Add files via upload

README.txt

@@ -0,0 +1,10 @@
+The sample code 'SNOWNET_sample_code.py' demonstartes how to use the AMSR2-only ('snownet.h5') 
+and the AMSR2+SMOS ('snownet_smos.h5') neural networks for snow depth estimation.
+
+To run the code you need a few packages installed including Tensorflow and Keras.
+To make sure the model can be loaded and to reproduce the same results as presented 
+in the paper, please install Tensorflow version 1.10.0 and Keras version 2.2.2.
+
+When using the neural networks or the code, please cite:
+'A. Braakmann-Folgmann and C. Donlon: Estimating Snow Depth on Arctic Sea Ice using 
+Satellite Microwave Radiometry and a Neural Network, The Cryosphere (Discussions), (2019)'

SNOWNET_sample_code.py

@@ -0,0 +1,142 @@
+ # -*- coding: utf-8 -*-
+"""
+January 2019
+
+This code demonstartes how to use the AMSR2-only ('snownet.h5') 
+and the AMSR2+SMOS ('snownet_smos.h5') neural networks for snow depth estimation.
+
+As input the brightness temperatures at 6.9V, 18.7V and 18.7H, 36.5V and 36.5H
+are required (18.7H is only needed for the Nasa Team algorithm to do an open water 
+correction). The AMSR2+SMOS neural network additionally relies on brightness 
+temperatures at 1.4V and 1.4H at 55deg incidence angle. All the input brightness 
+temperatures need to be collocated.
+
+The neural network is loaded with model=load_model('snownet.h5')
+and snow depth is calculated with snow_depth = model.predict(inputs).
+inputs has to be a samples x 3 matrix for the AMSR2-only and a samples x 4 matrix
+for the AMSR2+SMOS neural network. The first column is always the 37/19V gradient 
+ratio, the second column has to be the 19/7V gradient ratio, the third one the 
+polarization ratio at 37GHz. If the AMSR2+SMOS neural network is used the fourth
+column stores the gradient ratio at 1.4GHz.
+
+The snow depth output is always in meters.
+
+To run the code you need a few packages installed including Tensorflow and Keras.
+To make sure the model can be loaded and to reproduce the same results as presented 
+in the paper, please install Tensorflow version 1.10.0 and Keras version 2.2.2.
+
+When using the neural networks or this code, please cite:
+'A. Braakmann-Folgmann and C. Donlon: Estimating Snow Depth on Arctic Sea Ice 
+using Satellite Microwave Radiometry and a Neural Network, The Cryosphere 
+(Discussions), (2019)'
+
+@author: Anne Braakmann-Folgmann
+
+"""
+from IPython import get_ipython
+get_ipython().magic('reset -sf')
+
+import numpy as np
+# =============================================================================
+# you should install Tensorflow 1.10.0 and Keras 2.2.2 to reproduce the same results
+# =============================================================================
+from keras.models import Sequential, load_model
+from keras.layers import Dense, BatchNormalization
+
+
+#==============================================================================
+# generate AMSR2-like sample data
+#==============================================================================
+Tb6V=np.array([248.3, 245.44, 247.02, 245.02, 245.97, 245.02, 246.62, 245.86, 245.64, 246.64])
+Tb18H=np.array([209.45, 206.67, 208.47, 205.19, 206.92, 205.19, 208.86, 207.19, 206.61, 208.78])
+Tb18V=np.array([232.27, 230.65, 232.52, 229.04, 232.64, 229.04, 229.73, 231.26, 229.06, 229.53])
+Tb36H=np.array([191.38,192.49, 192.81, 189.02, 193.87, 189.02, 191.59, 192.86, 189.5, 191.2])
+Tb36V=np.array([207.07, 209.16, 209.51, 205.56, 212.2, 205.56, 206.3, 209.96, 205.22, 205.84])
+
+
+# =============================================================================
+# generate SMOS-like sample data
+# =============================================================================
+Tb1H=np.array([231.77, 233.74, 234.88, 228.06, 229.03, 222.89, 222.84, 223.26, 219.59, 218.55])
+Tb1V=np.array([261.13, 256.50, 257.23, 255.97, 254.17, 252.88, 252.88, 254.19, 251.27, 252.06])
+
+
+# =============================================================================
+# Calculate the gradient and polarization ratios
+# =============================================================================
+
+# SIC (can be calculated from the Nasa Team algorithm)
+SIC=np.array([0.93307081, 0.90422626, 0.91017262, 0.8984863, 0.88024428, 0.9106783, 
+     0.95138806,  0.90307585, 0.930586, 0.95366877])
+
+# Open water tie points from Ivanova et al. (2014)
+OW_6V=161.35
+OW_18H=108.46
+OW_18V=183.72
+OW_36H=145.29
+OW_36V=209.81
+
+# calculate brightness Temperature of ice part within footprint
+Tbice6V= np.divide((Tb6V - np.multiply((1-SIC), OW_6V) ), SIC)
+Tbice18V= np.divide((Tb18V - np.multiply((1-SIC), OW_18V) ), SIC)
+Tbice36H= np.divide((Tb36H - np.multiply((1-SIC), OW_36H) ), SIC)
+Tbice36V= np.divide((Tb36V - np.multiply((1-SIC), OW_36V) ), SIC)
+
+# gradient ratios
+GR3618V = np.divide((Tbice36V - Tbice18V), (Tbice36V + Tbice18V))
+GR186V = np.divide((Tbice18V- Tbice6V), (Tbice18V + Tbice6V))
+
+# polarisation ratios
+PR36=np.divide((Tbice36V-Tbice36H),(Tbice36V+Tbice36H)) 
+PR1=np.divide((Tb1V-Tb1H),(Tb1V+Tb1H)) 
+
+#==============================================================================
+# prepare the input data 
+#==============================================================================
+input_data=np.concatenate((np.expand_dims(GR3618V, axis=1), np.expand_dims(GR186V, axis=1), np.expand_dims(PR36, axis=1)), axis=1)
+
+input_data_smos=np.concatenate((input_data, np.expand_dims(PR1, axis=1)), axis=1)
+
+# =============================================================================
+#  load the neural network
+# =============================================================================
+# try to load the neural networks like this: 
+try: 
+    model=load_model('snownet.h5')
+    model_smos=load_model('snownet_smos.h5')
+
+# if an error occurs, you have a different version of Keras or Tensorflow installed. 
+# Try to set up the design yourself and only load the weights
+# this may however lead to slightly different results
+except:
+    print('You have a different version of Tensorflow or Keras installed. This may lead to slightly different results.')
+    model = Sequential()
+    model.add(Dense(15, input_shape=(3,), activation='sigmoid'))
+    model.add(BatchNormalization())
+    model.add(Dense(15, activation='relu'))
+    model.add(Dense(15, activation='relu'))
+    model.add(Dense(15, activation='relu'))
+    model.add(Dense(20, activation='relu'))
+    model.add(Dense(1, activation='tanh'))
+
+    model.compile(loss='mean_absolute_percentage_error', optimizer='adam')
+    model.load_weights('snownet_weights.h5')
+
+    model_smos = Sequential()
+    model_smos.add(Dense(15, input_shape=(4,), activation='sigmoid'))
+    model_smos.add(BatchNormalization())
+    model_smos.add(Dense(15, activation='relu'))
+    model_smos.add(Dense(15, activation='relu'))
+    model_smos.add(Dense(20, activation='relu'))
+    model_smos.add(Dense(1, activation='tanh'))
+
+    model_smos.compile(loss='mean_absolute_percentage_error', optimizer='adam')
+    model_smos.load_weights('snownet_smos_weights.h5')
+
+
+#==============================================================================
+# make predictions
+#==============================================================================
+snow_depth_AMSR2_only = model.predict(input_data)
+snow_depth_AMSR2_SMOS = model_smos.predict(input_data_smos)
+