Table 2 Quantitative comparison of forecasting performance across different methods on the MRMS and Nha Be datasets.

From: ThoR: A Motion-Dependent Physics-Informed Deep Learning Framework with Constraint-Centric Theory of Functional Connections for Rainfall Nowcasting

Dataset

Method

MSE \((\downarrow )\)

SSIM \((\uparrow )\)

CSI(1) \((\uparrow )\)

CSI(8) \((\uparrow )\)

CSI(16) \((\uparrow )\)

The MRMS Dataset (450, 1000)

PySTEPS8

107.52

0.75

0.402

0.198

0.104

NowcastNet36

42.33

0.88

0.734

0.612

0.297

TrajGRU13

91.23

0.81

0.508

0.312

0.102

PredRNN14

56.81

0.86

0.681

0.605

0.293

DiffCast26

43.10

0.90

0.722

0.604

0.345

ThoR (Ours)

40.24

0.94

0.702

0.653

0.351

The MRMS Dataset (1400, 3000)

PySTEPS8

102.38

0.76

0.387

0.182

0.088

NowcastNet36

39.61

0.89

0.756

0.531

0.305

TrajGRU13

89.33

0.80

0.488

0.294

0.101

PredRNN14

59.22

0.84

0.697

0.558

0.260

DiffCast26

42.17

0.85

0.704

0.560

0.286

ThoR (Ours)

44.25

0.91

0.712

0.575

0.291

The MRMS Dataset (1700, 5500)

PySTEPS8

101.27

0.79

0.352

0.132

0.096

NowcastNet36

35.53

0.88

0.835

0.528

0.198

TrajGRU13

83.61

0.81

0.504

0.230

0.065

PredRNN14

47.40

0.82

0.751

0.535

0.267

DiffCast26

40.12

0.89

0.821

0.539

0.291

ThoR (Ours)

34.12

0.89

0.755

0.556

0.260

The Nha Be Dataset

PySTEPS8

86.19

0.84

0.412

0.179

0.112

NowcastNet36

53.77

0.96

0.811

0.599

0.423

TrajGRU13

85.12

0.83

0.325

0.230

0.107

PredRNN14

55.48

0.83

0.758

0.605

0.377

DiffCast26

55.06

0.90

0.766

0.608

0.392

ThoR (Ours)

49.23

0.91

0.742

0.612

0.461

Back to article page