These simulations suggest the direction of farmland adjustment, which serves as a measure of private adaptation to respond to projected climate changes. Our projection results suggest that a potential decrease in winter chill hours could result in an increase in the percentage of perennial crops grown in the Central Valley.Our outcome variables are the land use shares of perennial, annual, and non-cultivated crops at the parcel level in our study region. To determine these agricultural land use, we rely on Cropland Data Layer , a raster-based land-use map, at 30×30 meter resolution for 2008 through 2021. The USDA, National Agricultural Statistic Service publishes CDL products using a machine learning model based on a combination of satellite imaging and agricultural ground data collected during the growing season . CDL products are available for the contiguous United States at a 30 m spatial resolution annually since 2008.7 Despite the widespread use of CDL data in agriculture climate research, CDL data has inherent errors that could potentially lead to uncertainty in land use change calculations . Although there are limitations, grow strawberry in containers the CDL data is still the primary source of land use information at the micro-level and is frequently used in the literature to influence land use policies . We are cautious in the use of CDL data in our study region and take the following steps to minimize any potential errors that may arise from using CDL data.
First, following Lark et al. and Reitsma et al. , we combine CDL classes into perennial, annual, and non-cultivated to minimize any errors related to CDL ability to distinguish spectrally similar land cover classes. Second, we use time-series CDL data for a given parcel to derive parcel-specific crops. Third, we compare the construction validity of the derived crop acreage in our sample by taking the ratio of the parcel area derived from the geographic information system to the parcel area reported in the assessor’s data. Any value greater than one means that the area of the parcel from the GIS exceeds the area reported in the assessor’s table. We drop observations above 95% of the distribution of measurement errors. Fourth, we compare aggregated crop types acreages obtained from CDL data to administrative data at the county level. To compare the satellite-derived Cropland Data Layer with administrative data, we obtained the county-specific annual crop statistics for California from the National Agricultural Statistics Services . The NASS report is based on the yearly crop reports of the California County Agricultural Commissioners. Specifically, we use county-level data on total harvested acres from annual crop reports between 2008 and 2021 to estimate the total harvested acres for particular crops in each county of California. We present two comparison plots using the CDL and NASS dataset that has been aggregated. First, a time series of total cropland from perennial and annual crop acreage obtained from the CDL and harvested acres from the NASS datasets. Second, we present the time series of the detrended CDL cropland area and detrended time series of NASS cropland area.
These figures are shown in Appendix Figure A2. We find that CDL cropland acreage data reflects NASS harvested acres from 2008 to 2021, except for 2009, 2018 and 2019. In 2009, the aggregated acreage obtained from CDL data had a substantial drop, while in 2018 and 2019, NASS harvested acres had a significant rise. Overall, we find that the CDL data in our sample is most comparable to the NAAS data for the years 2010 through 2017. In our robustness checks, we perform regression analysis on the main specifications for the years 2010 to 2017. Lastly, we take advantage of a stable climate regime and homogeneous biophysical characteristics of our study region, which also reduce false positives, which reduce inaccuracy in cropland data.Our main climate variables are growing degree days during summer and winter , the accumulated annual precipitation, and accumulated chill hours during winter derived at the parcel-level using the PRISM daily dataset for the years 2008–2021. For the purposes of our analysis and to preserve complete cropland data from 2008-2021, we define the climate normals over 27 years. The PRISM data is a high-resolution dataset that is suitable for agricultural-climate analysis and is utilized by researchers to design climate policy for California .The winter chill hour is a critical climate variable in the fruit and nut growing region of the Central Valley. We follow Jackson et al., to calculate the daily chill hours using the daily minimum temperature, mean temperature, daily maximum temperature, and the reference temperature . Winter chill hours are the sum of daily chill hours during plant’s dormancy period of November through February. Depending on the variety, a tree crop can require anywhere from 200 and 1500 chill hours during winter to produce flowers and fruits .
Appendix Figure A5 highlights the spatial variations in growing season used in our study region. During summer, there are significant variations in degree days in the central and southern parts of Central Valley, while there are small variations in degrees days in winter in the southern parts of Central Valley . In the northern and southern parts of the Central Valley, the change in average annual precipitation over a long period is greater, whereas in the central part, there is a decrease in average annual precipitation. The northern and central parts of the Central Valley experience a greater increase in winter chills hours. Overall, there are enough variations in our study region to identify long-term climatic impacts in our empirical design.To assess parcel’s suitability for agricultural production, we link the parcel-level cropland data to the dominant land capability class , an integrated measure of soil quality and agricultural potential, which is widely used in literature to measure land quality. We obtained LCC data for California from the California Soil Resource Lab at UC Davis, which is available in grid cells of 800 meters . We construct one indicator for high quality land and two indicators for low-quality land . On average, more than half of the sample is on high-quality land , while 40% of the sample is on low-quality land, and only 4% of the sample is on the lowest-quality land.Table 1 presents the descriptive statistics of 657,554 observations from 2008 to 2021. Perennial crops have the highest land-use shares on average at 0.52, followed by annual crops , and non-cultivated crops . About 21% of parcels do not have any share of perennial crops, while 34% and 41% of parcels do not have any share of annual crops and non-cultivated crops, respectively. In appendix Table A2, we present the annual composition of our dependent and explanatory variables. The share of perennial crops increased by 29 percent from 0.48 in 2008 to 0.62 in 2021, while the share of annual crops declined by 32 percent from 0.38 in 2008 to 0.26 in 2021. The non-cultivated crop shares, which include fallowed/idle land and natural vegetation, varied between 0.11 to 0.20, with more fallowed land during drought years. Overall, these descriptive statistics demonstrate that perennial crops have replaced annual crops. We explore further to examine crop-specific variations over time. For illustration purposes, we randomly split the sample into two periods: the first period from 2008 to 2014 and the second period from 2015 to 2021 . The land share allocated to perennial crops, particularly almonds, pistachios, and nuts, hydroponic nft channel increased by 8% in the second period. This increase was predominantly in high-quality land and in low-quality land . The land share allocated to annual crops declined by 12%, with a reduction of 10% in high-quality land and a 6% and 2% decline in low-quality land . Together, the trends in agricultural land use shares indicate further that substitution of annual crops for perennial crops, particularly allocation to almonds, pistachios, and nuts, has occurred on both high-quality and low-quality land . Next, we discuss the climate variables used in the study. As previously mentioned, we define climate normal over 27 years. During our study period, on average, there were 2,083 degree days in summer and 415 degree days in winter.
The long-term average total precipitation was 346 mm. Moreover, in winter, the Valley accumulates 962 hours of long-term chill hours, on average. The average degree days for both summer and winter are fairly uniform over the study period, with the summer average values between 2000 and 2200 degree days and the winter average between 400 and 4300 degree days . The precipitation levels in the long run fluctuated between 350 and 370 millimeters, but in 2009, they decreased significantly to almost 291 millimeters from 363 millimeters the previous year. The winter chill hours have decreased over time, from 1006 cumulative hours in 2008 to 892 cumulative hours in 2021. Although winter chill hours have decreased, the values for most tree crops are still above the upper bound thresholds. Finally, we use the appraisal value of farmland, divided by the acreage of the lot, to calculate the variable appraisal value per acre as a measure of net returns from the farmland. On average, the appraised value of farmland in the study area and period is 7.31 thousand dollars per acre. The dollar values are adjusted for inflation. The annual Gross Domestic Product obtained from the Federal Reserve Economic Database is used to convert nominal values to 2017 U.S. dollars .We first present a simple correlation analysis between agricultural land-use shares and the climate variables. Second, we present the transition probabilities among major crops grown in the Valley. Third, we present combined regression results to examine changes in parcel-specific crop types and the probability of switching between crop types. Fourth, we perform two robustness checks: we use restricted sample to address measurement error in land use change CDL data; and we include an additional regressor for distance to control for the correlation between the proximity of parcels to one another. Lastly, using the estimated coefficients from our econometric models, we simulate the changes in parcel-specific agricultural land-use shares across northern, central, and southern parts of the Central Valley in response to future climate projections. Appendix Figure A4 and present scatter plots that show a correlation between the share of perennial and annual crops and climate variables such as degree days in summer and winter, annual precipitation, and winter chill hours. During the summer, the share of perennial crops increases at an increasing rate, while it decreases during winter. The relationship between perennial crops and total annual precipitation shows a flat to downward slope, suggesting that excess precipitation may decrease the share of perennial crops. The relationship between annual crop shares and summer degree days and total precipitation decreases, but it increases with winter degree days.We present a probability estimate for the likelihood of a crop being grown in the next period among the major crops by land quality in our study region. In order to do that, we split the land use shares of perennial and annual crops into crop-specific land-use shares by land quality and year in the study region, as shown in the Appendix Table A3. The changes in land-use shares are displayed for two periods to maintain readability. Appendix Table A4 displays the probability that a typical grower in our study region will continue to grow the same crop in the next period . The probability of a grower cultivating almonds, pistachios, and nuts in the next period is 92% . Furthermore, growers are 63% and 37% likely to cultivate grapes, citrus, and other subtropical fruits respectively in the next period. The probability of annual crops growing in the next period is low , with the exception of alfalfa . Lastly, land that was fallowed or idled in the first period has an 80% probability of continuing to be fallowed or idled.We repeat the assessment of transition probabilities to examine how crops transition between different land classes . Columns 2, 3, and 4 of Appendix Table A4 report the results. The key findings are as follows. Almonds, pistachios, nuts, and crops like alfalfa have a very high probability of being grown in low quality land in the next period. While rice crops have the least probability of being grown on high quality land in the next period .Table 2 presents combined regression results to examine changes in parcel-specific crop types and the probability of switching between crop types . Quantitatively, the marginal effects derived from the panel fixed effects model and fractional multinomial logit model are alike, expect for annual precipitation.