An Improved CASA Model Based On The Grassland Comprehensive And Sequential Classification System And Its Application To The Grassland NPP In China

Net primary productivity (NPP) of vegetation, which is defined as the net flux ofcarbon from the atmosphere into green plants per unit time and unit area, plays crucialroles in global change and carbon balance. NPP is a major determinant of carbonsinks on land and a key regulator of ecological processes. Grassland NPP isdetermined together by soil, grass, and livestock in grassland ecosystem andenvironmental factors such as climate. Grassland NPP can directly reflect theproduction capacity of grassland communities in a natural environment.Comprehensive and sequential classification system of grasslands (CSCS) is aunique vegetation classification system (mainly for grassland) that is dependent onquantitive measurement indexes. In CSCS, differentiation between class groups isdetermined by an integrated index of>0oC average annual cumulative temperature(Σθ) and moisture index (K). Thus, the classification indicators of CSCS are clear andinvolve a great amount of information and it can be used as a reference for animalhusbandry on the grasslands. Additionally, CSCS is the first classification system thatmay retrieve quantitively by computer. Up to data, the distribution maps for grasslandclasses in Gansu province, China, and northern hemisphere have been completed byCSCS, respectively. The study of coupling of simulating grassland NPP and CSCS isnot only a creative research method, but also is supplement and development forCSCS.Carnegie Ames Stanford Approach (CASA) model is a process-based ecosystemdepiction of NPP, which has taken full account of environmental conditions andvegetation characters. As cumulative temperature (energy) and precipitation(substance) are two key factors that affect vegetation exist, the CASA model havebeen improved based on the relationship of the quantitive classification of CSCS andgrassland NPP. As a result, we established a new model for grassland NPP in thispaper. On the uniform GIS platform, the spatial overlay analysis of basic map,MODIS data, observed meteorological data, and measured data was carried out. Afterthe comparison with observed data and other NPP product data, the validation andprecision of the improved CASA model were analyzed. Then, the grassland NPP from2004to2008in China and its spatio-temporal variations were estimated and analyzed.We also simulated the potential trend of NPP of41grassland classes in China based on CSCS. The followings are the main results and conclusions of this study.1. The improvement and validation of CASA modelHere, Σθ and K in CSCS were introduced in CASA model, and the calculation ofW x,t in model was improved. To estimate the NPP of different grassland classesby the improved model, the maximum light-use efficiency (max) of41grasslandclasses in China was simulated according to the measured data. The values ofobserved data and estimation data of the improved model were distributed round thetrend line. There were likely linear relationship between observed data and estimationdata and the correlation coefficient (R~2) was0.715. The estimation data of41grassland classes was included in the range of observed data except IID23. Averagevalue of observed data was503.75gC/m~2/yr and estimation data was567.312gC/m~2/yr, indicating the difference between them was small. The variation andfluctuation of the simulated values was small, and its total standard error was108.598,which was lower than that of observed data (248.091). Correlation analysis ofobserved data and estimation data from12grassland classes which have manysamples revealed that the linear correlation trend was obvious (R2=0.54~0.93).The characters of errors distribution of observed data and estimation data from41grassland classes are as follows:(1) total average error was4.85gC/m~2/yr; thesamples of positive average error were more than those of negative average error;estimation data was higher than observed data.(2) the average absoluteness error was3.423～241.783, and its average was108.428gC/m~2/yr, indicating that theaccumulation value of error was relative high; the average relative error was-23.3%~51%, and its average was7.6%.(3) when the number of samples was enough,the relative error between observed data and estimation data was small, suggestingthat the estimation results were reliable.The light-use efficiency () in the improved CASA model was0.008～0.846,and its average was0.345, which were higher than that estimated by other researchers.The reason for this may be:(1) the span between the maximum value and theminimum value was high, resulting in the average of was high;(2) the estimationvalue ofmaxof some grassland classes was high.The estimation data of the improved model was lower than that of Miami model andThornthwaite Memorial model. The difference between the estimation data of Miamimodel or Thornthwaite Memorial model and the observed data was high, while theestimation data of the improved model was more precise. Statistically significant relationships were determined between the improved CASA model and Miami modelor Thornthwaite Memorial model. The results showed that there were likelysignificant (high) correlations between the improved CASA model and Miami model(R2=0.868). The improved model had relative low correlations with ThornthwaiteMemorial model, and its correlation coefficient was0.683. This confirmed that theprecision of the improved CASA model is enough for the estimation of differentgrassland classes in China.2. Spatio-temporal variations of NPP in China(1) Spatial variationsThe annual total and average of grassland NPP in China were6.79810~9gC/yrand489.361gC/m~2/yr from2004to2008, respectively. There were clearly strongregional variations in grassland NPP. The grassland NPP in east was higher than thatin west, and that in south was higher than that in north. It increased from NorthwestChina toward Southeast China except Tibetan Plateau. This was consistent with theindex chart for determining grassland class in CSCS, which may show thecharacteristic of horizontal and vertical distribution of Σθ and K.The grassland NPP increased with the increasing longitude and the decreasinglatitude, although there were some fluctuations. The characteristic of longitudezonation of grassland NPP had a highly correlation with longitudinal variabilities ofclimate. Climate becomes more benefit for the vegetation growth with the increasinglongitude. In CSCS, with the suitable natural landscape from desert and semidesert tomeadow and forest with the increasing K value, the NPP has an increasing trend. Thedistribution pattern of grassland NPP of different latitude reflected the NPP responseto thermal gradient. With the increasing latitude, climate became from hot to cold, andΣθ decreased accordingly. NPP decreased with the decreasing Σθ when K value wasthe same.(2) Temporal variationsFrom2004to2008, the annual total NPP of grassland in China ranges from5.9910~9to7.3610~9gC/yr, and its annual average ranges from428.9to527.5gC/m~2/yr. The trend of grassland NPP in China from2004to2008was increaseexcept some fluctuation, and the NPP increased by22.9%in these5years. From thetrend of monthly variation, it can be drawn that the NPP accumulative period wasmainly between April and October when the combination of water and thermal is in agood condition for grassland vegetation growth. The quantity of NPP in these seven months was957.017gC/m~2, which was about89.1%of the annual total. FromOctober to April in the next year, its NPP was about11%of the annual total for thetemperature is too low to inhibit grassland vegetation grow. The grassland NPP inspring, summer, autumn and winter was45.826gC/m~2,136.849gC/m~2,39.935gC/m~2and10.228gC/m~2, respectively, which was about18.6%,59.6%,17.4%and4.5%each of the total annual NPP.According to the annual, monthly and seasonal variation of grassland vegetationNPP, the suitable combination of water and thermal plays key roles in grasslandvegetation growth. In CSCS, differentiation between class groups is determined by anintegrated index of moisture (K-value) and temperature (Σθ). Thus, there existsinherent relationship between grassland classes in CSCS and its NPP.(3) NPP variation of different grassland classesThe variations of annual average NPP in different grassland classes weresignificant in China from2004to2008. VIF41(sub-tropical perhumid evergreenbroad leaved forest) had the highest annual average NPP value, and then VIIF42(tropical-perhumid rain forest). However, IVA4(warm temperate-extrarid warmtemperate zonal desert) had the lowest annual average NPP. During the experiment,VIF41had the highest total NPP value, and then VF40(warm-perhumiddeciduous-evergreen broad leaved forest) and IF36(frigid perhumid rain tundra,alpine meadow), VIA6(Subtropical-extrarid subtropical desert) had the lowest totalNPP, it agree with its minimum area. The K-values of VIF41and VIIF42were same,which were both greater than2.0. When K-value is same, the total grassland NPPdecreased with the decreasing Σθ. As the humidity grades of IVA4is extrarid and itsΣθ is high, the combination of water and thermal is not good for grassland vegetationgrowth, resulting in low NPP. The results further confirmed the idea that the suitablecombination of water and thermal is a key factor for vegetation NPP.The NPP showed a decrease from2004to2008for all grassland classes exceptIVB11(warm temperate-arid warm temperate zonal semidesert). VD26(warm-subhumid deciduous broad leaved forest) had the greatest increase range of52.7%, then IVE32(warm temperate-humid deciduous broad leaved forest,44.2%).IVC18(warm temperate semiarid warm temperate typical steppe), IVD25(warmtemperate-subhumid forest steppe), VC19(warm-semiarid subtropicalgrasses-fruticous steppe), VE33(warm-humid evergreen-deciduous broad leavedforest), and VID27(subtropical-subhumid sclerophyllous forest) all had the similar increase range of about40%. IF36had the lowest increase range of3.4%.System clustering analysis was conducted based on the annual average NPP of41grassland classes from2004to2008. The41grassland classes were clustered intothree groups: the groups1had high NPP for its high K-values and Σθ and suitableratio of moisture and temperature. However, the NPP of the groups2was low for itslow K-values and high Σθ. Therefore, the degree of coupling between grassland NPPand the classes and super-classes in CSCS was high.3. Relationship between grassland NPP in China and its influencefactors(1) Correlation analysis between NPP and influence factorsThere were significant positive linear correlations between the annual averageNPP in China from2004to2008and NDVI, precipitation, and K-values. Σθ had anegative direct effect on NPP. NPP had the highest correlation with NDVI and had theweakest correlation with solar radiation. The above results demonstrated that NDVI,precipitation, and K-values are the main factors for grassland NPP in the improvedCASA model.The correlation between NPP of various grassland super-classes and its influencefactors are different. The NPP of10zonal grassland super-class groups in China from2004to2008had positive correlation with NDVI, precipitation, and K-values. Intundra alpine steppe, temperate zonal humid grassland, and temperate zonal foreststeppe, annual NPP had negative correlation with solar radiation and Σθ. The NPP offrigid desert had relatively high correlation with NDVI and precipitation, while it hadrelatively weak correlation with Σθ. In semidesert, there were high correlationbetween annual NPP and NDVI. The annual NPP of steppe had high correlation withprecipitation. In sub-tropical zonal forest steppe, annual NPP had high correlationwith precipitation and K-values.According to the lag-linear correlations analysis, the response of grassland NPPin China from2004to2008to NDVI, precipitation, and K-values did not show a timelag effect. However, the time lag as to responses of NPP to solar radiation and Σθ wasaround1month and2months, respectively. There was no accumulated time lag effectof solar radiation and precipitation on grassland NPP. The accumulated time lagperiods as to responses of NPP to Σθ were4months.The results of partial correlation analysis showed that when NDVI, solar radiation,K-values, Σθ and precipitation were regarded as control varies respectively, the correlation coefficients between grassland NPP in China and K-values were all similar.Thus, the effects of NDVI, solar radiation, Σθ and precipitation on grassland NPPwere interaction but not independent. Among various influence factors, the correlationcoefficients between K-values and other factors were relatively low. NDVI andprecipitation had the highest correlation, and Σθ and precipitation had the second highcorrelation coefficients. Therefore, the effects of climate changes on grassland NPPwere complex, and the integration of moisture and thermal is the key factors for theincrease of grassland NPP.(2) Sensitive analysis on NPP to its influence factorsNDVI was the most sensitive variable to the grassland NPP in China simulatedby the improved CASA model in general in this study, followed by Σθ, K-values,precipitation, and solar radiation. K-values and Σθ were used to quantitiveclassification in CSCS, and these two parameters were sensitive to NPP. Thedifferences of NPP values of different grassland classes increased with the increasingdifference of Σθ and K-values. In conclusion, the coupling of CSCS and grasslandNPP has been completed to a certain extent in this study.