Estimating maize biomass and yield over large areas using high spatial and temporal resolution Sentinel-2 like remote sensing data

- Spatialization of agrometeorological model with high resolution remote sensing data
- A robust and generic method mainly based on remote sensing data is proposed.
- Modification of model equations leads to a better fitting of biomass at local scale.
- A very good adequacy of model and yield statistics was found at the regional scale.
- The methodology developed could be applied to Sentinel-2 images.

This study aims at developing a robust and generic methodology, based on the use of high resolution remote sensing data to provide accurate estimates of maize biomass and yield over large areas (i.e. at regional scale). We propose here a strategy of calibration and spatialization independent as much as possible of in situ measurements and reliable over large areas and under various climatic conditions. For this purpose, we combine the Simple Algorithm For Yield estimates (SAFY) model with high spatial and temporal resolution remote sensing data from several sensors: Formosat-2, SPOT4-Take5, Landsat-8 and Deimos-1. SPOT4-Take5 experiment conducted in 2013 was designed to simulate the temporal sampling of ESA's Sentinel-2 mission. This study led to a new version of the SAFY model that takes into account the seasonal variation of specific leaf area (SLA) and effective light use efficiency (ELUE). The study takes place in a temperate agrosystem located in the south west of France. The SAFY outputs were validated with local measurements of biomass and yield estimates at both local and regional scales using a multiannual dataset. Good results were obtained for both local biomass (R = 0.98; RRMSE = 14%) and yield (R = 0.81; RRMSE = 8.9%), and for yield estimations at regional scale (R = 0.96; RRMSE = 4.6%). Results also showed that the use of a double logistic function to interpolate Green Area Index (GAI) time series permits to improve the estimations of biomass and yield when remote sensing data are missing. This work demonstrates the potential of high resolution remote sensing data to calibrate a simple crop model without resorting to in situ data and thus foreshadows the future applications using Sentinel-2 data.