Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Maximum Temperature Prediction Using Remote Sensing Data Via Convolutional Neural Network

Lorenzo Innocenti, Giacomo Blanco, Luca Barco, Claudio Rossi

TL;DR

The paper tackles urban heat island challenges by predicting high-resolution, city-scale maximum daily temperatures for Turin using a CNN-based image-to-image regression that fuses Sentinel-3 satellite bands, meteorological indicators, and static topographic/land-use data. It formulates the task as per-pixel temperature regression I_t → Ť_t, employing an MSE loss on valid pixels and temporal encodings to capture daily and diurnal patterns. The authors compare shallow linear, ResNet-50, and ConvNext-tiny U-Net–style architectures, finding ConvNext to deliver the lowest MAE (2.09 $^ ext{o}$C) at 20 m/px, with consistent gains over coarser resolutions and qualitative alignment to urban features. This work demonstrates the feasibility of dense, multi-source urban climate mapping, offering a practically impactful tool for UHIs assessment and policy planning, and points to extensions via multistep forecasts and additional inputs such as NDVI and population density.

Abstract

Urban heat islands, defined as specific zones exhibiting substantially higher temperatures than their immediate environs, pose significant threats to environmental sustainability and public health. This study introduces a novel machine-learning model that amalgamates data from the Sentinel-3 satellite, meteorological predictions, and additional remote sensing inputs. The primary aim is to generate detailed spatiotemporal maps that forecast the peak temperatures within a 24-hour period in Turin. Experimental results validate the model's proficiency in predicting temperature patterns, achieving a Mean Absolute Error (MAE) of 2.09 degrees Celsius for the year 2023 at a resolution of 20 meters per pixel, thereby enriching our knowledge of urban climatic behavior. This investigation enhances the understanding of urban microclimates, emphasizing the importance of cross-disciplinary data integration, and laying the groundwork for informed policy-making aimed at alleviating the negative impacts of extreme urban temperatures.

Maximum Temperature Prediction Using Remote Sensing Data Via Convolutional Neural Network

TL;DR

The paper tackles urban heat island challenges by predicting high-resolution, city-scale maximum daily temperatures for Turin using a CNN-based image-to-image regression that fuses Sentinel-3 satellite bands, meteorological indicators, and static topographic/land-use data. It formulates the task as per-pixel temperature regression I_t → Ť_t, employing an MSE loss on valid pixels and temporal encodings to capture daily and diurnal patterns. The authors compare shallow linear, ResNet-50, and ConvNext-tiny U-Net–style architectures, finding ConvNext to deliver the lowest MAE (2.09 C) at 20 m/px, with consistent gains over coarser resolutions and qualitative alignment to urban features. This work demonstrates the feasibility of dense, multi-source urban climate mapping, offering a practically impactful tool for UHIs assessment and policy planning, and points to extensions via multistep forecasts and additional inputs such as NDVI and population density.

Abstract

Urban heat islands, defined as specific zones exhibiting substantially higher temperatures than their immediate environs, pose significant threats to environmental sustainability and public health. This study introduces a novel machine-learning model that amalgamates data from the Sentinel-3 satellite, meteorological predictions, and additional remote sensing inputs. The primary aim is to generate detailed spatiotemporal maps that forecast the peak temperatures within a 24-hour period in Turin. Experimental results validate the model's proficiency in predicting temperature patterns, achieving a Mean Absolute Error (MAE) of 2.09 degrees Celsius for the year 2023 at a resolution of 20 meters per pixel, thereby enriching our knowledge of urban climatic behavior. This investigation enhances the understanding of urban microclimates, emphasizing the importance of cross-disciplinary data integration, and laying the groundwork for informed policy-making aimed at alleviating the negative impacts of extreme urban temperatures.
Paper Structure (12 sections, 1 figure, 1 table)

This paper contains 12 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related works
  3. Dataset
  4. Temperature Data
  5. Satellite Data
  6. Weather Data
  7. Topological Data
  8. Methodology
  9. Problem Formulation
  10. Models Architecture
  11. Results
  12. Conclusion and future works

Figures (1)

  • Figure 1: Output sample of the ConvNext model, for the day of Monday, June 12 2023, at a resolution of 50 m/px, superimposed on a satellite image of Turin. Legend displays maximum and minimum predicted temperatures, in Celsius degrees.