Contributions of this work and reading guide - Introduction

Abstract

Based on $0.5° \times 0.5°$ 30-year climate normals and land cover data, a simple convolutional neural network is designed to learn how local climate translates into vegetation, which is the ecological meaning of climate. The outputs of an intermediate layer are extracted and processed by PCA to obtain the 15 climate features for the next step of clustering. We analyzed and interpreted the three main features, which are the core concepts through this work. Then, through a simple 322 SOM, the 322 scheme is obtained. Without many details, the 322 scheme is highly structured and easy to understand and interpret at a high level. Next, the detailed 533-5.2 scheme with 26 climate types is developed by 533 SOM and crystal clustering. Each type in this scheme has distinct climatic characteristics. The above describes the backbone of this study.

World climate map of the 322 scheme, 1991-2020 normals. Right-click for full image.

World climate map of the 533-5.2 scheme, 1991-2020 normals. Right-click for full image.

Contributions of this work

• The CNN for climate-vegetation mapping with high accuracy. (CNN structure Sec 1, accuracy Sec 2)
• Clear distinction between climate classification and biome classification by definition of climate features and biome features with outputs from two different NN layers. (Sec 3)
• The three principal axes of the climate space: Damp-coldness (awaiting a better name), Continentality and Seasonality. (Sec 3)
• An in-depth case study on the crystal clustering method based on the 533 climate centroids dataset. (Sec 6)
• Derived climate classification schemes: 322 (Sec 4) and 533-5.2 (Sec 7).
• Analysis of historical climate pattern shifts based on the three features and the 322 scheme. (Sec 5)

Reading guide

This is an interdisciplinary project covering various topics. The Preface is a background introduction to traditional climate classification systems and the high-level idea of this work.

• For readers who focus on the data science concepts and methodology, head to Sec 1, Sec 2 and Sec 6. Also check out this post for the introduction of the crystal clustering method.
• For readers who are interested in the global climate pattern, head to Sec 3, Sec 4 and Sec 7.
• For readers who are concerned about climate change, head to Sec 3, Sec 4 and Sec 5.

Pay attention that there are many images in Sec 4-7, especially in Sec 4 and Sec 7 the images are of large sizes. Take care of data usage.

This is a series of blog posts, so the writing style is generally casual and I avoid complex math representations. I hope you enjoy your journey through the beauty of data science and climate science.