0.072	0.063	0.064	0.062	0.065	0.058	0.058	0.069
Normalized	3	0.321	0.119	0.118	0.124	0.106	0.121	0.128	0.114	0.094	0.103	0.121
Normalized	4	0.381	0.046	0.046	0.046	0.041	0.042	0.041	0.042	0.038	0.038	0.043
Normalized	5	0.181	0.063	0.062	0.064	0.059	0.060	0.060	0.061	0.068	0.056	0.062
16-bit	1	0.083	0.031	0.032	0.029	0.032	0.028	0.029	0.029	0.031	0.029	0.030
16-bit	2	0.350	0.072	0.070	0.071	0.072	0.070	0.075	0.074	0.071	0.071	0.068
16-bit	3	0.321	0.119	0.127	0.145	0.129	0.131	0.121	0.125	0.130	0.114	0.111
16-bit	4	0.381	0.054	0.061	0.052	0.066	0.043	0.046	0.048	0.043	0.044	0.061
16-bit	5	0.181	0.062	0.069	0.064	0.067	0.062	0.062	0.061	0.066	0.059	0.065

Visual Results

True Color Normalized

True Color 16-bit

False Color Normalized

False Color 16-bit

Citation

Wang, P., Aksoy, S., & Sertel, E. (2026). A Benchmark Dataset for Landsat-to-Sentinel Image Generation Using Deep Learning-Driven Super-Resolution Techniques. Advances in Space Research. https://doi.org/10.1016/j.asr.2026.01.049