Estimating Nonharvested Crop Residue Cover Dynamics Using Remote Sensing

1. Introduction

Non harvested above and below ground carbon must be continuously replaced to maintain the soil resilience and adaptability. The soil organic carbon (SOC) maintenance requirement is the amount of non-harvested carbon (NHC) that must be added to maintain the SOC content at the current level (NHC_m) (Mamani-Pati et al., 2010; Mamani-Pati et al., 2009). To understand the maintenance concept a basic understanding of the carbon cycle is needed (Mamani-Pati et al., 2009). The carbon cycle is driven by photosynthesis that produces organic biomass which when returned to soil can either be respired by the soil biota or contribute to the SOC. The rates that non-harvested biomass is converted from fresh biomass to SOC and SOC is converted to CO₂ are functions of many factors including, management, climate, and biomass composition. First order rate mineralization constants for nonharvested carbon (k_NHC) and SOC (k_SOC) can be used to calculate half lives and residence times. Carbon turnover calculations are based on two equations,

In these equations, SOC is soil organic C, NHC_a is the non-harvested carbon returned to soil, NHC_m is the nonharvested carbon maintenance requirement, k_soc is the first order rate constant for the conversion of SOC to CO₂, and k_NHC is the first order rate constant for the conversion of NHC to SOC (Clay et al., 2006). These equations state that the temporal change in SOC (dSOC/dt) is equal to the non-harvested carbon mineralization rate constant (k_NHC) times the difference between the amounts of carbon added to the soil (NHC_a) and the maintenance requirement (NHC_m) and that at the SOC equilibrium point (SOC_e), the rate that non-harvested C (NHC) is converted into SOC (k_NHC × NHC_m) is equal to the rate that SOC is mineralized into CO₂ (k_SOC × SOC_e). Through algebraic manipulation, these equations can be combined to produce the equation,

When fit to a zero order equation, the y-intercept and slopes are kSOCkNHCand1kNHC×SOCe, respectively.

Based on this equation, NHC_m, k_NHC, and k_SOC can be calculated using the equations, NHC_m = b × SOC_e; k_NHC = 1/ (m × SOC_e); and k_SOC = b/(m × SOC_e). This approach assumed that above and below ground biomass make equal contributions to SOC; that the amount of below ground biomass is known; and NHC is known and that initial (SOC_e) and final (SOC_final) SOC values are near the equilibrium point. Advantages with this approach are that k_SOC and k_NHC are calculated directly from the data and the assumptions needed for these calculations can be tested. A disadvantage with this solution is that surface and subsurface NHC must be measured or estimated. Remote sensing may provide the information needed to calculate surface NHC, through estimating the spatial variation of crop residues which are a major source of NHC.

Traditionally crop residue cover estimates have relied on visual estimation through road side surveys, line-point transect or photographic methods (CTIC, 2004; McNairn and Protz, 1993; Serbin et al., 2009 a). However, such ground-based survey methods tend to be time-consuming and expensive and therefore are inadequate for crop residue quantification over large areas (Daughtry et al., 2005; Daughtry et al., 2006). The need to improve these estimates has prompted much research on the extraction of surface residue information from aerial and satellite remote sensing (Bannari et al., 2006; Daughtry et al., 2005; Gelder et al., 2009; Serbin et al., 2009 a& b; Thoma et al., 2004). Previous research has shown crop residues may lack the unique spectral signature of actively growing green vegetation making the discrimination between crop residues and soils difficult (Daughtry et al., 2005). Daughtry and Hunt (2008) reported that dry plant materials have their greatest effect in the short wave infra-red (SWIR) region between 2000 and 2400 nm related to the concentration of ligno-cellulose in dry plant residue.

Other studies have proposed the Cellulose Absorption Index (CAI), the Lignin Cellulose Absorption index (LCA) and the Shortwave Infrared Normalized Difference Residue Index (SINDRI) for estimating field residue coverage (Daughtry et al., 2005; Daughtry et al., 2006; Thoma et al., 2004; Serbin et al., 2009 c). However, neither CAI, LCA nor SINDRI are currently practical for use in spaceborne platforms (Serbin et al., 2009 a). For example, EO-1 Hyperion which was sensitive to the spectral ranges of CAI and LCA (2100 and 2300 nm wavelengths), is past its planned operational lifetime and suffers bad detector lines (USGS, 2007), while the shortwave infrared (SWIR) detector for ASTER satellite failed in April, 2008 (NASA, 2010; Serbin et al., 2009 c). Therefore, indices that utilize the multispectral wavelength ranges (450-1750 nm) appear to be the most viable economical alternative. The objective of this research was to assess if remote sensing can be used to evaluate surface crop residue cover, and the amount of nonharvested biomass returned to soil.

2. Materials and methods

2.3 Statistical analysis

Proc Mixed available within the Statistical Analysis System (SAS Institute, North Carolina) software, was used to determine reflectance differences in the residue removed and returned plots. Correlation (r) coefficients between reflectance values and weights of stover returned and % surface residue cover were determined. Finally, graphs of percent residue cover versus spectral band and indexfor the models with the highestcorrelations were compared.

3. Results and discussion

In 2009, 28.8 % of the soil was covered with residue in the removed (baled) plots, while in 2010, 54% of the soil was covered with residue (Table 2). In the residue returned (not baled) plots, the surface cover was 100 and 70%, in 2009 and 2010, respectively. The residue removal plots (28.8% cover) in 2009 had the lowest reflectance in the green, red, and NIR bands, while the residue returned plots in 2010 had the highest reflectance in the green, red, NIR, and MIR bands. These results imply that corn residues have a relative high albedo, compared to soil. Slightly different results would be expected in soybean (Glycine max) where Chang et al. (2004) did not detect reflectance differences between bare and soybean residue covered soil.

The correlation coefficient between the residue returned in ton/ha and percent residue cover with surface reflectance are shown in Table 3. The correlation coefficients between residue returned and reflectance ranged from 0.002 in the MIR band to 0.39 in the blue band. For the % surface residue cover, higher r values were observed. These results suggest that surface reflectance measurements were better at predicting the crop residue coverage than residue amount. The highest r value between % ground cover and reflectance was observed for the blue band. The different bands previously have been reported for different uses (http://landsat.usgs.gov/best_spectral_bands_to_use.php). The blue band is useful for distinguishing soil from vegetation, while green is useful for assessing plant vigor. The NIR (770-900 nm) and short-wave infrared (1550 – 1750 nm) discriminates biomass content from soil moisture content. Although blue has a high correlation with surface residue cover, atmospheric scattering may reduce its suitability for space-based sensors (Lillesand and Kiefer, 2000; Wang et al., 2010).

The amount of residue retained on the soil was correlated to NDVI, GNDVI, and NDWI, while the percent coverage was correlated to NDVI and NDWI. Of the indices determined, NDWI_w had the highest r value with percent residue cover followed by NDVI_w, GNDVI_w, and BNDVI_w respectively. It is important to note, that the results are limited by the boundary conditions of the experiment. Although only the percent residue cover and residue weights versus the reflectance were analyzed, other factors such as soil cover, color or moisture could have impacted the reflectance values (Barnes et al. 2003; Daughtry et al., 2005; Daughtry et al., 2006; Pacheco and McNairn, 2010; Thoma et al., 2004).

A comparison of the reflectance data across years for the blue band suggests that a zero order equation (y = 4.31 + 0.036x; r² = 0.38) could explain the relationship between reflectance and surface coverage (Fig. 1). Slightly different results were observed for the NDWI_w indices where a second order quadratic equation (y = -0.22 + 0.005x – 0.0000263 x²; R² = 0.26) was used to describe the relationship with surface coverage. The graph of % residue cover versus NDWIw flattens after the 60 % residue cover which implies that NDWI_w may saturate with increasing coverage. A limitation of this study is that only one site was analyzed, therefore any models generated would be suitable for the specific site and only after fall harvest. Other errors can occur when extrapolating plot measurements data to the whole field coverage. In future, research that confirms the finding for other sites and harvesting approach needs to be conducted.

4. Conclusion

The main objective of this study was to investigate the potential of remote sensing to assess residue coverage. The research showed that surface reflectance was more closely correlated with percent cover than the amount of residue returned. Of the spectral band widths measured, reflectance in the blue range provided the most consistent results across the two years. NDWI_w had a higher correlation with residue returned and % cover than NDVI_w, GNDVI_w, or BNDVI_w. Future studies should not only consider more field sites, but incorporate factors such as the decomposition rates of residues on spectral reflectance and harvesting approaches (see Daughtry et al. 2010), so as to develop an accurate and standard approach for mapping residue cover in real time over large geographic areas.