Regional Crop Gross Primary Productivity and Yield Estimation Using Fused Landsat-MODIS Data

Mingzhu He, John Kimball, Marco Maneta, Bruce Maxwell, Alvaro Moreno, Santiago Beguería, Xiaocui Wu
2018 Remote Sensing  
Accurate crop yield assessments using satellite remote sensing-based methods are of interest for regional monitoring and the design of policies that promote agricultural resiliency and food security. However, the application of current vegetation productivity algorithms derived from global satellite observations is generally too coarse to capture cropland heterogeneity. The fusion of data from different sensors can provide enhanced information and overcome many of the limitations of individual
more » ... ensors. In thitables study, we estimate annual crop yields for seven important crop types across Montana in the continental USA from 2008-2015, including alfalfa, barley, maize, peas, durum wheat, spring wheat and winter wheat. We used a satellite data-driven light use efficiency (LUE) model to estimate gross primary productivity (GPP) over croplands at 30-m spatial resolution and eight-day time steps using a fused NDVI dataset constructed by blending Landsat (5 or 7) and Terra MODIS reflectance data. The fused 30-m NDVI record showed good consistency with the original Landsat and MODIS data, but provides better spatiotemporal delineations of cropland vegetation growth. Crop yields were estimated at 30-m resolution as the product of estimated GPP accumulated over the growing season and a crop-specific harvest index (HI GPP ). The resulting GPP estimates capture characteristic cropland productivity patterns and seasonal variations, while the estimated annual crop production results correspond favorably with reported county-level crop production data (r = 0.96, relative RMSE = 37.0%, p < 0.05) from the U.S. Department of Agriculture (USDA). The performance of estimated crop yields at a finer (field) scale was generally lower, but still meaningful (r = 0.42, relative RMSE = 50.8%, p < 0.05). Our methods and results are suitable for operational applications of crop yield monitoring at regional scales, suggesting the potential of using global satellite observations to improve agricultural management, policy decisions and regional/global food security. yield for various cropping systems [1, [3] [4] [5] [6] [7] . Agricultural surveys provide a reliable way to estimate crop yield, but are less effective over larger regions due to excessive time and budget constraints [8] . In recent decades, satellite remote sensing has been employed for agricultural applications, including crop yield monitoring [9] [10] [11] . Current operational satellite records, including Landsat and MODIS (MODerate resolution Imaging Spectroradiometer), are sensitive to photosynthetic vegetation cover and provide frequent observations with global coverage and consistent sampling, as well as relatively long-term overlapping records. Traditional crop yield estimation methods have used empirical relationships between vegetation biomass and remote sensing spectral vegetation indices to estimate yields [3, 11, 12] . For example, crop yield derived from MODIS NDVI data from 2000 to 2006 in the Canadian Prairies for barley, canola, field peas and spring wheat accounted for 48 to 90%, 32 to 82%, 53% to 89% and 47 to 80% of the variability in reported crop yield from Statistics Canada, respectively [13] . However, these empirical models are fundamentally simple and specific to the limited areas and conditions from which they were developed and cannot easily be extended to other areas. Another approach involves estimating crop yield as the product of vegetation gross primary productivity (GPP) and an empirical harvest index (HI) specific to different crop types. GPP, representing the total carbon uptake by plant photosynthesis, can be estimated at spatial and temporal scales suitable for cropland applications using a light use efficiency (LUE) model driven by remote sensing inputs [14] [15] [16] . Two global operational GPP products are currently produced using the satellite data-driven LUE model logic, including the NASA MODIS MOD17 and SMAP (Soil Moisture Active Passive) Level 4 Carbon (L4C) products [15, 17] . The MOD17 product provides continuous GPP estimates with eight-day temporal fidelity and 500-m spatial resolution (Version 6) spanning all global vegetated ecosystems and extending from 2000 to the present. The L4C product uses a similar LUE model framework driven by combined satellite information from MODIS and SMAP sensors to estimate GPP and underlying environmental constraints to vegetation growth, including soil moisture-related water supply controls; the L4C product is derived globally from 2015 to the present and provides daily temporal fidelity and 1 to 9-km resolution. However, while operational GPP products derived from global satellite observations provide consistent and frequent temporal sampling, the spatial scale of these products may be too coarse for many agricultural applications; the global LUE algorithm parameterizations for croplands used in the MOD17 and L4C products also only distinguish general crop functional types (e.g., cereal vs. broadleaf), which can degrade GPP accuracy for agricultural ecosystems [17] [18] [19] [20] [21] . Alternatively, GPP products derived using LUE model parameterizations that distinguish a greater number of crop types, and with finer spatial resolution (e.g., 30-m) and suitable temporal fidelity (e.g., eight-day), may overcome many of the above limitations while enhancing the utility of these data for agricultural applications. The Landsat TM and ETM+ sensors on the Landsat 5 and 7 platforms provide 30-m resolution imagery that is well suited for capturing surface spectral reflectance heterogeneity at the level of individual agricultural fields [22, 23] . However, Landsat has limited temporal coverage due to a long revisit cycle (16-day), data loss from atmosphere aerosol and cloud contamination and failure of the Landsat 7 sensor Scan Line Corrector (SLC) in May 2003 [23, 24] ; these factors contribute to degraded Landsat utility for cropland monitoring [23, 25] . MODIS provides similar spectral information as Landsat, but with more frequent eight-day composite global observations of surface conditions, albeit at a coarser (250 to 1000 m) spatial resolution that is less suitable for heterogeneous agricultural landscapes [22] . Data fusion methods have been used to reduce the constraints of single sensor remote sensing by blending similar spectral information from Landsat and MODIS to generate harmonized multi-sensor observations, providing both relatively fine-scale spatial resolution and frequent temporal sampling [22, [25] [26] [27] [28] [29] [30] . The spatial and temporal adaptive reflectance fusion model (STARFM), developed by [25] , has been widely used for blending surface reflectance data from Landsat and MODIS. The STARFM approach was modified and improved by [23] for more complex situations. However, the STARFM model is computationally expensive, which can impose a constraint on regional applications. Alternatively,
doi:10.3390/rs10030372 fatcat:4kfdokcr7jdr7ivwhucp2bkxqq