Retrieval of wheat leaf area index from AWiFS multispectral data using canopy radiative transfer simulation

Highlights

• Use of canopy radiative transfer model for retrieval of regional wheat LAI from AWiFS data.

• Comparison of canopy radiative transfer and empirical model for wheat LAI retrieval.

• Canopy radiative transfer model captured better spatial LAI heterogeneity than empirical model with better accuracy.

Abstract

Accurate representation of leaf area index (LAI) from high resolution satellite observations is obligatory for various modelling exercises and predicting the precise farm productivity. Present study compared the two retrieval approach based on canopy radiative transfer (CRT) method and empirical method using four vegetation indices (VI) (e.g. NDVI, NDWI, RVI and GNDVI) to estimate the wheat LAI. Reflectance observations available at very high (56 m) spatial resolution from Advanced Wide-Field Sensor (AWiFS) sensor onboard Indian Remote Sensing (IRS) P6, Resourcesat-1 satellite was used in this study. This study was performed over two different wheat growing regions, situated in different agro-climatic settings/environments: Trans-Gangetic Plain Region (TGPR) and Central Plateau and Hill Region (CPHR). Forward simulation of canopy reflectances in four AWiFS bands viz. green (0.52–0.59 μm), red (0.62–0.68 μm), NIR (0.77–0.86 μm) and SWIR (1.55–1.70 μm) were carried out to generate the look up table (LUT) using CRT model PROSAIL from all combinations of canopy intrinsic variables. An inversion technique based on minimization of cost function was used to retrieve LAI from LUT and observed AWiFS surface reflectances. Two consecutive wheat growing seasons (November 2005–March 2006 and November 2006–March 2007) datasets were used in this study. The empirical models were developed from first season data and second growing season data used for validation. Among all the models, LAI-NDVI empirical model showed the least RMSE (root mean square error) of 0.54 and 0.51 in both agro-climatic regions respectively. The comparison of PROSAIL retrieved LAI with in situ measurements of 2006–2007 over the two agro-climatic regions produced substantially less RMSE of 0.34 and 0.41 having more R² of 0.91 and 0.95 for TGPR and CPHR respectively in comparison to empirical models. Moreover, CRT retrieved LAI had less value of errors in all the LAI classes contrary to empirical estimates. The PROSAIL based retrieval has potential for operational implementation to determine the regional crop LAI and can be extendible to other regions after rigorous validation exercise.

Introduction

Plant biophysical traits are generally used to model energy transfer processes between land and atmosphere (Bonan, 1991, Chen et al., 2007, Liu et al., 1997, Seller et al., 1997). Accurate crop leaf area index (LAI) is required as forcing parameter to predict the crop yield using crop simulation model and for efficient crop management at farm-scale. Any impact due to abiotic (e.g. temperature, water and nutrients) and biotic (e.g. pests and diseases) stresses would significantly change the crop LAI (Hameed et al., 2002, Jackson, 1986). Therefore, monitoring of LAI at farm to regional-scale is of practical importance for crop monitoring and agricultural management to settle damage claim under crop insurance. Satellite optical bands provide the spatial and temporal variation of spectral vegetation characteristics. The shape and form of canopy reflectance spectra depend on several factors like canopy structure, leaf biochemical composition, soil background and view and illumination geometry. The visible part of the canopy reflectance spectra is strongly controlled by leaf chlorophyll content (Gitelson et al., 2003). LAI has a large impact on reflectance spectra especially in the near-infrared (NIR) wavelength regions (Botha et al., 2007, Price and Bausch, 1995, Yi et al., 2008). Leaf water content is the primary determinant of canopy reflectance in the middle infrared (MIR) wavelength region.

Crop LAI can be estimated either from empirical models based on spectral vegetation indices (VIs) or through physically explicit canopy radiative transfer simulation model using inversion. Empirical models are generally restricted to site, time and crop. Therefore, these models cannot be extrapolated to large-scale for operational implementation (Baret and Guyot, 1991, Chaurasia et al., 2011, Colombo et al., 2003, Gobron et al., 1997). The physical models are proved as promising alternative because they describe the transfer and interaction of radiation within a canopy at regular spectral interval in optical region (Jacquemoud et al., 1996, Jacquemoud et al., 2000). These models provide an intrinsic connection between the plant biophysical characteristics and canopy reflectances. Various techniques have been proposed for the inversion of these models, including numerical optimization method (Jacquemoud et al., 1995, Jacquemoud et al., 2000), look-up table (LUT) approach (Combal et al., 2002) and neural network method (Danson et al., 2003, Leshno et al., 1993). The LUT approach is faster one and relies on a large set of database for the retrieval of specific variable. With the advancement in the efficiency of computational power, this methodology has the potential to retrieve the biophysical variables such as LAI, through inversion at various spatial scales.

Agriculture contributes approximately 14.2% in Gross Domestic Product (GDP) of India, whereas, wheat alone contributes 30.5% of total food production (Department of Agriculture and Cooperation, Ministry of Agriculture, Govt. of India, 2011). It is the dominant crop of north and north-central India during winter (rabi) season and the second highest staple food grain after rice. There is a wide variability in wheat grain productivity which varies between 1600 and 4600 kg ha⁻¹ (Bhattacharya et al., 2011). Its yield prediction has a paramount importance in India to decide the import–export policies and minimum selling price for the farmers. The Ministry of agriculture, Government of India recently established a dedicated centre named Mahalanobis National Crop Forecasting Centre (MNCFC) (http://agricoop.nic.in) to provide crop acreage and production of major crops in advance on national scale. The operational pre-harvest yield forecast of crop such as wheat requires LAI as input to crop simulation model. The objectives of this study are (i) to develop a methodology for wheat LAI retrieval using CRT model and AWiFS (Advanced Wide Field Sensor) data, (ii) validation of retrieved LAI with in situ measurements and (iii) comparison of CRT retrieved LAI with empirical estimates from VI–LAI models.