Automatic mapping of rice fields in the Sacramento Valley for water resources management

Abstract:

Water use by rice fields is one of the most important components in hydrologic model simulation of the Sacramento Valley, California. In this study, rice fields were mapped by an automatic approach using Landsat imagery. The automatic approach is advantageous for its capacity of mapping rice fields repeatedly, consistently and timely without the need to collect training data. Seasonal dynamics of Enhanced Vegetation Index (EVI) and Normalized Difference Moisture Index (NDMI) were employed to identify rice based on its phenological characteristics. Classification could be conducted around planting date for early response to cropland use change, or for the full growing season to monitor rice growth. Two studies are illustrated as the applications of this mapping method:

1. A rice map was produced before mid-June to forecast rice acreage and water use in the 2015 drought. Due to continuous drought, rice acreage in the Sacramento Valley reached the historical minimum of the past 20 years in 2014, and further reduction is occurring in 2015. A quantitative measure of rice field extent is needed to forecast rice water use as early as possible. The automatic mapping method utilized the spectral dynamics during initial flooding to identify rice fields. Based on the map product, the forecast of rice water demand was made to facilitate the simulation of current-year hydrologic conditions.

2. Rice field extent has been mapped since 1989 and phenological metrics have been derived to study the change in growing season. The increasing use of short-season rice varieties and special weather condition (like El Nino in 2015) may alter the seasonal pattern of water demand by rice. Rice fields were identified based on the temporal profiles of NDMI and EVI derived from series of segmented images. Validation using field survey data and other land use maps showed a promising accuracy. The start and the end of the growing season and other phenological metrics were extracted from object-level temporal profiles of EVI by a curve-fitting method. The monthly pattern of water demand by rice fields was depicted for each region and each year to improve the calibration of hydrologic simulation models.