Detection of Crop Hail Damage with a Machine Learning Algorithm Using Time Series of Remote Sensing Data

Leandro Sosa,Ana Justel,Íñigo Molina

doi:10.3390/agronomy11102078

Abstract

Hailstorms usually result in total crop loss. After a hailstorm, the affected field is inspected by an insurance claims adjuster to assess yield loss. Assessment accuracy depends largely on in situ detection of homogeneous damage sectors within the field, using visual techniques. This paper presents an algorithm for the automatic detection of homogeneous hail damage through the application of unsupervised machine learning techniques to vegetation indices calculated from remote sensing data. Five microwave and five spectral indices were evaluated before and after a hailstorm in zones with different degrees of damage. Dual Polarization SAR Vegetation Index and Normalized Pigment Chlorophyll Ratio Index were the most sensitive to hail-induced changes. The time series and rates of change of these indices were used as input variables in the K-means method for clustering pixels into homogeneous damage zones. Validation of the algorithm with data from 91 soybean, wheat, and corn plots showed that in 87.01% of cases there was significant evidence of differences in average damage between zones determined by the algorithm within the plot. Thus, the algorithm presented in this paper allowed efficient detection of homogeneous hail damage zones, which is expected to improve accuracy and transparency in the characterization of hailstorm events.

Highlights

In the last decades, both the frequency and intensity of extreme weather and climate events have increased worldwide, leading to huge economic losses [1]
Hail damage detection using satellite data is based on the premise that strong winds, rainfall, and hail impacts on crop architecture may alter the physical characteristics of the canopy and underlying soil, changing the energy backscattered from the crop to active and passive sensors [44]
We have identified some particular situations in plots with no significant evidence of differences in degrees of crop damage between the three Homogeneous Damage Zones (HDZs) defined by the algorithm

Summary

Introduction

Both the frequency and intensity of extreme weather and climate events have increased worldwide, leading to huge economic losses [1]. The Pampas region of Argentina, the most important crop production area in the country, is located in midlatitudes, and the time of higher frequency of hail events overlaps both the stage of the highest vulnerability of winter crops (such as wheat, which reaches the fruiting stage in the warm season) and the whole growth period of summer crops such as soybeans and corn. Farmers aim to mitigate the consequences of natural catastrophes through management strategies within their farming operations, but certain risks, such as hail events, are beyond their financial means. For such cases, farmers usually transfer the risk to specialized risk management companies by taking out agricultural insurance policies, with hail damage policies being very widespread in the country

Objectives

Methods

Results

Conclusion