Application of remote sensing in crop growth simulation and an ensembles approach to reduce model uncertainties
Abstract
A semi-empirical model derived from the water-cloud model was used to convert synthetic- aperture radar (SAR) backscattering data into LAI. The SAR-based LAI at early rice growth stages were in a close agreement (90%) with LAI derived from MODIS data for the same study location in Nueva Ecija, Philippines. ORYZA2000 simulated rice yield of 4.5 Mg ha-1 for the 2008 wet season in Nueva Ejica, Philippines when using LAI inputs derived from SAR data, which is closer to the observed yield of 3.9 Mg ha-1, whereas simulated yield without SAR-derived LAI inputs was 5.4 Mg ha-1. The dynamic water and nitrogen balances were accounted in these simulations based on site-specific soil properties and actual fertilizer N and water management. The use of remote sensing data was promising for model application to approximate actual growth conditions and to compensate for limitations in the model due to relevant underlining processes absent in model formulations such as detailed tillering, leaf shading effect, etc., and also limiting factors not accounted in the model such as biotic factors and abiotic factors other than water and N shortages. This study also demonstrated the use an ensembles approach for provincial level rice yield estimation in the Philippines. Such ensembles approach involved statistical classifications of agronomic management settings into 25% percentile, median, and 75% levels followed by generation of factorial combinations. For irrigated lowland system, 4 factors were considered that include transplanting date, plant density, fertilizer N rate, and amount of irrigation water. For rainfed lowland system, there were 3 agronomic management factors (transplanting date, plant density, fertilizer N) and 1 soil parameter (depth of ground water table). These 4 management/soil factors and 3 statistical levels resulted in 81 total factorial combinations representing simulation scenarios for each area of interest (province in the Philippines) and water environments (irrigated vs. rainfed). Finally a normal distribution was assumed and applied to the simulations outputs. This ensembles approach provided an efficient and yet effective method of aggregating point-based crop model results into a larger spatial level of interest. Lack of access to accurate model parameters (e.g. depth of ground water table) could be solved with this approach. The use of process-based crop growth model was critical because the ultimate aim of this study was not just to establish a reliable rice yield estimation system but also to allow yield estimation outputs explainable by the underlining agronomic practices such as transplanting date, fertilizer N application, and water management.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2012
- Bibcode:
- 2012AGUFMGC13B1077S
- Keywords:
-
- 0402 BIOGEOSCIENCES / Agricultural systems;
- 0466 BIOGEOSCIENCES / Modeling;
- 0480 BIOGEOSCIENCES / Remote sensing;
- 9320 GEOGRAPHIC LOCATION / Asia