In the spirit of my last post, I ran benchmarks and built prediction models of temperature in all Wallonia. It was a good way for me to be more familiar with mlr. My models have been built from a benchmark run on 2 months. Preparation of the benchmark First, I had to define the target, the learners, the tasks and the resampling strategy. The target is the temperature, and the resampling strategy is LOOCV (Leave-One-Out Cross Validation).

I detailed before in this post why we use mlr in the context of the AGROMET project. Machine learning is very powerful to build models. I will present my methods and my results. Data preparation Reshaping data I have already explained it in my last post but this is an important part of the workflow. mlr package needs a data frame to work correctly. Then, we need to focus on its preparation.

When I began my internship in the CRA-W, I had no idea of what machine learning was. Its weight in the AGROMET project required from me that I learn basis and some applications to apply it in R. That’s why, writing a brief introduction on machine learning and on its utilisation in R seems to be a good idea. What is machine learning ? Machine learning is a subset of deep learning or Artificial Intelligence that provides an ability to “learn” with data.

Solar irradiance is a parameter of interest for building models to predict weather parameters like temperature. That’s why we need to access to these data and prepare them to be included in the models. Get data To obtain solar irradiance data from Wallonia, those have been included in the AGROMET API from EUMETSAT website. Data that we have are DSSF (Downward Surface Shortwave Flux) with a frequency of 1 day (MDSSF) or 30 minutes (DIDSSF) and measured in W/m².

In the context of the AGROMET project, we need CORINE land cover data and digital terrain model to improve models. I refer you to my other posts about CORINE land cover (CLC) : How to get CLC data from Wallonia Solve a problem in computing the area of each polygon inside a buffer This post deals with the methodology which was applied to prepare data before using them in our models.

In the context of the AGROMET project, we searched to include data from CORINE land cover in our models. I refer you to my post How to get CORINE land cover data from Wallonia where I explained how we got data and how we reclass land covers in groups. During my work, I wanted to calculate the part of each land cover around every station and I am arrived at a problem.

The objective of the AGROMET project is to provide hourly 1km² gridded datasets of weather parameters with the best accuracy (i.e. spatialize hourly records from the stations on the whole area of Wallonia). Weather parameters will be predicted with explanatory variables. These explanatory variables are : Digital elevation model and its derivatives like aspect and slope Solar irradiance

In the context of my internship, my work mainly consists in development. However, working in development often imposes to be methodic. That’s why, when I have began my internship, I installed my working environment to be more productive. Installing Linux First, I have installed Ubuntu GNOME, a distribution of Linux on my laptop. Indeed, this OS is prefered by developers and open-source addicts thanks to the high contribution to improve distributions.

One of the objectives of the AGROMET project is to make maps to view spatialized data. That’s why I tried to create a template map to view the different tests of spatialization. To do that, I used data from the AGROMET API using methods I have already detailed before. Get map of Wallonia Source : Raster data acquisition We need an outline map to limit the map to Wallonia. We can get the font of Wallonia with getData() function.

The API has 20 years of records of weather data. When you need some data, you have to use the API to load them. For my first exercise, I had to give an usable file with data about the relative humidity throughout the year near to Frasnes-lez-Anvaing. Find the nearest station Fisrt, I checked where was Frasnes-lez-Anvaing on the platform to find the nearest station.