Comparison of Phenology Models for Predicting the Onset of Growing Season over the Northern Hemisphere

Yuan W (2014) Comparison of Phenology Models for Predicting the Onset of Growing Season over the Northern Hemisphere. PLoS ONE 9(10): e109544. doi:10.1371/journal.pone.0109544 Comparison of Phenology Models for Predicting the Onset of Growing Season over the Northern Hemisphere Yang Fu 0 Haicheng Zhang 0 Wenjie Dong 0 Wenping Yuan 0 Ben Bond-Lamberty, DOE Pacific Northwest National Laboratory, United States of America 0 1 State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University , Beijing , China , 2 State Key Laboratory of Cryospheric Sciences, Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences , Lanzhou, Gansu , China Vegetation phenology models are important for examining the impact of climate change on the length of the growing season and carbon cycles in terrestrial ecosystems. However, large uncertainties in present phenology models make accurate assessment of the beginning of the growing season (BGS) a challenge. In this study, based on the satellite-based phenology product (i.e. the V005 MODIS Land Cover Dynamics (MCD12Q2) product), we calibrated four phenology models, compared their relative strength to predict vegetation phenology; and assessed the spatial pattern and interannual variability of BGS in the Northern Hemisphere. The results indicated that parameter calibration significantly influences the models' accuracy. All models showed good performance in cool regions but poor performance in warm regions. On average, they explained about 67% (the Growing Degree Day model), 79% (the Biome-BGC phenology model), 73% (the Number of Growing Days model) and 68% (the Number of Chilling Days-Growing Degree Day model) of the BGS variations over the Northern Hemisphere. There were substantial differences in BGS simulations among the four phenology models. Overall, the Biome-BGC phenology model performed best in predicting the BGS, and showed low biases in most boreal and cool regions. Compared with the other three models, the two-phase phenology model (NCD-GDD) showed the lowest correlation and largest biases with the MODIS phenology product, although it could catch the interannual variations well for some vegetation types. Our study highlights the need for further improvements by integrating the effects of water availability, especially for plants growing in low latitudes, and the physiological adaptation of plants into phenology models. - Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. All relevant data are within the paper and its Supporting Information files. Funding: This study was supported by the National Science Foundation for Excellent Young Scholars of China (41322005), National Natural Science Foundation of China (41201078), Program for New Century Excellent Talents in University (NCET-12-0060) and the Fundamental Research Funds for the Central Universities. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. Phenology refers to the timing of recurring biological cycles, and is considered a sensitive indicator of climate change [13]. In particular, as research interest in global change increases, determining the beginning of the growing season (BGS) of land vegetation has become an important research subject [4]. Previous studies revealed that plant activity is more sensitive to climatic changes in spring than other seasons; and changes in the BGS would strongly impact the seasonal energy balance and net carbon dioxide (CO2) flux of terrestrial ecosystems [5,6]. Large uncertainties, however, in present phenology models make accurate assessment of BGS a challenge. Two classes of process-based models have been developed for simulating the spring phenological phases. Models belonging to the first class, the one-phase models, are the simplest and have been used in agronomy since the 18th century [7]. This kind of model implicitly assumes that bud dormancy is fully released after a fixed sum of accumulated temperatures has been reached. The second class of models, the two-phase models, considers the breaking of two dormancy phases [8]. The first phase is a period when buds remain dormant due to plant endogenous factors, and the second phase is a period when buds remain dormant due to unfavorable environmental factors [9]. Many studies have described the breaking of the first phase and overcoming the second phase in terms of chill accumulation to break the first phase followed by a period of forcing temperature to overcome the second phase [10,11]. The two-phase models are of more recent development, and are conceptually based on experimental studies which highlighted that chilling was the major factor responsible for dormancy release [1215]. Many phenology observations have focused on cultivated rather than natural plants [16,17]. Geographically, most of the observations were conducted in North America and Europe [1820]. Due to the limited availability of phenological observation data on a large scale, most phenology models are calibrated at local scales [21] and thus are unlikely to accurately predict BGS across different vegetation types. These phenology models might underestimate or overestimate the BGS when applied to a regional or global scale [22]. For example, a comparison of phenology models in 14 terrestrial biosphere models indicated that almost all models failed to track the phenology, and most predicted an earlier BGS, overestimating the gross ecosystem photosynthesis by 20% [23]. Figure l. Vegetation distribution map of the Northern Hemisphere retrieved from the V005 MODIS Land Cover Type Product (MCD12Q1). Grey areas are either excluded vegetation types like croplands, or areas with no seasonal cycle detectable by satellite. doi:10.1371/journal.pone.0109544.g001 Remote sensing data from satellites provide broad coverage of useful information on vegetation phenology for diverse ecosystems at various scales, and help to calibrate the phenology models [24 28]. For example, Yang et al. [22] parameterized three budburst models in New England using 11 years of remotely sensed phenology and climate data. Nowadays, remote sensing-based phenology has been significantly improved with the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra and Aqua satellites [29]. Since 2009, the latest version of the MODIS Land Cover Dynamics Product (MCD12Q2) has been available [30], which provides valuable phenology data for the present study. Based on the global satellite-based phenological observations, the primary objectives of this study are to (1) calibrate four phenology models; (2) compare the relative strengths of four phenology models; and (3) assess the spatial pattern and interannual variability of BGS in the Northern H (...truncated)