A Machine Learning Approach to Long-Term Drought Prediction Using Normalized Difference Indices Computed on a Spatiotemporal Dataset

Aug 6, 2026

Upload by Jiven

Climate change intensifies drought, threatening agricultural productivity and livestock supply. This research introduces a machine-learning framework for long-term drought prediction focused on water-basin areas. The approach performs next-frame prediction of the Normalized Difference Drought Index (NDDI) using the SEN2DWATER spatiotemporal database. Two methods are compared by changing where NDDI is computed to estimate NDDI values over a target land region. The study supports proactive drought mitigation and sustainable water access via proof-of-concept short- and long-term resource change forecasting.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages4

Reading Time10 min

AccessFree to download

Abstract

Introduction

Dataset

Index computation (NDVI, NDMI, NDDI)

Image preparation and study regions

A MACHINE LEARNING APPROACH TO LONG-TERM DROUGHT PREDICTION USING NORMALIZED DIFFERENCE INDICES COMPUTED ON A SPATIOTEMPORAL DATASET arXiv :2302 .02440v2 [ ee ss .IV] 1 Jun 2023 Veronica Wairimu Murigaa,1, Benjamin Richa, Francesco Mauro b, Alessandro Sebastianellic and Silvia Liberata Ullob a Massachusetts Institute of Technology, Boston, USA b Engineering Department, University of Sannio, Benevento, Italy c ϕ-lab, European Space Agency, Frascati, Italy ABSTRACT Climate change and increases in drought conditions affect the lives of many and are closely tied to global agricultural output and livestock production. This research presentsa novel approach utilizing machine learning frameworks for drought prediction around water basins. Our method focuses on the next-frame prediction of the Normalized Difference Drought Index (NDDI) by leveraging the recently developed SEN2DWATER database. We propose and compare two prediction methods for estimating NDDI values over a specific land area. Our work makes possible proactive measures that can ensure adequate water access for drought-affected communities and sustainable agriculture practices by implementing a proof-of-concept of short and long-term drought prediction of changes in water resources. Index Terms— Climate change, Drought, Sentinel-2, Water Indices, Deep Learning, Late/Early Stage Computation 1. INTRODUCTION Previous studies in remote sensing and drought assessment have led to the development of several indices computed from multispectral and hyperspectral data [1, 2] . Among these indices, the Normalized Difference Moisture Index (NDMI) [3] is valuable for monitoring vegetation moisture levels, while the Normalized Difference Vegetation Index (NDVI) [4, 5] helps assess the presence of live green vegetation in a landscape. The Normalized Difference Drought Index (NDDI), which combines information from NDVI and NDMI, has been shown to be particularly useful for mapping drought severity and identifying drought-prone areas at an early stage [6, 7] . Machine learning (ML) has opened new avenues for climate and weather forecasting. Recent advancements have demonstrated the successful application of various neural network architectures to remote sensing data [8–12] . Machine 1 Corresponding author. Email addresses: [wmuriga](wmuriga@mit.edu)[@](wmuriga@mit.edu)[mit.edu](wmuriga@mit.edu)[ ](wmuriga@mit.edu)(VWM), [brrich](brrich@mit.edu)[@](brrich@mit.edu)[mit.edu](brrich@mit.edu) (BR), [f.mauro](f.mauro@studenti.unisannio.it)[@](f.mauro@studenti.unisannio.it)[studenti.unisannio.it](f.mauro@studenti.unisannio.it) (FM), [Alessandro.Sebastianelli](Alessandro.Sebastianelli@esa.int)[@](Alessandro.Sebastianelli@esa.int)[esa.int](Alessandro.Sebastianelli@esa.int) (AS), [ullo](ullo@unisannio.it)[@](ullo@unisannio.it)[unisannio.it](ullo@unisannio.it) (SLU) learning techniques have additionally been employed to predict changes in vegetation health, water content, quality, and moisture levels using normalized indices [4, 13, 14] . Among various frameworks for predicting such indices, Time Distributed Convolutional Neural Networks (TD-CNNs) show promise in time series forecasting as they combine CNN layers for pattern recognition in images with a Long-Short Term Memory (LSTM) layer capable of identifying changes across time series data (Using TD-CNNs with Keras) . In this study, we expand on the application of ML methods to normalized difference indices, focusing on NDDI. Previous research has shown that the NDDI, computed using visible, near-infrared, and short-wave infrared channels, exhibits enhanced sensitivity as a drought indicator compared to other indices [7] . Therefore, we investigate the effectiveness of NDDI as an early indicator of drought or significant environmental changes using data from European landscapes near water bodies. We train TD-CNNs on NDVI, NDMI, and NDDI data obtained from SEN2DWATER, a 2-dimensional dataset derived from multispectral S

FAQs

What is the main goal of the machine learning approach in the study?

To predict long-term drought conditions by forecasting NDDI values over space and time for areas near water basins.

How does the method generate inputs for prediction?

It computes NDVI, NDMI, and NDDI from multispectral Sentinel-2 bands using the SEN2DWATER spatiotemporal dataset.

Why is NDDI emphasized instead of other indices?

NDDI is described as a drought indicator that combines NDVI and NDMI information and shows enhanced sensitivity for mapping drought severity early.

What two prediction methods does the paper compare?

Two approaches are compared by changing the point at which NDDI computation is performed when training and predicting with TD-CNN models.

Aug 6, 2026

Upload by Jiven