Cesar Aybar

Geographical Engineering

UNMSM

Hi, Cesar here. Welcome to my personal website!

I am a Peruvian data analyst, with knowledge in web development and DevOps, that love to read about computer vision and deep learning. Previously, I held academic positions at National Weather Service and Hydrology of Peru (SENAMHI) and San Marcos University (UNMSM).

I am currently writing a paper that proposes the use of attention based Convolution Neural Networks for identifying sources of errors caused by humans in manned meteorological stations. Past projects include the development of a high-resolution gridded rainfall dataset for Peru and the generation of indices and maps to understand better the climate and extreme events from the Central Altiplano of Peru and Bolivia. I have worked on the development of a number of R/Python packages for automatizing theoretical variograms fitting, improving the control quality of rain gauges, and detecting deforestation. I am the author of rgee, a binding library that permits you to use Google Earth Engine from R.

I have taught Introduction to Algorithms a couple of times at undergraduate level at San Marcos University, Peru, and I have been invited to deliver talks and training courses on geospatial analysis and geostatistic at international workshops. I have also created online educational materials most of them to learn how to use Google Earth Engine with Python.

I received my Bachelor's in Geographical Engineering from the University of San Marcos, Peru.

Algorithms & Maps

Make a GeoViz or learn how Geospatial algorithms really work, it probably gives you tonnes of fun, isn’t that right?. If this isn’t your case yet, remember that Geospatial thinking has positioned itself as an essential skill for solving problems in industry and academia, so, learn some geo-stuff will boost definitely your CV. There is a lot to talk about it, but, in this blog, we want to target to:

- Reproducible examples of Deep Learning and Computer Vision.
- State-of-the-art algorithms clearly explained.
- Tips about good practices in coding (most of the time Python and R).
- And much more!

I promise to post a new entrance every first week of each month (with probability 0.4).

Download the GLC30 product for Peru and Ecuador

By Cesar Aybar on 2019-10-10

Land Cover information is fundamental for environmental change studies, land resource management, sustainable development, and many other societal benefits. In Google Earth Engine existing different coarse Land use/cover products (e.g. MCD12Q1-500m) that are not able to capture the most significant human impacts on land systems due to its spatial resolution. Therefore, this short post aims to introduce and explain step by step how to download the GLC30 a relative new Global Land Cover product at 30-meter spatial resolution.

Continue reading

Integrating Earth Engine with Tensorflow II - U-Net

By Cesar Aybar on 2019-06-21

This notebook has been inspired by the Chris Brown & Nick Clinton EarthEngine + Tensorflow presentation. It shows the step by step how to integrate Google Earth Engine and TensorFlow 2.0 in the same pipeline (EE->Tensorflow->EE). OBS: I will assume reader are already familiar with the basic concepts of Machine Learning and Convolutional Networks. If it is doesn’t, I firstly highly recommend taking the deep learning coursera specialization available here.

Continue reading

Integrating Earth Engine with Tensorflow I - DNN

By Cesar Aybar on 2019-05-30

This notebook has been inspired by the Chris Brown & Nick Clinton EarthEngine + Tensorflow presentation. It shows the step by step how to integrate Google Earth Engine and TensorFlow 2.0 in the same pipeline (EE->Tensorflow->EE). Topics Create a training/testing dataset (in a TFRecord format) using Earth Engine. Create functions for parse data (TFRecord -> tf.

Continue reading

Head Start Data Science I: Titanic Challenge

By Cesar Aybar on 2019-05-19

1. Introduction The Titanic challenge is an excellent way to practice the necessary skills required for ML. In my first attempts, I blindly applied a well-known ML method (Lightgbm); however, I couldn’t go up over the Top 20% :(. To have success in this competition you need to realize an acute feature engineering that takes into account the distribution on train and test dataset. This post is the perfect opportunity to share with you my Python package preml and show how can you BEAT THE 97% OF LB.

Continue reading