 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
In order to understand how the climate anomalies associated with the ENSO phenomenon affect agriculture, we study individual events at the continental scale (encompassing both natural and managed ecosystems) and time-series of vegetation responses in agricultural regions. We are developing remote sensing techniques to characterize both scales of vegetation responses. As examples, we utilize the normalized difference vegetation index (NDVI) (Tucker and Sellers, 1986) to study the biospheric response to the 1997-98 El Niño over Africa and the response of the U.S. Cornbelt to ENSO over the period 1982-97.
Over Africa, the period evaluated encompasses the short rains (October- November) and the long rains (March-May) season in eastern Africa and the growing season over southern Africa (November-March). As shown by the NDVI anomalies (Figs. 1 and 2), the terrestrial biosphere shows a positive response over large areas of eastern Africa from October 1997 to March 1998, greener than normal conditions over southern Africa during the early part of the season (October to November 1997), but drier than normal conditions later in the season, as evidenced in March 1998. This pattern agrees with the nature of ENSO teleconnection patterns over Africa (Ropelewski and Halpert, 1987). However, the magnitude of the departure patterns in both precipitation and NDVI exceeds the normal response patterns during previous warm ENSO events. This was partly due to the enhancement of the ENSO-related anomalies resulting from the warmer than normal western equatorial Indian Ocean. In March 2000, La Niña conditions by contrast brought somewhat opposite patterns.
 Figure 1: Normalized difference vegetation index (NDVI) for Africa in March 1998. (Click for larger image) |
 Figure 2: NDVI for Africa in March 2000. (Click for larger image) |
We applied satellite imagery and ancillary data in the effort to understand the spatial and temporal responses of agricultural regions to ENSO phenomena. Here we demonstrate our methodology in the U.S. Cornbelt, a highly productive, primarily rainfed region with weak but significant responses to the ENSO cycle (Wannebo and Rosenzweig, 1999). The ENSO teleconnections are manifested as short-term local climate perturbations that influence corn yields. La Niña summers, in particular, tend to be warmer and drier than normal summers in the region, creating unfavorable conditions for corn crop development and lower than average yields (Mauget and Upchurch, 1998; Phillips et al., 1999). For example, the drought of 1988, during a La Niña year, is estimated to have cost the Midwest $39 billion in damages. On the other hand, the hot dry summer of 1983, associated with the El Niño event of 1982, also resulted in crop losses, estimated at $10 billion.
To explore the ENSO-crop linkages, we integrate four types of data over forty crop reporting districts designated as the U.S. Cornbelt: 8-km AVHRR NDVI data, Pacific SSTs from the NINO3 region, precipitation, and crop yields. Results indicate that NDVI anomalies during the growing season (June, July and August) do explain some variability in corn yields (r = .441 at 95% level) and are correlated with growing season cumulative rainfall (r = .358 at the 95% level). While the NDVI values seem to reflect local climate conditions related to yields, the physical linkage between the ENSO phenomenon and crop yields requires additional investigation.
A one-way analysis of variance (ANOVA) was used to test if the mean NDVI values differed by ENSO phase (El Niño, La Niña, neutral). Classifying the ENSO phase using SST values for spring and summer of the same year as the growing season showed that both La Niña and El Niño years were significantly lower than neutral years, but not significantly different from each other.
To investigate the spatial component of ENSO impacts, the vegetation condition index (VCI) is used to relate the growing season NDVI value of one year to the minimum and maximum of the time series (Unganai and Kogan, 1998). The VCI is expressed as a percent of the optimal; a VCI value of less than 60% indicates vegetation stress. The mean La Niña, El Niño and neutral VCI values as well as the VCI values for the individual years within each grouping of the simultaneous classification method were mapped (Fig. 6). The VCI was able to capture drought conditions such as the extreme heat waves during both 1983 and 1988. While the mean VCI values for each ENSO phenomenon reflect the general trend of lower than normal NDVI values during both La Niña and El Niño years, no consistent spatial pattern characteristic of either La Niña or El Niño emerged. The lack of evident spatial pattern has implications for preparing for predicted ENSO events, since it characterizes the nature of at least part of the uncertainty facing agricultural production in the region.
For the agriculture component of our project, we devised a consistent set of steps to assess regional ENSO impacts and to aid in utilization of ENSO predictions.
Our work has shown that GCM-based predictions can indicate the direction and degree of seasonal climate variation at gridbox scales. These temporal and spatial scales differ from the scales of the crop models generally used to study agricultural responses to climate perturbations. Agricultural impact models typically require daily sequences of weather variables to drive crop growth and development, reflecting crop sensitivity to climate on the daily time-scale. Furthermore, crop growth is spatially represented at the field-scale. We have tested several approaches for creating climate scenarios appropriate for driving site-based crop simulation models at a daily time-step: (1) direct use of daily GCM output; (2) forcing a regional climate model with GCM results; and (3) training a weather generator with observed weather data and GCM anomalies. Due to the spread of rainfall over an entire grid box, GCM daily data show too many days with small rainfall amounts, rather than a bimodal distribution of dry and heavier rainfall categories. Crop soil-moisture dynamics become unrealistic under these rainfall distributions, even if the monthly totals are well represented. We overcome these limitations by downscaling via a regional climate model and weather generators.
A Regional Climate Model (RCM) developed at Florida State University (Fulakeza, 1998) has been tested over Zimbabwe (Druyan et al., 1999b). A 50-km grid resolves Zimbabwe into 168 space elements and therefore has the potential to represent climate heterogeneity within the country. A pilot study for December 1982 compared with the GCM hindcast indeed showed that the RCM produced regional and temporal sub-GCM-scale differentiations resembling observations. Subsequently, the same method has been used over southern and western Africa, and is to be used in northeastern South America in a domain that includes Nordeste, Amazonia and the Cerrados.
Weather generators are useful in scaling monthly GCM-derived climate anomalies at grid-box resolution down to daily, site-level weather data (Wilks, 1999). We have tested a newly-developed weather generator (Rajagopalan and Lall, 1999) at a number of sites in Zimbabwe using predictions from the 4°x5° nine-layer GISS GCM ensemble runs. The weather generator is distinctive in that it uses a bootstrap approach of random selection of daily rainfall events from the historical database, constrained by time-order dependence, rather than the more common approach of weather generators that use statistical description of the historical base from which to produce new weather scenarios (Richardson, 1981). To analyze the ability of the weather generator to capture local-scale climate dynamics, we chose the strong El Niño event of 1982-83, in which the GCM-predicted precipitation anomaly was well correlated with observed all-Zimbabwe seasonal rainfall. This generated weather dataset was then used to drive a crop simulation model parameterized for a maize hybrid commonly grown in the region, with appropriate soils and fertility levels. Implications for crop management were then drawn from crop simulation results. The negative anomaly in seasonal precipitation predicted by the GCM for 1982-83 produced crop simulations at Masvingo, Zimbabwe (20.07°S 30.83°E) that indicate a potential delay in planting due to lack of soil moisture, but also imply faster crop development because of higher temperatures. Predicted maize yield distribution shifts downward, with the likelihood of crop failure increasing from 18% to 39%. The implication of the regional forecast for Masvingo, as interpreted through a weather generator and a series of crop simulations, is that maize yields are likely to suffer unless mitigative management strategies are applied (Phillips et al., 1998).
Regional collaborations have been established and focused research conducted in major crop-growing regions known to experience an influence of ENSO-cycle variability. These include:
The U. S. Cornbelt indicates a statistically significant relationship among ENSO phase, climate and agricultural production, with lower crop yields during the La Niña phase (Handler, 1986; Piechota and Dracup, 1996; Carlson et al., 1996; Phillips et al., 1999). Our research in this area involves a comparison of NINO3 SSTs with temperature, precipitation and crop yield (Phillips et al., 1999), showing that the negative impacts of warmer and dryer La Niñas outweigh the benefits of El Niño-related rainfall and cooler temperatures. The possibility of providing advance warning of both ENSO phases offers the potential of improving crop management in the Cornbelt.
Zimbabwe experiences a drier climate and lowered crop production during El Niños (Ropelewski and Halpert, 1987; Janowiak, 1988; Cane et al.; 1994). We quantified the relationship between ENSO phase and rainfall utilizing data from approximately 1500 stations covering the period 1972-96. While modeled maize yields were generally lowest in El Niño years, simulations showed that early planting in both El Niño and La Niña years could improve yields. Large variability in both rainfall patterns and yields within ENSO phases suggests the need for more precise definition of ENSO climate teleconnections.
Uruguay experiences wetter climates and better production in El Niño years, with the reverse in La Niña years (Ropelewski and Halpert,1987,1989; Bidegain and Krecl, 1998; Pisciottano et al., 1994; Grimm et al., 1998; Grondona et al., 1998; Diaz et al., 1998; Baethgen, 1997; Podesta et al., 1998; Myneni et al, 1996). Here we have studied the relationship between tropical Pacific SSTs and the historical national yield of a winter crop (wheat) and a summer crop (maize). We have also assessed the capability of two GCMs (GISS and NCEP) and a weather generator to create synthetic weather data with the variability associated with ENSO. Crop simulation models and generated weather data have been used to explore agronomic management practices best suited to the expected conditions defined by the climate forecasts. These research activities are linked to an existing IFDC-INIA collaborative project for seasonal climate outlooks. We have found that maize production, more than wheat, is associated with tropical SSTs, that weather generators worked best to provide estimates of rainfall variability when the SST anomaly is strong, that GCM provide better assessment of seasonal than of interannual variability, and that, in the CERES-MAIZE crop model, a combination of short-season maize hybrids with delayed sowing dates can help maintain yields in La Niña years.
For the agricultural research, we have developed a set of steps to facilitate regional analysis of ENSO effects on agriculture (Rosenzweig, 2000, 2001). A key step is the assessment of the temporal and spatial patterns of ENSO effects through analysis of satellite data (AVHRR NDVI and EOS products). We are now focusing on the utilization of ENSO forecasts for commercial and subsistence agriculture in Uruguay and Nordeste, Brazil, 1996-2000.
Utilization of ENSO Forecasts for Agriculture in Uruguay and Nordeste, Brazil, 1996-2000:
CAFE Home *
Summary *
Team
Components:
ENSO,
Climate,
Marine,
Vegetation
Education *
Instruments *
Acquisitions *
Publications
|
|
|
