#!/usr/bin/env python
# coding: utf-8

# # DeepCube Use Case 3: Fire Hazard Forecasting in the Mediterranean

#  ![use case3 icon](static/uc3-logo.png) 

# # Notebook 1: Datacube access and basic analytics

# This notebook shows basic utilities to load and process the [greece_wildfire_datacube](https://zenodo.org/deposit/4943354) with [xarray](http://xarray.pydata.org/en/stable/). 
# 
# It contains basic instructions showing how to access, process and visualize the data.

# ## Initial Imports
# 
# We start by importing the libraries we need. Make sure to have them installed in your system. 

# In[1]:


import xarray as xr
import fsspec
import zarr

import numpy as np
import pandas as pd

import csv
import numpy as np
import matplotlib.pyplot as plt

import random
import os
from tqdm import tqdm
import gc
from pathlib import Path


# ## Download (optional) and access Datacube
# 
# We either download the datacube from [Zenodo](https://zenodo.org/record/4943354) in the form of a single netcdf file named `dataset_greece.nc`, or directly use the cloud-hosted version  of the datacube (access will be slower).
# 
# Then we open the datacube with xarray.

# In[2]:


# First method (Download and access)
# !wget -O dataset_greece.nc https://zenodo.org/record/4943354/files/dataset_greece.nc?download=1
# ds = xr.open_dataset('./dataset_greece.nc')

# Second method (Use cloud-hosted datacube instance)
url = 'https://storage.de.cloud.ovh.net/v1/AUTH_84d6da8e37fe4bb5aea18902da8c1170/uc3/uc3cube.zarr'
ds = xr.open_zarr(fsspec.get_mapper(url), consolidated=True)


# In[3]:


ds['1 km 16 days NDVI'][0].plot()


# In[4]:


ds


# We see that the dataset has x,y and time dimensions.
# 
# More specifically it contains 4314 days (from 06/03/2009 to 06/12/2020) of 700x562 rasters
# 
# Dynamic variables like the `burned_areas` have all three dimensions, while static variables like `clc_2012` misses the temporal component and only has x, y dimensions.

# In[5]:


import random
random.sample(list(range(10)), 10)


# ## Plotting and Analytics

# In this section, we'll see some basic functionality showing how to select and visualize the variables in the datacube.

# Corine Land Cover 2012 (clc_2012) is a static dataset, containing Land Cover values . Let's see what it looks like

# In[6]:


ds['clc_2012'].plot(cmap='Spectral', vmin=0, vmax=44, figsize=(10,8))


# Maximum Temperature for a specific period of time for a specific pixel (300x300) in the dataset. We can plot both the time-series of the temperature and a histogram of the values

# In[7]:


fig, axes = plt.subplots(ncols=2, figsize=(10,4))
(ds['era5_max_t2m']).isel(x=300, y=300).plot(ax=axes[0])
(ds['era5_max_t2m']).isel(x=300, y=300).plot.hist(ax=axes[1])
plt.tight_layout()
plt.draw()


# The same but only for year 2019

# In[8]:


fig, axes = plt.subplots(ncols=2, figsize=(10,4))
ds['era5_max_t2m'].sel(time=slice("2019-01-01", "2019-12-31")).isel(x=300, y=300).plot(ax=axes[0])
ds['era5_max_t2m'].sel(time=slice("2019-01-01", "2019-12-31")).isel(x=300, y=300).plot.hist(ax=axes[1])
plt.show()


# A 10-rolling mean of the NDVI for a specific period of time and a specific pixel

# In[9]:


ds['1 km 16 days NDVI'].isel(dict(time=slice(4000, 4200), x=300, y=300)).rolling(time=10, center=True).mean().plot()


# Plots of land surface temperature for a series of days for the whole Greece region

# In[10]:


ds['LST_Day_1km'].isel(time=slice(4157, 4166)).plot(x="x", y="y", col="time", col_wrap=3)


# We can plot the fire weather index for the same days

# In[11]:


ds['fwi'].isel(time=slice(4157, 4166)).plot(x="x", y="y", col="time", col_wrap=3)


# Or the burned areas from a fire that started those days

# In[12]:


ds['burned_areas'].isel(time=slice(4157, 4166)).plot(x="x", y="y", col="time", col_wrap=3)


# Just a few pixels of burned areas and a complex interaction of the fire drivers. It is going to be quite challenging to model that... Stay tuned for our first models in the next notebooks that we will publish.

# # Discussion

# We saw how easy it is to access, process and visualize data in the datacube, even if the actual data is very large and cannot be loaded in memory.
# 
# In a later notebook we'll see how to develop some ML models using our datacube to predict the burnt area from next day's fire. Stay tuned!

# # Acknowledgements
# 
# Research funded by the EU H2020 project DeepCube ’Explainable AI pipelines for big Copernicus data’, grant agreement No 101004188.

# Notebook Authors: *Ioannis Prapas (https://github.com/iprapas), Spyros Kondylatos (https://github.com/skondylatos)*