Advanced Estimation of Winter Wheat Leaf’s Relative Chlorophyll Content Across Growth Stages Using Satellite-Derived Texture Indices in a Region with Various Sowing Dates

TL;DR

This study improves the estimation of chlorophyll in winter wheat leaves using satellite data and texture indices across different growth stages and sowing dates.

Contribution

The novel approach integrates satellite-derived texture indices with machine learning to estimate chlorophyll content across multiple wheat growth stages.

Findings

The SVR-RBFVIs+TIs model achieved high accuracy (R2 = 0.8131) in estimating SPAD values.

Texture indices significantly improved estimation accuracy compared to using vegetation indices alone.

The model showed good transferability to different locations with varying sowing times and soil types.

Abstract

Accurately estimating leaves’ relative chlorophyll contents (widely represented by Soil and Plant Analysis Development (SPAD) values) across growth stages is crucial for assessing crop health, particularly in regions characterized by varying sowing dates. Unlike previous studies focusing on high-resolution UAV imagery or specific growth stages, this research incorporates satellite-derived texture indices (TIs) into a SPAD value estimation model applicable across multiple growth stages (from tillering to grain-filling). Field experiments were conducted in Jiangsu Province, China, where winter wheat sowing dates varied significantly from field to field. Sentinel-2 imagery was employed to extract vegetation indices (VIs) and TIs. Following a two-step variable selection method, Random Forest (RF)-LassoCV, five machine learning algorithms were applied to develop estimation models. The newly developed model (SVR-RBFVIs+TIs) exhibited robust estimation performance (R2 = 0.8131, RMSE = 3.2333, RRMSE = 0.0710, and RPD = 2.3424) when validated against independent SPAD value datasets collected from fields with varying sowing dates. Moreover, this optimal model also exhibited a notable level of transferability at another location with different sowing times, wheat varieties, and soil types from the modeling area. In addition, this research revealed that despite the lower resolution of satellite imagery compared to UAV imagery, the incorporation of TIs significantly improved estimation accuracies compared to the sole use of VIs typical in previous studies.

Figures