Add FAPAR
This commit is contained in:
parent
3770c4dc2f
commit
b92c7be4dc
185
sat.py
185
sat.py
|
@ -1,7 +1,8 @@
|
||||||
from PIL import Image
|
from PIL import Image
|
||||||
import requests
|
import requests
|
||||||
from sat7_pointer import *
|
from sat7_pointer import *
|
||||||
from numpy import np
|
import numpy as np
|
||||||
|
from tqdm import tqdm
|
||||||
|
|
||||||
# Load Landsat 7 band 1, 2, 3 TIF images and create a composite image
|
# Load Landsat 7 band 1, 2, 3 TIF images and create a composite image
|
||||||
# from the three bands.
|
# from the three bands.
|
||||||
|
@ -14,17 +15,17 @@ from numpy import np
|
||||||
#
|
#
|
||||||
# The composite image is saved as a PNG file.
|
# The composite image is saved as a PNG file.
|
||||||
|
|
||||||
band1 = Image.open('LE07_L1TP_177025_20210723_20210818_02_T1_B1.TIF')
|
band1 = Image.open("LE07_L1TP_177025_20210723_20210818_02_T1_B1.TIF")
|
||||||
band2 = Image.open('LE07_L1TP_177025_20210723_20210818_02_T1_B2.TIF')
|
band2 = Image.open("LE07_L1TP_177025_20210723_20210818_02_T1_B2.TIF")
|
||||||
band3 = Image.open('LE07_L1TP_177025_20210723_20210818_02_T1_B3.TIF')
|
band3 = Image.open("LE07_L1TP_177025_20210723_20210818_02_T1_B3.TIF")
|
||||||
|
|
||||||
composite = Image.merge('RGB', (band3, band2, band1))
|
composite = Image.merge("RGB", (band3, band2, band1))
|
||||||
|
|
||||||
# Load corner coordinates of the image.
|
# Load corner coordinates of the image.
|
||||||
#
|
#
|
||||||
# The coordinates are stored in a MTL text file.
|
# The coordinates are stored in a MTL text file.
|
||||||
|
|
||||||
mtl_data = load_metadata('LE07_L1TP_177025_20210723_20210818_02_T1_MTL.txt')
|
mtl_data = load_metadata("LE07_L1TP_177025_20210723_20210818_02_T1_MTL.txt")
|
||||||
|
|
||||||
# Fetch coordinates of a city.
|
# Fetch coordinates of a city.
|
||||||
#
|
#
|
||||||
|
@ -35,14 +36,18 @@ mtl_data = load_metadata('LE07_L1TP_177025_20210723_20210818_02_T1_MTL.txt')
|
||||||
#
|
#
|
||||||
# # City is Belgorod, Russia.
|
# # City is Belgorod, Russia.
|
||||||
|
|
||||||
url = 'http://nominatim.openstreetmap.org/search?q=Belgorod,+Russia&format=json'
|
url = "http://nominatim.openstreetmap.org/search?q=Belgorod,+Russia&format=json"
|
||||||
response = requests.get(url)
|
response = requests.get(url)
|
||||||
data = response.json()
|
data = response.json()
|
||||||
|
|
||||||
# Convert bounding box coordinates to image coordinates.
|
# Convert bounding box coordinates to image coordinates.
|
||||||
|
|
||||||
x0, y1 = lat_lot_to_pixel(data[0]['boundingbox'][0], data[0]['boundingbox'][2], mtl_data)
|
x0, y1 = lat_lot_to_pixel(
|
||||||
x1, y0 = lat_lot_to_pixel(data[0]['boundingbox'][1], data[0]['boundingbox'][3], mtl_data)
|
data[0]["boundingbox"][0], data[0]["boundingbox"][2], mtl_data
|
||||||
|
)
|
||||||
|
x1, y0 = lat_lot_to_pixel(
|
||||||
|
data[0]["boundingbox"][1], data[0]["boundingbox"][3], mtl_data
|
||||||
|
)
|
||||||
|
|
||||||
print(x0, y0)
|
print(x0, y0)
|
||||||
print(x1, y1)
|
print(x1, y1)
|
||||||
|
@ -51,7 +56,7 @@ print(composite.size)
|
||||||
|
|
||||||
cropped = composite.crop((x0, y0, x1, y1))
|
cropped = composite.crop((x0, y0, x1, y1))
|
||||||
|
|
||||||
cropped.save('cropped.png')
|
cropped.save("cropped.png")
|
||||||
|
|
||||||
###############################################################################
|
###############################################################################
|
||||||
# LAB 2
|
# LAB 2
|
||||||
|
@ -60,7 +65,7 @@ cropped.save('cropped.png')
|
||||||
# Load landsat band 4 TIF image.
|
# Load landsat band 4 TIF image.
|
||||||
# Band 4 is near infrared.
|
# Band 4 is near infrared.
|
||||||
|
|
||||||
band4 = Image.open('LE07_L1TP_177025_20210723_20210818_02_T1_B4.TIF')
|
band4 = Image.open("LE07_L1TP_177025_20210723_20210818_02_T1_B4.TIF")
|
||||||
|
|
||||||
# Claculate the NDVI.
|
# Claculate the NDVI.
|
||||||
#
|
#
|
||||||
|
@ -116,7 +121,8 @@ def get_color(value):
|
||||||
else:
|
else:
|
||||||
return (0, 0, 0)
|
return (0, 0, 0)
|
||||||
|
|
||||||
for x in range(ndvi.size[0]):
|
|
||||||
|
for x in tqdm(range(ndvi.size[0]), desc="NDVI"):
|
||||||
for y in range(ndvi.size[1]):
|
for y in range(ndvi.size[1]):
|
||||||
r = red.getpixel((x, y))
|
r = red.getpixel((x, y))
|
||||||
nir = ndvi.getpixel((x, y))
|
nir = ndvi.getpixel((x, y))
|
||||||
|
@ -125,7 +131,7 @@ for x in range(ndvi.size[0]):
|
||||||
else:
|
else:
|
||||||
result.putpixel((x, y), get_color((nir - r) / (nir + r)))
|
result.putpixel((x, y), get_color((nir - r) / (nir + r)))
|
||||||
|
|
||||||
result.save('ndvi.png')
|
result.save("ndvi.png")
|
||||||
|
|
||||||
|
|
||||||
# Calculate FAPAR (Fraction of Absorbed Photosynthetically Active Radiation)
|
# Calculate FAPAR (Fraction of Absorbed Photosynthetically Active Radiation)
|
||||||
|
@ -133,28 +139,141 @@ result.save('ndvi.png')
|
||||||
#
|
#
|
||||||
# Bands 1, 3, 4 are used.
|
# Bands 1, 3, 4 are used.
|
||||||
|
|
||||||
# Coefficients of polynominal g_0
|
solar_zenith_angle = np.radians(float(mtl_data["SUN_ELEVATION"]))
|
||||||
# l0,1 l0,2 l0,3 l0,4 l0,5 l0,6
|
sensor_zenith_angle = np.radians(0)
|
||||||
# 0.27505 0.35511 −0.004 −0.322 0.299 −0.0131
|
sun_sensor_relative_azimuth = np.radians(float(mtl_data["SUN_AZIMUTH"]))
|
||||||
l_0 = np.array([0.27505, 0.35511, -0.004, -0.322, 0.299, -0.0131])
|
|
||||||
|
|
||||||
# Coefficients of polynominal g_1
|
gain = [float(mtl_data["RADIANCE_MULT_BAND_" + str(i)]) for i in [1, 3, 4]]
|
||||||
# l1,1 l1,2 l1,3 l1,4 l1,5 l1,6 l1,7 l1,8 l1,9 l1,10 l1,11
|
offset = [float(mtl_data["RADIANCE_ADD_BAND_" + str(i)]) for i in [1, 3, 4]]
|
||||||
# −10.036 −0.019804 0.55438 0.14108 12.494 0 0 0 0 0 1.0
|
|
||||||
l_1 = np.array([-10.036, -0.019804, 0.55438, 0.14108, 12.494, 0, 0, 0, 0, 0, 1.0])
|
|
||||||
|
|
||||||
# Coefficients of polynominal g_2
|
dsol = float(mtl_data["EARTH_SUN_DISTANCE"])
|
||||||
# l2,1 l2,2 l2,3 l2,4 l2,5 l2,6 l2,7 l2,8 l2,9 l2,10
|
|
||||||
# 0.42720 0.069884 −0.33771 0.24690 −1.0821 −0.30401 −1.1024 −1.2596 −0.31949 −1.4864
|
|
||||||
l_2 = np.array([0.42720, 0.069884, -0.33771, 0.24690, -1.0821, -0.30401, -1.1024, -1.2596, -0.31949, -1.4864])
|
|
||||||
|
|
||||||
l = [l_0, l_1, l_2]
|
pic = [0.643, 0.80760, 0.89472]
|
||||||
|
k = [0.76611, 0.63931, 0.81037]
|
||||||
|
theta = [-0.10055, -0.06156, -0.03924]
|
||||||
|
|
||||||
# Function to calculate ratio of polynominals
|
k = [0.63931,0.81037, 0.76611]
|
||||||
def g_n(n, x, y):
|
pic = [0.80760, 0.89472, 0.643]
|
||||||
if n == 0:
|
theta = [-0.06156, -0.03924, -0.10055]
|
||||||
return l_0[0] * y - l_0[1] * x + l_0[2] / (l_0[3] - x)**2 + (l_0[4] - y)**2 + l_0[5]
|
|
||||||
if n == 1:
|
|
||||||
return l_1[0]
|
|
||||||
|
|
||||||
l[n][0] * (x + l[n][1])**2 + l[n][2] * (y + l[n][3])**2 + l[n][4] * x * y / l[n][5] * (x + l[n][6])**2 + l
|
E0 = [1969, 1551, 1044]
|
||||||
|
|
||||||
|
cosg = np.cos(solar_zenith_angle) * np.cos(sensor_zenith_angle) + np.sin(
|
||||||
|
solar_zenith_angle
|
||||||
|
) * np.sin(sensor_zenith_angle) * np.cos(sun_sensor_relative_azimuth)
|
||||||
|
G = (
|
||||||
|
np.tan(solar_zenith_angle) ** 2
|
||||||
|
+ np.tan(sensor_zenith_angle) ** 2
|
||||||
|
- 2
|
||||||
|
* np.tan(solar_zenith_angle)
|
||||||
|
* np.tan(sensor_zenith_angle)
|
||||||
|
* np.cos(sun_sensor_relative_azimuth)
|
||||||
|
) ** 0.5
|
||||||
|
|
||||||
|
|
||||||
|
polynoms = np.array(
|
||||||
|
[
|
||||||
|
[0.27505, 0.35511, -0.004, -0.322, 0.299, -0.0131, 0, 0, 0, 0, 0],
|
||||||
|
[-10.036, -0.019804, 0.55438, 0.14108, 12.494, 0, 0, 0, 0, 0, 1],
|
||||||
|
[
|
||||||
|
0.42720,
|
||||||
|
0.069884,
|
||||||
|
-0.33771,
|
||||||
|
0.24690,
|
||||||
|
-1.0821,
|
||||||
|
-0.30401,
|
||||||
|
-1.1024,
|
||||||
|
-1.2596,
|
||||||
|
-0.31949,
|
||||||
|
-1.4864,
|
||||||
|
0,
|
||||||
|
],
|
||||||
|
]
|
||||||
|
)
|
||||||
|
|
||||||
|
blue = band1.copy().crop((x0, y0, x1, y1))
|
||||||
|
red = band3.copy().crop((x0, y0, x1, y1))
|
||||||
|
nir = band4.copy().crop((x0, y0, x1, y1))
|
||||||
|
result = cropped.copy()
|
||||||
|
|
||||||
|
f1 = [
|
||||||
|
((np.cos(solar_zenith_angle) * np.cos(sensor_zenith_angle)) ** (k[i] - 1))
|
||||||
|
/ (np.cos(solar_zenith_angle) + np.cos(sensor_zenith_angle)) ** (1 - k[i])
|
||||||
|
for i in range(3)
|
||||||
|
]
|
||||||
|
f2 = [
|
||||||
|
(1 - theta[i] ** 2) / (1 + 2 * theta[i] * cosg + theta[i] ** 2) ** (3 / 2)
|
||||||
|
for i in range(3)
|
||||||
|
]
|
||||||
|
f3 = [1 + (1 - pic[i]) / (1 + G) for i in range(3)]
|
||||||
|
F = [f1[i] * f2[i] * f3[i] for i in range(3)]
|
||||||
|
|
||||||
|
def get_color_fapar(value, rho):
|
||||||
|
if (0 < rho[0] and rho[0] < 0.257752) \
|
||||||
|
and (0 < rho[1] and rho[1] < 0.48407) \
|
||||||
|
and (0 < rho[2] and rho[2] < 0.683928) \
|
||||||
|
and (rho[0] <= rho[2]) \
|
||||||
|
and (rho[2] >= 1.26826*rho[1]):
|
||||||
|
return get_color(value)
|
||||||
|
if (rho[0] <= 0) or (rho[1] <= 0) or (rho[2] <= 0):
|
||||||
|
return (0, 0, 0)
|
||||||
|
if (rho[0] >= 0.257752) or (rho[1] >= 0.48407) or (rho[2] >= 0.683928):
|
||||||
|
return (255, 255, 255)
|
||||||
|
if (0 < rho[0] and rho[0] < 0.257752) \
|
||||||
|
and (0 < rho[1] and rho[1] < 0.48407) \
|
||||||
|
and (0 < rho[2] and rho[2] < 0.683928) \
|
||||||
|
and (rho[0] >= rho[2]):
|
||||||
|
return (0, 0, 255)
|
||||||
|
if (0 < rho[0] and rho[0] < 0.257752) \
|
||||||
|
and (0 < rho[1] and rho[1] < 0.48407) \
|
||||||
|
and (0 < rho[2] and rho[2] < 0.683928) \
|
||||||
|
and (rho[0] <= rho[2]) \
|
||||||
|
and (1.25*rho[1] > rho[2]):
|
||||||
|
return (255, 150, 150)
|
||||||
|
if (rho[1] < 0) or (rho[2] < 0):
|
||||||
|
return (0, 0, 0)
|
||||||
|
if value < 0 or value > 1:
|
||||||
|
return (0, 0, 0)
|
||||||
|
|
||||||
|
return (int(180.0 * (1 - value)), 255, 255)
|
||||||
|
|
||||||
|
for x in tqdm(range(result.size[0]), desc="FAPAR"):
|
||||||
|
for y in range(result.size[1]):
|
||||||
|
bands = [blue.getpixel((x, y)), red.getpixel((x, y)), nir.getpixel((x, y))]
|
||||||
|
|
||||||
|
rho_i = [
|
||||||
|
(
|
||||||
|
(np.pi * (gain[i] * bands[i] + offset[i]) * dsol ** 2)
|
||||||
|
/ (E0[i] * np.cos(sensor_zenith_angle))
|
||||||
|
)
|
||||||
|
/ F[i]
|
||||||
|
for i in range(3)
|
||||||
|
]
|
||||||
|
|
||||||
|
g1 = (
|
||||||
|
(polynoms[1, 0] * (rho_i[0] + polynoms[1, 1]) ** 2)
|
||||||
|
+ (polynoms[1, 2] * (rho_i[1] + polynoms[1, 3]) ** 2)
|
||||||
|
+ polynoms[1, 4] * rho_i[0] * rho_i[1]
|
||||||
|
) / (
|
||||||
|
polynoms[1, 5] * (rho_i[0] + polynoms[1, 6]) ** 2
|
||||||
|
+ polynoms[1, 7] * (rho_i[1] + polynoms[1, 8]) ** 2
|
||||||
|
+ polynoms[1, 9] * rho_i[0] * rho_i[1]
|
||||||
|
+ polynoms[1, 10]
|
||||||
|
)
|
||||||
|
g2 = (
|
||||||
|
(polynoms[2, 0] * (rho_i[0] + polynoms[2, 1]) ** 2)
|
||||||
|
+ (polynoms[2, 2] * (rho_i[2] + polynoms[2, 3]) ** 2)
|
||||||
|
+ polynoms[2, 4] * rho_i[0] * rho_i[2]
|
||||||
|
) / (
|
||||||
|
polynoms[2, 5] * (rho_i[0] + polynoms[2, 6]) ** 2
|
||||||
|
+ polynoms[2, 7] * (rho_i[2] + polynoms[2, 8]) ** 2
|
||||||
|
+ polynoms[2, 9] * rho_i[0] * rho_i[2]
|
||||||
|
+ polynoms[2, 10]
|
||||||
|
)
|
||||||
|
FAPAR = ((polynoms[0, 0] * g2) - polynoms[0, 1] * g1 - polynoms[0, 2]) / (
|
||||||
|
(polynoms[0, 3] - g1) ** 2 + (polynoms[0, 4] - g2) ** 2 + polynoms[0, 5]
|
||||||
|
)
|
||||||
|
|
||||||
|
result.putpixel((x, y), get_color_fapar(FAPAR, rho_i))
|
||||||
|
|
||||||
|
result.save('fapar.png')
|
||||||
|
|
Loading…
Reference in a new issue